University of Southern California Dissertations and Theses

Exploring the pernicious effects of redlining and discriminatory policies on an American city: a spatio-temporal case study of New York City

(USC Thesis Other)

Exploring the pernicious effects of redlining and discriminatory policies on an American city: a spatio-temporal case study of New York City

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Exploring the Pernicious Effects of Redlining and Discriminatory Policies on an American City:

A Spatio-Temporal Case Study of New York City

by

Christopher S. Hayner

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 2022

Copyright © 2022 Christopher S. Hayner

ii

To Klausy and Kahlo, for being good cats, and brightening my world considerably at the height
of the pandemic.

To Alana and Renee, for listening to my rants on racist practices, for giving me constructive
criticism on the scale and scope of this thesis, for being patient with me when I became absorbed
in my work (and neglected the housework), and mostly, for your constant love and support.

iii
Acknowledgements
I am grateful to my thesis advisor, Dr. Elisabeth Sedano, for her guidance not only in this thesis,
but throughout much of this program at the University of Southern California. Her collaboration
not only supported the technical prowess to conduct this analysis, but often reinforced a passion
for urban equity issues. I am very appreciative of the feedback and guidance of my thesis
committee, Dr. Darren Ruddell and Dr. Robert Vos. Lastly, I would like to thank Dr. Masayoshi
Oka for graciously answering questions on conducting local spatial segregation measures.

iv
Table of Contents
Acknowledgements ........................................................................................................................ iii
List of Tables ................................................................................................................................ vii
List of Figures ................................................................................................................................ ix
Abbreviations ............................................................................................................................... xiii
Abstract ........................................................................................................................................ xiv
Chapter 1 Introduction .................................................................................................................... 1
1.1. Overview .............................................................................................................................1
1.2. Study Area ..........................................................................................................................4
1.3. Motivation .........................................................................................................................10
Chapter 2 Background and Related Work .................................................................................... 14
2.1. Historical Backdrop to Institutional Racism in Planning Issues .......................................14
2.1.1. The Great Migration and Initial Exclusionary tactics ..............................................15
2.1.2. The Home Owners’ Loan Corporation and “Redlining” .........................................16
2.1.3. Suburbanization and the Rise of Single-Family Zoning ..........................................19
2.1.4. Slum Clearance, Urban Renewal, and Suburban Mortgages ...................................20
2.1.5. Separate and Unequal ..............................................................................................22
2.1.6. New York Backdrop ................................................................................................25
2.2. Literature review ...............................................................................................................34
2.2.1. Literature Informing the Historic Backdrop and Variable Selection .......................34
2.2.2. Literature Informing Spatial Research Methods ......................................................38
Chapter 3 Methods ........................................................................................................................ 48
3.1. Data Description and Preparation .....................................................................................49
3.1.1. Dependent Variables ................................................................................................49
3.1.2. Independent Variables .............................................................................................59

v
3.2. Research Design................................................................................................................79
3.2.1. Exploratory Spatial Analyses ...................................................................................80
3.2.2. Comparative Regression Analyses ..........................................................................84
Chapter 4 Results .......................................................................................................................... 91
4.1. Exploratory Spatial analyses .............................................................................................91
4.1.1. Population Pattern Analyses ....................................................................................91
4.1.2. Segregation Analyses ...............................................................................................98
4.2. Comparative Regression Analysis ..................................................................................101
4.2.1. Population Density .................................................................................................104
4.2.2. Population Percentage ............................................................................................114
4.2.3. Population Density Change....................................................................................123
4.2.4. Dissimilarity ...........................................................................................................126
4.2.5. Isolation..................................................................................................................128
4.2.6. Spatial Heterogeneity .............................................................................................130
Chapter 5 Discussion .................................................................................................................. 139
5.1. Overall Findings..............................................................................................................139
5.1.1. Persistence of White Privilege ...............................................................................139
5.1.2. Path Dependency Between Policies .......................................................................144
5.2. Policy Recommendations................................................................................................149
5.2.1. Remove Exclusionary Zoning Tools .....................................................................149
5.2.2. Reconsider Contextuality .......................................................................................150
5.2.3. Allow Infill Opportunities on Campuses ...............................................................150
5.2.4. Equitably Distribute Zoning and Planning Resources ...........................................151
5.3. Data and Other Research Limitations .............................................................................151
5.3.1. Subway and Highway Datasets ..............................................................................151

vi
5.3.2. Zoning Datasets .....................................................................................................154
5.3.3. Market-based Counterfactuals ...............................................................................156
5.4. Additional Avenues of Inquiry .......................................................................................158
5.4.1. Gentrification .........................................................................................................158
5.4.2. 1916 Zoning ...........................................................................................................158
5.4.3. Comparative Measures Between Cities .................................................................159
5.4.4. Other Segregation Studies ......................................................................................160
5.5. Conclusion ......................................................................................................................160
References ................................................................................................................................... 161
Appendix A: Detailed Regression Results – Population density change ................................... 172
Appendix B: Detailed Regression Results – Dissimilarity ......................................................... 178
Appendix C: Detailed Regression Results – Isolation ................................................................ 184
Appendix D: Detailed Regression Results – HOLC boundaries ................................................ 190

vii
List of Tables
Table 1. Response variables in regression analyses ..................................................................... 50
Table 2. Independent variables to investigate in regression analyses .......................................... 77
Table 3. OLS Regression results, Black and White population density, 1960 .......................... 105
Table 4. SLM and SEM Regression results, Black and White population density, 1960 .......... 107
Table 5. OLS Regression results, Black and White population density, 1990 ........................... 108
Table 6. SLM and SEM Regression results, Black and White population density, 1990 .......... 109
Table 7. OLS Regression results, Black and White population density, 2020 ........................... 110
Table 8. SLM and SEM Regression results, Black and White population density, 2020 .......... 111
Table 9. OLS Regression results, Black and White population percentage, 1960 .................... 115
Table 10. SLM and SEM Regression results, Black and White population percentage, 1960 ... 116
Table 11. OLS Regression results, Black and White population percentage, 1990 ................... 117
Table 12. SLM and SEM Regression results, Black and White population percentage, 1990 ... 118
Table 13. OLS Regression results, Black and White population percentage, 2020 ................... 119
Table 14. SLM and SEM Regression results, Black and White population percentage, 2020 ... 120
Table 15. Range of GWR coefficients for select variables ......................................................... 131
Table 16. Regression results, Black population density change, 1930-1960 .............................. 172
Table 17. Regression results, White population density change, 1930-1960 ............................. 173
Table 18. Regression results, Black population density change, 1960-1990 .............................. 174
Table 19. Regression results, White population density change, 1960-1990 ............................. 175
Table 20. Regression results, Black population density change, 1960-1990 .............................. 176
Table 21. Regression results, White population density change, 1960-1990 ............................. 177
Table 22. Regression results, Black-other dissimilarity, 1960 ................................................... 178
Table 23 Regression results, White-other dissimilarity, 1960 .................................................... 179
Table 24. Regression results, Black-other dissimilarity, 1990 ................................................... 180

viii
Table 25. Regression results, White-other dissimilarity, 1990 ................................................... 181
Table 26. Regression results, Black-other dissimilarity, 2020 ................................................... 182
Table 27. Regression results, White-other dissimilarity, 2020 ................................................... 183
Table 28. Regression results, Black isolation, 1960 ................................................................... 184
Table 29. Regression results, White isolation, 1960 ................................................................... 185
Table 30. Regression results, Black isolation, 1990 ................................................................... 186
Table 31. Regression results, White isolation, 1990 ................................................................... 187
Table 32. Regression results, Black isolation, 2020 ................................................................... 188
Table 33. Regression results, White isolation, 2020 ................................................................... 189
Table 34. Regression results, Regression results, HOLC A boundaries, 1960 ........................... 190
Table 35. Regression results, Regression results, HOLC B boundaries, 1960 ........................... 191
Table 36. Regression results, Regression results, HOLC C boundaries, 1960 ........................... 192
Table 37. Regression results, Regression results, HOLC D boundaries, 1960 ........................... 193
Table 38. Regression results, Regression results, HOLC A boundaries, 1990 ........................... 194
Table 39. Regression results, Regression results, HOLC B boundaries, 1990 ........................... 195
Table 40. Regression results, Regression results, HOLC C boundaries, 1990 ........................... 196
Table 41. Regression results, Regression results, HOLC D boundaries, 1990 ........................... 197
Table 42. Regression results, Regression results, HOLC A boundaries, 2020 ........................... 198
Table 43. Regression results, Regression results, HOLC B boundaries, 2020 ........................... 200
Table 44. Regression results, Regression results, HOLC C boundaries, 2020 ........................... 200
Table 45. Regression results, Regression results, HOLC D boundaries, 2020 ........................... 201

ix
List of Figures

Figure 1. The disproportionate killing of Black and Hispanic Americans by police officers ........ 1
Figure 2. Images of the expanding subway system in NYC ........................................................... 6
Figure 3. Generational shifts in housing patterns in NYC ............................................................. 7
Figure 4. Population growth, by borough, of NYC ........................................................................ 7
Figure 5. Distribution and density of population in NYC, by race, in 2020 ................................... 9
Figure 6. HOLC map for Buffalo, NY, with inset denoting the supermarket ............................. 11
Figure 7. Images showing the impact of urban renewal and housing policies in St. Louis . ........ 12
Figure 8. Black population in the US, circa 1900. ........................................................................ 15
Figure 9. HOLC maps from Minneapolis and Baltimore ............................................................. 18
Figure 10. Land zoned for single-family homes in select US cities ............................................. 20
Figure 11. Images showing highway-building and other urban renewal impacts in NYC .. ........ 21
Figure 12. Different trajectories stemming from the Housing Act of 1949 .................................. 23
Figure 13. Images of White suburbanites resisting integration .................................................... 24
Figure 14. Proposed location of, and protest against, the Lower Manhattan Expressway ........... 28
Figure 15. HOLC D area designation writeup for West Village versus Bedford-Stuyvesant ...... 30
Figure 16. HOLC maps of Manhattan compared with the Slum Clearance boundaries .............. 30
Figure 17. Change in Black and White population between 2010 and 2020 ................................ 33
Figure 18. Illustration of the checkerboard problem .................................................................... 45
Figure 19. Dimensions of spatial segregation ............................................................................... 46
Figure 20. White population density per acre: 1960, 1990, and 2020 .......................................... 52
Figure 21. Black population density per acre: 1960, 1990, and 2020 .......................................... 52
Figure 22. White percentage of population of census tract: 1960, 1990, and 2020 ...................... 53
Figure 23. Black percentage of population of census tract: 1960, 1990, and 2020 ...................... 53

x
Figure 24. White population change per acre: 1930-1960, 1960-1990, and 1990-2020 .............. 55
Figure 25. Black population change per acre: 1930-1960, 1960-1990, and 1990-2020 ............... 55
Figure 26. Formula for local spatial dissimilarity indices ........................................................... 56
Figure 27. White – others dissimilarity, by census tract: 1960, 1990, and 2020 .......................... 57
Figure 28. Black – others dissimilarity, by census tract: 1960, 1990, and 2020 .......................... 57
Figure 29. Formula for local spatial isolation index .................................................................... 58
Figure 30. White isolation, by census tract: 1960, 1990, and 2020 .............................................. 58
Figure 31. Black isolation, by census tract: 1960, 1990, and 2020 .............................................. 59
Figure 32. HOLC areas, subway adjacent areas, highway adjacent areas ................................... 62
Figure 33. Public housing units by building: 1960, 1990, and 2020 ............................................ 63
Figure 34. URA plan examples in Harlem ................................................................................... 64
Figure 35. Urban renewal areas: 1960, 1990, and 2020 ............................................................... 65
Figure 36. NYC zoning district spectrum .................................................................................... 66
Figure 37. Semi-detached homes in Queens ................................................................................ 67
Figure 38. Comparison of potential built form in non-contextual versus contextual districts .... 68
Figure 39. Rezoning map for the 2007 Bedford-Stuyvesant rezoning ........................................ 69
Figure 40. Demographic change, 2010-2020 in Bedford-Stuyvesant NTAs ............................... 70
Figure 41. HOLC A and B areas around Central Park compared with R10 zoning .................... 72
Figure 42. Zoning designations: 1960, 1990, and 2020 ............................................................... 73
Figure 43. Example of zoning changes in South Richmond, Staten Island, 1975 and 2006 ....... 75
Figure 44. Example of zoning changes in Astoria, Queens, 1990 and 2020 ............................... 75
Figure 45. Special purpose districts: 1960, 1990, and 2020 ......................................................... 76
Figure 46. Historic districts: 1960, 1990, and 2020 ..................................................................... 77
Figure 47. Process diagram for thesis methodology ..................................................................... 80
Figure 48. Z-scores resulting from Incremental Spatial Autocorrelation tool .............................. 83

xi

Figure 49. Black population density and hot spot analysis, 1910 – 1930 ..................................... 92
Figure 50. Black population density and hot spot analysis, 1940 – 1960 ..................................... 94
Figure 51. Black population density and hot spot analysis, 1970 – 1990 .................................... 95
Figure 52. Black population density and hot spot analysis, 2000 – 2020 .................................... 96
Figure 53. Moran’s I values, 1910 – 2020 .................................................................................... 97
Figure 54. Population of NYC 1910 – 2020, by race .................................................................. 97
Figure 55. Local Moran’s I, 2020, with public housing insets .................................................... 98
Figure 56. Dissimilarity, diversity, information theory, and exposure / isolation index results ... 99
Figure 57. White and Black pop. density surfaces, comparing 1930, 1960, 1990, and 2020 ..... 101
Figure 58. Histograms showing the skewness in select explanatory variables........................... 102
Figure 59. Fitted line and Q-Q plot of a standard OLS 3 run ..................................................... 103
Figure 60. Residual versus Fitted Line and Q-Q Plots for Black and White pop. density ......... 107
Figure 61. Regression coefficients explaining Black versus White population density, 1960 .. 112
Figure 62. Regression coefficients explaining Black versus White population density, 1990 .. 113
Figure 63. Regression coefficients explaining Black versus White population density, 2020 ... 114
Figure 64. Regression coefficients explaining Black versus White pop. Percentage, 1960 ...... 121
Figure 65. Regression coefficients explaining Black versus White pop. Percentage, 1990 ...... 122
Figure 66. Regression coefficients explaining Black versus White pop. Percentage, 2020 ...... 123
Figure 67. Regression coefficients explaining Black versus White density change, 1960 ........ 124
Figure 68. Regression coefficients explaining Black versus White density change, 1990 ........ 125
Figure 69. Regression coefficients explaining Black versus White density change, 2020 ........ 126
Figure 70. Comparison of SEM dissimilarity coefficients for 1960, 1990 and 2020 ................ 127
Figure 71. Regression coefficients explaining Black versus White isolation, 1960 .................. 128
Figure 72. Regression coefficients explaining Black versus White isolation, 1990 .................. 129
Figure 73. Regression coefficients explaining Black versus White isolation, 2020 .................. 130

xii
Figure 74. Select GWR coefficients influence on Black and White pop. percentage, 1960 ..... 133
Figure 75. Select GWR coefficients influence on Black and White pop. percentage, 1990 ..... 136
Figure 76. Select GWR coefficients influence on Black and White pop. percentage, 2020 ..... 138
Figure 77. Comparison of SEM population density coefficients for 1960, 1990, and 2020 ..... 140
Figure 78. Comparison of SEM pop. percentage coefficients for 1960, 1990, and 2020 .......... 140
Figure 79. Select SEM population density and percentage coefficients .................................... 141
Figure 80. Select SEM population density and percentage coefficients .................................... 143
Figure 81. Select SEM population density and percentage coefficients .................................... 144
Figure 82. Patterns of layered exclusion in Riverdale, the Bronx ............................................. 145
Figure 83. Public housing density and percent urban renewal area by HOLC area .................. 146
Figure 84. Percentage of zoning districts within each historic HOLC area ............................... 147
Figure 85. Percentage of historic districts within each historic HOLC area .............................. 147
Figure 86. Correlation between post-war policies and HOLC areas .......................................... 148

xiii
Abbreviations
AIC Akaike information criteria
BP Breusch-Pagan
CBD Central Business District
DCP Department of City Planning
FHA Federal Housing Administration
GWR Geographically weighted regression
HOLC Home Owners’ Loan Corporation
HPD Department of Housing Preservation and Development
HUD Department of Housing and Urban Development
LM Lagrange multiplier
MAUP Modifiable areal unit problem
MTA Metropolitan Transportation Authority
NTA Neighborhood Tabulation Area
NYC New York City
OLS Ordinary least squares
SEM Spatial error model
SLM Spatial lag model
URA Urban Renewal Area
VIF Variance inflation factor

xiv
Abstract
In the summer of 2020, sustained violence against Black Americans by law enforcement erupted
into nationwide protests following the callous murder of George Floyd. The cultural zeitgeist
prompted a call to action, not only to rethink our policing, but also to examine larger systemic
and institutionalized racism in our society. In urban planning circles, this discussion often begins
with an examination of the role “redlining” maps created in the 1930s by the federal government,
which controversially appraised lending risk with a racial lens, stigmatizing areas with Black
residents, outlined in red, as risky for investment, and contributing to ensuing segregation.
Through examination of the nation’s largest metropolis, New York, this thesis evaluates
whether redlining was only one factor of government policy – federal or municipal – entrenching
segregation in the landscape. Global and local spatial clustering and segregation measures were
conducted in 10-year intervals from 1910 to 2020 to evaluate underlying shifts in the spatial
patterns of Black and White population segments over time. Linear regression, spatial error, and
spatial lag models were then constructed to evaluate the degree to which redlining, urban
renewal designations, public housing concentrations, zoning designations and historic districting
contributed to the spatial segregation of Black and White populations in three distinct years:
1960, 1990 and 2020. The findings showed each era of new urban planning policy contributed to
persisting segregation. The findings also showed that oftentimes a new generation of policy
would spatially reference a prior era, to the benefit or detriment of a particular population: Urban
renewal designations mimicked redlined areas and disproportionately concentrated public
housing into increasingly Black enclaves, while exclusionary zoning tools like single-family
zoning, often mimicked the safest investment designations in redlining maps, prolonging the
privilege of predominantly White communities.
1

Chapter 1 Introduction
This introductory chapter begins with an overview of the project, then provides a description of
the study area, New York City, and lastly, outlines the principal motivations for embarking on
the themes of this thesis: persisting segregation in the urban environment and the role various
government planning policies have played.
1.1. Overview
In the summer of 2020, while most of the nation was in lockdown from COVID-19, a
teenage girl took a video on her cell phone of the murder of George Floyd, a Black man in
Minneapolis, by a White officer, Derek Chauvin. The callous violence captured by her video
sparked riots and protests throughout the nation and prompted calls for police reform and the
examination of larger systemic and institutionalized racism in our society. This murder of a
Black man by White police officers was not an isolated incident, and Minneapolis is not uniquely
violent; from Eric Garner in Staten Island to Michael Brown in Ferguson to Breonna Taylor in
Louisville, nationwide, Black Americans are killed by police officers at rates almost twice as
high as that of White Americans (see Figure 1) (The Washington Post 2021).

Figure 1. The disproportionate killing of Black and Hispanic Americans by police officers
(The Washington Post 2021)
2

Government officials have long attributed these crimes to rogue police officers or the
unique circumstances of particular communities
1
, a misdirect that conveniently glosses over the
fact that the neighborhoods where these crimes disproportionately occur have a shared legacy of
being clustered within and segregated from the larger urban area, suffering from inferior
amenities and public services (Rothstein 2014). Importantly, this segregation is not happenstance
and did not emerge purely from the preferences or biases of people to live amongst those with
similar skin color; rather, this spatial pattern was shaped by discriminatory government policies
(Rothstein 2017). In the introspective moment following the death of George Floyd, city
planning departments across the nation, including those representing its two largest cities, New
York and Los Angeles, finally began to acknowledge the historical role they have played
codifying segregation into the spatial pattern of American cities and their responsibility in
devising policies that “played a large role in perpetuating racism and even violence against Black
and Brown Americans” (NYC DCP 2020).
Central to many cities’ discussions have been the historical “redlining” maps associated
with the Home Owners’ Loan Corporation (HOLC) and the Federal Housing Administration
(FHA) that informed public and private housing lending practices in the first half of the 20th
century. In these maps, the federal government demarcated the purported investment risk of
different neighborhoods, directing how and where bank loans could be issued, and which
mortgages the federal government would insure (Jackson 1987). The federal mapmakers
deliberately singled out areas with high concentrations of racial minorities – chiefly Black

1
The Justice Department, upon investigating Michael Brown’s death in Ferguson, stated “The
Department of Justice is investigating the death of Michael Brown and the racial practices of Ferguson's
police department, but has not suggested recent events reflect anything broader than Ferguson's unique
problems.” (Rothstein 2014)

3

communities – as high risk, thereby limiting re-investment (Rothstein 2017). These racist
practices caused massive repercussions to the urban fabric of most American cities. Because
government officials favored the in vogue suburban home, and assigned the highest value to
White enclaves, they helped usher in an era of suburbanization, further segregated cities and the
surrounding regions, and left many older city centers (and the Black residents in them) without
access to capital for acquiring homes, or reinvesting in their neighborhoods (Coates 2014).
Local land use policies like municipal zoning and urban renewal area designations
layered onto this federal framework and often interlocked with its discriminatory nature,
accelerating the disparity between White and minority neighborhoods (Rothstein 2017).
Collectively, these policies not only helped engineer severe differences in socio-economic status
and generational wealth (Rothstein 2017).
This thesis aims to study the spatial implications of redlining and subsequent federal,
state and municipal policies and planning efforts, on Black communities through the lens of the
nation’s largest city, New York. The goal is to analyze how pervasive the legacy of HOLC
mapping designations, along with other government planning efforts, have been by analyzing
their influence in explaining Black and White spatial patterns over time.
To evaluate these hypotheses, several analyses are conducted. Spatial autocorrelation and
segregation analyses are used to generate initial global measures of Black and White spatial
patterns in the city. Linear regression analyses (ordinary least squares (OLS)) and spatial
regression techniques (spatial lag models (SLM) and spatial error models (SEM)) are run to
determine the correlation between various explanatory variables and demographic-oriented
response variables. Geographically weighted regression (GWR) tools further probe at prominent,
and statistically significant variables to explore their spatial heterogeneity.
4

The analysis places a particular emphasis on land use and other planning policies,
including zoning, urban renewal designations, and redlining demarcations to determine whether
they are significant in explaining persisting segregation, either directly or indirectly, or whether
other factors, like conventional market forces, including subway access or distance to a Central
Business District (CBD), might better explain the phenomenon. Running these same analyses at
different decennial census intervals illuminates the degree to which the residue of institutional
racism persists in the urban fabric.
1.2. Study Area
New York City (NYC) is a dynamic urban juggernaut; it is consistently reinventing itself,
yet for generations has maintained its allure for the prospect of economic opportunity. The
original settlement of the city by European colonists occurred on the island of Manhattan, at the
southern tip. The colonial boundary – Wall Street – has evolved into the center of New York’s
colossal financial district, a staggering, but representative example of Manhattan’s breakneck
pace of development and redevelopment. The surrounding boroughs – the Bronx, Queens,
Brooklyn, and Staten Island – developed after Manhattan, but are now large urban constellations
in their own right. With a population of over 8.8 million people as of the 2020 census, New York
is by far the largest city in the United States (NYC DCP n.d.; US Census Bureau n.d.). In fact, if
divided, four of the five boroughs would be top 10 cities in the US in terms of population, led by
Brooklyn, which would be the fourth largest city at over 2.7 million people (NYC DCP 2021).
The neighborhoods that accommodate these 8.8 million residents within the city limits
vary widely – people live in dense residential towers, single-family detached homes, and
numerous building types in between. The type of buildings associated with a particular
neighborhood are a product of many factors, including the neighborhood’s age and vintage of its
5

regulatory framework, proximity to the Manhattan core and public transit, and localized market
dynamics.
Each boroughs’ population growth, and the neighborhoods therein, varied widely and
was associated with national immigration policies, cultural trends, the development of the
subway system, and later, federal housing policies.
The pre-19th-century city was concentrated in lower Manhattan and near the Brooklyn
waterfront, as access to services, amenities, and employment, was largely determined by a
walking radius. As Manhattan’s population exploded after the 1850s with immigrants pouring in
for economic opportunity and cultural freedoms, neighborhoods like the Lower East Side
became some of the densest conglomerations of people on Earth. Reformers sought to protect the
welfare of the most destitute and instituted regulations that mandated light and air into tenement
buildings and incrementally mandated basic fire safety and quality of life measures (Plunz 2016).
These measures also sought to de-densify immigrant enclaves, through provisions like height
limits, for public health and safety purposes.
With the advent of urban omni-buses, elevated trains, and then the subway system,
population dispersion could accelerate in earnest beginning in the early 20th century (Jackson
1987). When the subway system was expanded in many locations in the outer boroughs, there
was farmland or small villages, such as in rural Queens (Figure 2). Within decades many
subway-adjacent had been completely transformed, including the area around the same transit
line in Queens (also in Figure 2).

6

Figure 2. Images of the expanding subway system in NYC (Queens, 1917 and the 1950s)
(Cohen 2017; Keller Williams Realty 2017)
The manner in which the 20th-century city formed and the racially segregated
byproducts, is largely the subject of this thesis; while market forces certainly interplayed with the
city’s massive infrastructure investments, federal, state, and local policies steered investment and
influenced the social and physical patterns that would come to characterize the urban landscape
The sequence of images in Figure 3 visualize the physical differences in generational housing
construction; from left, old law tenements in the Lower East Side rose in the 19th-century city,
while new law tenements were erected in the Bronx after the subway was constructed, “tower-in-
the-park” style urban renewal transformed Stuyvesant Town in the early post-war years, while
suburban subdivisions emerged in Staten Island after the construction of the Cross-Staten Island
Expressway. Each represents a microcosm that figures into the demographic and socio-economic
pattern of the city.
7

Figure 3. Generational shifts in housing patterns in NYC (Google 2022)
A graph showing the population growth of New York, by borough, is shown in Figure 4
(Gibson and Jung 2005). The initial dominance of Manhattan can be seen, as well as the eventual
dispersion of population to Brooklyn, Queens, and the Bronx. Staten Island is not interconnected
to the subway system, and thus remains mostly suburban. The graph also shows the precipitous
drop in population in the post-war years, along with a renaissance where the initial 1950s
population peak was overtaken in the 2000s.

Figure 4. Population growth, by borough, of NYC
-
2,000,000
4,000,000
6,000,000
8,000,000
10,000,000
Bronx Brooklyn Manhattan Queens Staten Island
8

The map in Figure 5 shows the relative density and spatial pattern of its principal races
and ethnicities. Manhattan south of 96th Street, western and southern Brooklyn, western Queens,
and much of Staten Island is White, while Harlem, the South Bronx, northern Bronx, central
Brooklyn, central and southeastern Queens, and northern Staten Island are segregated pockets of
Black and Latino populations. Melting pot that it is, there are moments of relative mixing and
integration, such as southern Brooklyn, northeastern Queens, and eastern Bronx, but by and
large, New York is highly segmented along racial lines.
9

Figure 5. Distribution and density of population in NYC, by race, in 2020

10

1.3. Motivation
While this thesis is centered around New York City, the motivation for its subject matter
was inspired by violence against the Black community in cities across America.
In May of 2022, a racially motivated mass killing targeting a Buffalo supermarket in a
low-income, predominantly Black neighborhood drew national attention and an outpouring of
grief for the victims. Many in the community itself were vocal about the hypocrisy of this
ephemeral concern, as it followed generations of sustained apathy for their wellbeing – Few
rallied to protect their neighborhood from being targeted for running a highway through it, stood
up and questioned why their children were disproportionately under-represented from the nearby
magnet school, or fought to bring a supermarket to the neighborhood when it lacked an
affordable healthy food option. In the words of resident Marlene Brown: “We don’t want to be
protected after the fact. We want to be protected and treated like we matter, without it taking a
White supremacist shooting up our community. Time and time again they’ve shown nobody
cares about us here. It’s a pattern” (Closson 2022).
Ms. Brown’s quote is apt; the long arc of institutional racism is often glossed over, along
with its role in generating the long-standing disparity between communities. The shooting in a
Tops supermarket (the location of which is denoted by a star in Figure 6) occurred just a few
blocks from one of the only areas north of the downtown area that housed a Black population in
the 1930s and was redlined for their “infiltration”. Today, the diagonal arterial of Main Street
remains a dividing line in the city, with the east side being a low-income Black community. The
racism that stigmatized their community nearly one hundred years ago, seems to have ossified
the urban fabric in time; encapsulating (and potentially even intensifying) its segregation.
11

Figure 6. HOLC map for Buffalo, NY, with inset denoting the supermarket (Nelson, et al. n.d.)
Since much of this violence occurs in urban America, and many Black communities are
hyper-segregated, it seems opportune to explore the role of governmental policies that might
have informed the current division, this “pattern” in Ms. Brown’s words, and trace its effects –
from redlining to exclusionary zoning, to urban renewal plans. Bringing light to these practices
and examining the degree to which their perniciousness has persisted, is timely and important.
The eminent researcher on government-brokered segregation in cities, Richard Rothstein,
used St. Louis to provide a clear example that illustrate how these racist policies link up with
present-day violence against Black Americans. In the 1930s, areas near the historic downtown
were redlined by the HOLC because they housed Black residents (Nelson, et al. n.d.). In the
1950s, these areas became targeted for urban renewal projects, and neighborhoods were razed to
construct the Gateway Arch, a museum, a stadium, interstate highways, and middle-class
housing (Figure 7). Roughly 78% of all families displaced by urban renewal in St. Louis were
12

Black families (ed. Nelson and Ayers n.d.). Meanwhile, public housing projects like the Pruitt-
Igoe houses, were constructed in more isolated areas, with less access to jobs and opportunity,
and housed large numbers of displaced Black residents in segregated enclaves because the
private housing market would often not lend or rent to them. By the 1970s, Pruitt-Igoe had
become a symbol of the dysfunction of public housing, and disinvestment was so bad that the
federal government dynamited the complex and relocated the residents (also in Figure 7).
Residents were given vouchers by the St. Louis Housing Authority and were predominantly
assisted in moving to inner-ring suburbs whose zoning permitted multi-family housing in the
form of apartments – communities like Ferguson. Outer ring suburbs, whose White residents
were more affluent, incorporated themselves and used exclusionary zoning tactics like single-
family zoning districts to effectively block Black residents through socio-economic barriers to
entry. In 1970, less than 1% of Ferguson was comprised of Black residents; by 1980, it was
around 14%; by 1990, 25%; by 2000, 52%; and by 2010, 67% (Rothstein 2014).

Figure 7. Images showing the impact of urban renewal (1950s) and housing policies in St. Louis
(1972) (Rothstein 2014)
In providing this history of St. Louis that ties the historical planning practices of
redlining, urban renewal, public housing failures, and zoning, with the emergence of Ferguson as
13

an enclave of low-income Black Americans – the type of place that leaves young Black men like
Michael Brown vulnerable to discrimination and police brutality – Richard Rothstein
underscored: “Every policy and practice segregating St. Louis over the last century was
duplicated in almost every metropolis nationwide” (Rothstein 2015, 169). It is difficult if not
impossible to redress issues like police brutality when the underlying institutional racism that has
segregated cities for over a century, and its role in fostering a vast divide in political power,
societal privilege, public safety, and economic opportunity, remains veiled (Rothstein 2017).
Identifying the persistence of this division is important to advance the discussion of equity in
communities across the nation.
Fixing the problems of historic racist policies will require interventions to dismantle the
spatial barriers that still exist in our cities’ built environments and will necessitate course
corrections in their land-use frameworks. Identifying explanatory variables for the segregated
patterns of today is important, even if the degree of influence of such variables differ from city to
city (Logan and Stults 2021). In the short term, isolating these variables can inform policymakers
where to target their efforts to improve integration and access. The racism embedded in the
American city will take the collective work of many planners, and policymakers, in conjunction
with other civic actors, to dismantle completely. This project humbly aims to advance that goal
towards advocating greater urban integration and equitable access to opportunities, amenities,
resources, and protections amongst the various populations that characterize American cities.

14

Chapter 2 Background and Related Work
This section is divided into two core components. First, a core historical backdrop pertaining to
the role governmental policies in the role of accelerating housing segregation in US cities is
surveyed to convey the salient policy issues, but also illuminate the relevancy of specific
variables in the regression analyses. This first section also includes a NYC backdrop where race
and its intersectionality with planning and land use planning is discussed, along with the local
figures that played an outsized role in its manifestation. Second, after this backdrop, a literature
review details the pre-eminent research informing the backdrop, and details other studies that
informed the research methods. In each of these sections, the limitations associated with the
existing research are briefly outlined.
2.1. Historical Backdrop to Institutional Racism in Planning Issues
In 1619, Africans were sold as slaves to European colonists in modern-day America for
the first time, propelling the future nation into a regrettable trajectory of white supremacy; an
ideology whose scars remain in the social fabric even after 400 years (Elliott and Hughes 2019).
In the decades prior to, and immediately after the American Revolution, plantations that profited
from slave labor were largely concentrated along the Southern coasts. As indigenous peoples
were forcibly removed from inland areas and more valuable farmland acquired by White
Americans, a “Black Belt” emerged by the 1850s in the Deep South – a double entendre
reflecting the rich soil of the Piedmont and the alluvial floodplains around the Mississippi-Yazoo
Delta and the high concentration of African-origin slaves working the plantations in these areas
(Figure 8) (Ayers, Madron, and Ayers 2021). In the aftermath of the Civil War, most
emancipated Blacks remained closely tethered to rural, southern farms, often working as
sharecroppers on White-owned cotton fields. Despite the constant attempts from White
15

Southerners to subvert their newfound political rights and freedoms through overt terrorism or
covert, discriminatory legal machinations (Coates 2014), these Black Americans often did not
migrate; they did not have the economic means to do so, and the Deep South only became well
connected to the North via railroads in the 19th century (Ayers, Madron, and Ayers 2021).

Figure 8. Black population in the US, circa 1900 (Gannett 1904)
2.1.1. The Great Migration and Initial Exclusionary tactics
Beginning in the early part of the 20th century, the spatial geography of Black Americans
began to shift. The outbreak of the boll weevil crop pest, followed by the economic devastation
of the Great Depression, and mechanization in harvesting crops, wreaked havoc on the livelihood
of southern Black sharecroppers, prompting a “Great Migration” of African-Americans from
southern fields, through the expanding rail network, into northern and later western cities in
search of factory work (Lemann 1991). As Black families began settling in northern cities, they
had newfound political liberties, but were nonetheless confronted with hostilities not altogether
16

alien to those they confronted in the Jim Crow south, particularly when Black families
unwittingly disrupted the prevailing opinion that neighborhood stability depended on racial
homogeneity. To preemptively segregate arriving Black migrants from White neighborhoods,
cities first used racial zoning ordinances as an exclusionary tactic. This practice was short-lived,
and deemed unconstitutional in 1917, because of its overt discrimination (and lack of nexus with
the powers vested with cities to “zone” for land uses), and so following that, many property
owners placed racial covenants on their deeds to effectively achieve the same results albeit more
covertly. This exclusionary practice became widespread in White communities; the Mapping
Prejudice project at the University of Minnesota, for instance, has found over 20,000 deeds in
Hennepin County with racial covenants issued prior to 1948, the year they were struck down as
unconstitutional by the Supreme Court (Kaul 2019; University of Minnesota Libraries n.d.). The
National Association of Real Estate Boards promoted neighborhood exclusivity and, amongst its
broker members, actively discouraged sales to minority groups, believing that integration would
erode property values
2
(Rothstein 2017).
2.1.2. The Home Owners’ Loan Corporation and “Redlining”
The professionalization of the real estate industry and its coalescence into a powerful
lobbying group quickly led discriminatory practices to be adopted by the federal government. As
part of the range of programs created by the New Deal to help Americans contend with the
effects of the Great Depression, the HOLC was established by the federal government to help
homeowners refinance their mortgages and avoid defaulting. Shortly after its inception, roughly

2
From the ethics handbook of the National Association of Real Estate Boards, 1924: “A Realtor should
never be instrumental in introducing into a neighborhood a character of property or occupancy, members
of any race or nationality, or any individuals whose presence will clearly be detrimental to property
values in that neighborhood” (Kaul 2019).
17

between 1935 and 1940, agents of the HOLC devised a grading system to evaluate “residential
security” – essentially a proxy for the lending risk a bank would assume. Within most major
American cities, neighborhoods were carved up into A, B, C, and D categories, with associated
colors of green, blue, yellow and red. The “A” category (green) equated to the “best” housing,
the “B” areas (blue) meant “still desirable”, “C” (yellow) represented “definitely declining” and
“D” (red) meant “hazardous” (Jackson 1987; Nelson, et al. n.d.). The rating system accounted for
a variety of factors that customarily accompany real estate appraisal in defining its categories,
such as the type and age of construction, but also added extraneous factors like the racial and
ethnic composition of neighborhoods as part of the grading rubric. Pockets of Black
neighborhoods were almost uniformly given C and D ratings within cities, effectively barring
them from access to loans. The red borders that demarcated the D zones led to the pejorative
term “redlining” for any type of racially discriminatory land use or housing-related practice
based on a community’s demographics
3
. The urban historian Kenneth Jackson (1987)
summarized the situation thusly: The HOLC “devised a rating system that undervalued
neighborhoods that were dense, mixed, or aging…[and] applied these notions of ethnic and racial
worth to real-estate appraising on an unprecedented scale” (Jackson 1987, 197). While the
HOLC maps have become notorious, grading neighborhoods for underwriting was also common

3
Descriptions of each designation were written by HOLC agents to help local lenders and give an
unadulterated vantagepoint into their mentality. A “hazardous” D zone in Minneapolis was assigned as
such because:

At the present time, many Jews and Scandinavians and negroes reside in the easterly half of this
area. The westerly half has many of the shifting populations occupying the cheap apartments and
rows. Down in the southeast corner of the area near the Adams school there is a considerably
large negro settlement. The whole area is close to the business center of the city, it contains many
rooming houses occupied by salaried employees and laboring people because of the nearness to
the business district. The age is 20 to 50 years. (Nelson, et al. n.d.)

18

practice by banking institutions and other government lenders, such as the FHA
4
(Hillier 2003).
The interlocking levels of discriminatory lending practices meant that the stratification of 1920s
cities started to become ossified in the urban landscape, as there was a special pattern already
emerging: inner-city areas, near the core, often received C and D ratings, while peripheral areas
transitioning to suburbs, received A and B designations. Figure 9 shows this radial type of
pattern in Minneapolis and Baltimore, where risk is diminishing as one moves outward from the
center city.

Figure 9. HOLC maps from Minneapolis and Baltimore (Nelson, et al. n.d.)

4
In instructing the creation of maps to assess appraisal and risk throughout the US, staff economist of the
FHA, Homer Hoyt, explained:

‘The maps, if prepared carefully according to the suggestions now to be given, should lay the
groundwork for the rating of neighborhoods,’… [and then] instructed staff responsible for
creating the maps to draw lines in specific colors around certain types of areas. In the first set of
maps, red was to be used to mark off areas with concentrations of an “undesirable element” such
as distinct racial, national, or income groups. (Hillier 2003, 402)
19

2.1.3. Suburbanization and the Rise of Single-Family Zoning
The disparagement of dense, inner-city, minority neighborhoods by federal underwriting
manuals coincided with a championing of single-family detached homes in peripheral suburban
areas. These were typically located in the areas of cities that HOLC and FHA maps graded
“best” or “still desirable” areas, and appraisers often noted legal protections like the presence of
racial covenants or exclusionary local zoning ordinances as reasons for banks to extend credit
and assume low risk. While the Supreme Court had determined that it was illegal to explicitly
discriminate through zoning in the early 1900s, planners quickly sought to effectively achieve
the same results through physical criteria, like minimum lot sizes, and an extremely limited range
of permissible land uses (Kahlenberg 2019). By explicitly prohibiting nearly every type of land
use other than single-family homes on large lots from swaths of the city, a de-facto segregation
emerged, as the types of homes lower-income residents could afford, like duplexes, triplexes,
and rental apartments, were expressly prohibited. Since barriers to economic opportunity often
meant non-White racial populations were low-income renters and could not afford to purchase
single-family homes, these exclusionary practices were upheld in courts despite yielding the
same effects as the explicit discrimination deemed unconstitutional. They spread quickly to
zoning codes throughout the country
5
where they came to “protect the neighborhood character”

5
The landmark Supreme Court case that upheld zoning, Euclid v. Ambler, discussed apartment buildings
as follows:

…very often the apartment house is a mere parasite, … interfering by their height and bulk with
the free circulation of air and monopolizing the rays of the sun which otherwise would fall upon
the smaller homes, and bringing, as their necessary accompaniments, the disturbing noises
incident to increased traffic and business, and the occupation, by means of moving and parked
automobiles, of larger portions of the streets, thus detracting from their safety and depriving
children of the privilege of quiet and open spaces for play, enjoyed by those in more favored
localities – until, finally, the residential character of the neighborhood and its desirability as a
place of detached residences are utterly destroyed (Justice Sutherland, US Supreme Court 1926).

20

of large swathes of the peripheral areas of most cities, not to mention their suburbs, from the
encroachment of higher-intensity uses and the unseemly demographics associated with them
(Figure 10) (Rothstein 2017). Zoning ordinances, therefore, often worked in tandem with the
mortgage lending practices to completely alter the 19
th
-century urban fabric of major cities
6
.
They had the effect of simultaneously suburbanizing White residents into homogenous
communities and consolidating Blacks and other minority groups into denser, urban ghettos
without access to loans (Rothstein 2017; Hirsh 1998).

Figure 10. Land zoned for single-family homes in select US cities (Bui and Badger 2019)
2.1.4. Slum Clearance, Urban Renewal, and Suburban Mortgages
The Housing Act of 1949 was landmark federal legislation that accelerated these divisive
urban and suburban trends in three primary ways. First, Title I of the bill granted federal monies
towards slum clearance in urban renewal projects designated by US cities (Committee on
Banking and Currency, US Senate 1949). These areas were often designated in lower-income,
minority communities and were used for a variety of functions, such as to reinvest in new civic

6
In The Color of Law, Richard Rothstein notes that “the FHA had its biggest impact on segregation, not
in its discriminatory evaluations of individual mortgage applicants, but in its financing of entire
subdivisions, in many cases entire suburbs, as racially exclusive white enclaves. (Rothstein 2017, 70)”
21

amenities (such as Lincoln Square in New York, as shown in Figure 11), to build middle-class
housing (such as Stuyvesant Town in New York), to construct highways, (particularly after the
Federal Highway Act was passed in 1956), and to construct public housing (in concert with Title
III funding through the same legislation) (Rothstein 2017; Hirsh 1998). Even where families
were not displaced by urban renewal transformations, their effects could be pernicious;
highways, for instance, often divided previously vibrant minority neighborhoods, causing a
vicious cycle of economic and social side effects. Figure 11, for example, shows Interstate 95
being cut through Tremont in the Bronx (Avila 2014; Caro 1973). Title II of the Housing Act
expanded the mortgage insurance program which, coupled with the discriminatory lending
practices and exclusionary zoning, expanded home-ownership opportunities for mostly White
citizens by offering securitized down payments and low-cost mortgage rates (Rothstein 2017).

Figure 11. Images showing highway-building (Tremont, Bronx, 1949) and other urban renewal
impacts in NYC (Lincoln Square, Manhattan, 1969) (lbennett 2019; Williams 2017)

22

2.1.5. Separate and Unequal
In the decades after World War II, as returning veterans sought to start families, this
policy armature was in place and ushered in modest suburban enclaves throughout the nation and
lifted millions of Americans into the middle class by securitizing their appreciating home assets
(Jackson 1987). Yet for Black Americans, a dramatically different narrative emerged; they were
locked out of the lending markets, barred from this federally subsidized wealth accumulation,
and confined to certain areas of the city where the limited housing supply left them dependent on
public housing or vulnerable to unscrupulous landlords
7
and predatory lending schemes like
contract housing. Early post-war suburbs like Levittown, NY, on Long Island, as shown in
Figure 12, were affordable to returning veterans through the GI Bill. However, despite federal
underwriting, they still included racial covenants barring Black owners (Sheidlower 2020).
Factors such as overcrowding, minimal maintenance from absentee landlords (whether
private or public entities), and neglect from city administrators, often led to slum conditions in
Black neighborhoods, and Black residents were stigmatized as the initiators rather than the
victims of this circumstance. Public housing projects like Pruitt-Igoe, in St. Louis, also shown in
Figure 12, had little money budgeted towards maintenance and operation, and so slum conditions
emerged from disrepair (Newman 1976). This, in turn, made White enclaves more biased against
integration, fearing slum conditions would follow Black residents (Rothstein 2014). Real estate
agents preyed on these White fears, and used a tactic called “block-busting” to acquire properties
at a lower price from White sellers for fear of Black residents, and then resold them to Black

7
Ta-Nahesi Coates (2014) explains: “In Chicago and across the country, whites looking to achieve the
American dream could rely on a legitimate credit system backed by the government. Blacks were herded
into the sights of unscrupulous lenders who took them for money and for sport. ‘It was like people who
like to go out and shoot lions in Africa. It was the same thrill,’ a housing attorney [gloated]... ‘The thrill
of the chase and the kill.’”
23

families at much higher prices as contract housing (where no equity was obtained), as families
were so desperate to leave the ghetto (Rothstein 2017; Coates 2014).

Figure 12. Different trajectories stemming from the Housing Act of 1949 (Levittown, NY,
1950s, and Pruitt-Igoe, St. Louis, 1960s) (Sheidlower 2020; Newman 1996)
As the Great Migration continued in earnest, with five million Black Americans
migrating between the 1940s and 1970s, the multi-level, discriminatory government policy
framework was in full force and intensified the burgeoning schism. Attempts to course-correct,
such as through Civil Rights legislation, and the Supreme Court mandated desegregation of
school districts, prompted swift reaction from White suburbanites (Figure 13), and the lofty
ambitions of the 1960s Great Society began to fade as this cohort became an important voting
bloc. In what would become one of the last bold moves for suburban integration, President
Nixon’s Secretary of the Department of Housing and Urban Development (HUD), George
Romney, proposed denying federal investments in sewers, water projects, parks, or other
redevelopment in White suburbs unless exclusionary zoning tactics were repealed and subsidized
low- and moderate-income housing accepted (Lamb 2005; Rothstein 2015). In a foreshadowing
of the dramatic shift in the Republican Party platform, Romney was dismissed, and the
integration policy reversed in order to mollify Nixon’s [perhaps not so] “silent majority”
24

(Jackson 1987). Figure 13 shows White Bostonians protesting the court-ordered desegregation of
schools and White suburbanites in Warren, Michigan, a suburb of Detroit, protesting HUD
Secretary Romney’s proposed suburban integration plan (Detroit News 1970). Romney was
familiar with inner-city strife as he was the Governor of Michigan during the 1967 race riot in
Detroit, and suburban residents were also familiar with him.
The exodus of White people from inner city areas to the suburbs – “White flight” –
continued into the 1970s and 80s, and as their associated wealth fled with them, the tax bases of
many cities plummeted, spurring a decline in the quality of municipal services. As crime rose,
schools and other municipal services deteriorated, and soon a vicious cycle emerged wherein
people of means left the city, leaving municipalities to grapple with more problems with fewer
resources, which in turn spurred further decline and exodus (Hirsh 1998). Drug epidemics in the
1980s and 90s further crippled these disenfranchised communities. Reciprocally, suburban
resources increased in a virtuous cycle, as the leafy, low-crime oases with high-performing
schools beget more transplants, aided by new federal highways, and a steady migration of
businesses and manufacturers to cheaper exurban land (Sugrue 2005).

Figure 13. Images of White suburbanites resisting integration
(Hannah-Jones 2019; Detroit News 1970)
25

In 1967, while the National Guard had been called in to quell a race riot in Detroit, a
Commission was assembled by President Johnson to try and understand the underlying psyche
leading to the string of rioting and looting being incited primarily by young Black men in cities
throughout America. While many expected an indictment of the Black “agitators”, the Kerner
Commission instead laid the blame squarely on White America, bluntly stating that the reason
for economic strife and social discontent amongst the Black in inner-city ghettos was a direct
consequence of the White privilege concentrated in the suburbs:
Our nation is moving toward two societies, one black, one white – separate and
unequal…What white Americans have never fully understood – but what the Negro can
never forget – is that white society is deeply implicated in the ghetto. White institutions
created it, white institutions maintain it, and white society condones it (The National
Advisory Commission on Civil Disorders 1968, 1).

Over fifty years later, the warning of the Kerner report remains largely unheeded, and
nearly every new episode of violence in Black communities reignites a probe into the underlying
structural inequality dividing Black and White America. Understanding the pernicious side
effects of redlining and other discriminatory tactics in American cities is vitally important if
progress is to be made in desegregation, and this thesis aims to contribute to that knowledge -
base.
2.1.6. New York Backdrop
New York is heralded as a bastion of liberal politics and progressive policies, but beneath
this veneer, the city has struggled with race for centuries.
Central Park, one of the largest municipal infrastructure projects ever contemplated, is
one of the city’s most defining features. Yet, when the Olmstead and Vaux plan was selected as
the winning entry to steer the park’s design in the 1850s, the northern portion of the land to be
condemned was home to a thriving Black community, Seneca Village, whose residents had
26

sought refuge there to escape persecution further downtown. The residents, who had constructed
roughly fifty homes, three churches, and a cemetery, were displaced entirely by 1857 (The
Central Park Conservancy 2018).
Less than a decade later, during the American Civil War, the city’s working-class, mostly
immigrant residents in Lower Manhattan rioted over being conscripted to fight in the war, and in
the process, the event degenerated into race riot, with mobs burning the homes, and inflicting
violence and death on innumerable Black residents. While this obviously was not the result of a
particular land use policy, it represented the largest insurrection in American history to that point
– apart from the Southern secession itself. The violent racism of its residents instilled such fear
and horror that many Black residents relocated to other parts of the city (Foner 1988).
In the late 1800s, a new Black enclave emerged on the fringes of the vice district of the
day, the Tenderloin. In 1900, a race riot emerged when a police officer was struck and killed by a
Black man intervening in the arrest of a Black woman for soliciting. White gangs burned down
Black homes and attacked Black residents. Within a decade, the city condemned four blocks
housing predominantly Black populations to construct its second rail station, Pennsylvania
(Penn) Station, displacing residents to other Black enclaves in the city (Jonnes 2008).
In the 20th century, a discussion of New York’s planning narrative, and the intersection
of federal and local policies, would not be complete without invoking Robert Moses and Jane
Jacobs. Moses, a city and state bureaucrat, embodied top-down urban planning, making
sweeping urban changes to the city without the input of those impacted, while Jacobs, an activist,
represented a bottom-up approach, where officials would make informed decisions from
community participation (Gratz 2010). Their powerful, and often opposing ideologies, have not
27

only had a lasting influence on the city, but embody a shift in larger planning mentalities over
time.
Robert Moses’ life and contribution to the city – both for good and ill – have been
meticulously documented by his biographer, Robert Caro. The emergence of government
infrastructure investment programs like the Works Progress Administration during the Great
Depression created a perfect scenario for this visionary, realpolitik power broker to rise through
the ranks of state and city governments. Moses was a fixture in nearly every agency involved
with physical planning in the region without ever being elected to office. He ran the Parks
Department, was a City Planning Commissioner, and otherwise operated as an infrastructure
czar, overseeing the construction of bridges, parkways, highways, stadiums, and public housing
(Caro 1973). This authority gave him an outsized influence over urban renewal policies and their
on-the-ground manifestation across the city, and he is known for his callous displacement of
minority communities
8
(Boeing 2021; Caro 1973).
His decrees were not always implemented. Between the 1940s and 60s he issued several
plans for a Lower Manhattan Expressway that would connect the Holland Tunnel on the west
side with the Manhattan and Williamsburg Bridges on the east side (Figure 14). While this would
efficiently connect New Jersey with Brooklyn and, through other highways, Queens, and greater
Long Island, it would displace residents in SoHo, Little Italy, the Lower East Side, and
Chinatown. Nearby residents in Greenwich Village would also be impacted, and activists like
Jane Jacobs (Figure 14) made resistance a cause célèbre (Chatelain n.d.).

8
In response to Robert Caro’s seminal biography, and the criticism of his legacy, Moses said: “I
raise my stein to the builder who can remove ghettos without moving people as I hail the chef
who can make omelets without breaking eggs” (Boeing 2021, 1).
28

Figure 14. Proposed location of, and protest against, the Lower Manhattan Expressway
(Sagalyn 2016; Seamon 2019)
The South Village also had a Slum Clearance and redevelopment plan proposed by
Moses under the auspices of the Title I of the Housing Act of 1949, wherein several blocks south
of Washington Square Park would be cleared for housing projects (The Committee on Slum
Clearance Plans 1951). Jacobs, in her seminal Death and Life of Great American Cities, which is
a foundational book for urban planners, delivered a blistering critique of monolithic superblocks,
and the banal residential architecture characterizing many urban renewal redevelopments.
Instead, she championed four principles that underpin a healthy neighborhood: a mix of uses to
encourage day-round vitality; short, walkable blocks; a varied building stock of old and new
buildings; and a healthy density to stimulate interactions (Jacobs 1961). The underlying
anecdotes that grounded her practical urban sociology were largely derived from her home in the
Village and were deeply critical of not just the urban renewal in her neighborhood, but what she
saw as a mistaken pedagogy of urban planning in general.
The narrative of Jane Jacobs’ success in stopping Robert Moses’ plans in the Village is a
well-trodden David versus Goliath type of story, which fueled a new generation of community
activism to try and shape local land-use politics (Gratz 2010). What is less explored in the
29

literature is the reasonings for targeting the Village for renewal in the first place. While
significant areas of the Upper East Side, the West Village, SoHo, and Lower Manhattan around
the CBD are some of the most affluent zip codes in the nation today, they were redlined in
HOLC maps and subsequently designated for slum clearance because the primary building stock
were aging tenements that possessed substandard access to light and air. Figure 15, for example,
shows a description from the West Village and one from Bedford-Stuyvesant (Nelson, et al.
n.d.). Notably, there are no “Negro” inhabitants in the West Village; instead, inhabitants are
comprised of a mix of white and blue collar, foreign-born and native-born, White residents. The
detrimental influences are simply the age and obsolescence of the building stock and lax zoning
regulations. This contrasts with the “infiltration” by African Americans in central Brooklyn.
Little literature has explored the connection between redlining maps, the slum clearance
proposals (such as in Figure 16), the actual racial composition, and how White communities that
were redlined have fared in subsequent decades contrast to Black or other minority communities
that were redlined.

30

Figure 15. HOLC D area designation writeup for West Village (top) versus
Bedford-Stuyvesant (bottom) (Nelson, et al. n.d.)

Figure 16. HOLC maps of Manhattan compared with the Slum Clearance boundaries
(Nelson, et al. n.d.; Laurence 2006)
NYC’s Zoning Resolution is one of the principal planning tools of the Department of
City Planning (DCP). The Department’s 1961 Zoning Resolution was a comprehensive rewrite
of the 1916 Zoning Resolution and was decades in the making. It sought to vastly simplify the
31

districting framework of its predecessor while also reducing the overall zoned capacity and
increasing open space (NYC DCP 2018). However, because the Zoning Resolution was the
result of decades worth of planning work, the underlying presumptions characterizing
development allowances, height, and setback regulations, and nearly all building bulk
regulations, were generally that of the ‘tower-in-a-park’ campuses typified by urban renewal
plans. This large-scale, super block framework, with towers set far from the street amidst open
space, contrasted in nearly every way with the existing character of most transit-adjacent
neighborhoods, where the building stock was defined by more granular 25-40’ wide tenement
buildings that were comparatively squat and hugged the sidewalk (NYC DCP 2018).
As urban renewal became reviled, accelerated by the growth of the historic preservation
movement (particularly after the loss of Penn Station in 1963), piecemeal modifications to the
Zoning Resolution were adopted in the form of alternative zoning districts, called “contextual
districts.” As the name suggests, these rules sought to ensure new development would mimic the
built pattern of well-established neighborhoods and differentiated themselves from the original
1961 zoning districts by having fixed height limits and a squatter built form (NYC DCP 2018).
While the negative reaction by community activists to urban renewal, and its zoning
affiliates, was legitimate, problematically, as the acuteness of government overreach diminished,
and the propensity to raze neighborhoods waned, the affinity for these contextual districts has
persisted. Over several decades, different coalitions of affluent and privileged stakeholders
became increasingly vocal about any change, saying “not in my backyard” about new
construction (Adams 2022). These NIMBY-ists were intent on preserving the “character” of a
neighborhood by shrink-wrapping the code around its existing form. Contextualization, at nearly
every density, has been one of the major themes to zoning amendments after 1961, especially as
32

district tools became more widely available in the late 1980s (NYC DCP 2018), and has been a
tool used by the Department as readily as communities themselves (Laskow 2014). Perhaps
unsurprisingly, as layers of land use regulations have increased, and more limitations on where
new housing can be built have been added, housing prices in New York have drastically
increased.
The Zoning Resolution has been amended a dizzying number of times (NYC DCP 2018).
This fact, combined with the complexity of the code, and difficulty in teasing out definitively
whether a modification increased or decreased zoned capacity, by changing the zoning map or
the zoning text, makes it difficult to evaluate if changes that have occurred to the Zoning
Resolution have disproportionately benefited White communities. For example, have legislative
changes in White affluent areas led to an outsized reduction in development capacity (by virtue
of a reduction in zoned rights (a “downzoning”)? Reciprocally, if housing supply is heavily
restricted in these affluent White areas, but demand to live in New York persists, has this led to a
disproportionate reliance on lower-income minority neighborhoods for shouldering a
disproportionate share of growth? Acknowledging housing supply is complex, and there are
exogenous inputs impacting NYC, including regional housing supply, there is little research
connecting the genesis, and effects of zoning modifications (e.g., the elasticity in housing
supply) with race and privilege.
The 2020 Census shows drastic demographic shifts in traditional Black communities like
Harlem, Crown Heights, and Bedford Stuyvesant, as can visually be seen in the maps in Figure
17. The loss of Black population in these neighborhoods corresponds very closely with
corresponding White gains. Overall, there was nearly a five percent loss in the Black population
between 2010 and 2020 (NYC DCP 2021) .
33

Figure 17. Change in Black and White population between 2010 and 2020
Given the tremendous loss of Black residents in the last ten years, despite the growth of
the city, consideration must be made for one of the more controversial urban phenomena of the
twenty-first century – gentrification. This phenomenon most traditionally involves wealthier
residents moving into lower-income areas, and, because they bring greater disposable incomes,
they accelerate neighborhood change (Freeman 2006). Retail store fronts change (with higher
priced goods), rents can be raised, and soon lower income populations are often undergoing
housing stress and at risk of being displaced (Freeman 2006).
In cities across America, this trend has more recently involved a racial component –
White residents moving into traditionally Black neighborhoods – as part of the renaissance many
cities have experienced since the late 1990s (Chronopoulos 2020). The trend is sometimes
doubly painful in cities like New York because it is occurring within the very same
neighborhoods that Whites abandoned in the 1960s and 70s (Chronopoulos 2020). While a
cursory glance at census data might reveal a seemingly integrated neighborhood, it could be
cloaking a prolonged segregation that is now in transition and masking Black displacement.
34

While gentrification has certainly received considerable attention, there has been little
research at a granular level that seeks to connect the correlation of planning policies in these
recent demographic shifts; exploring whether these land use regulations are new covert stand-ins
for overt segregation in housing policy 100 years ago, with a similar callousness in perpetuating
displacement.
2.2. Literature review
Several pieces of prior research informed different portions of this thesis; literature was
surveyed to inform the background understanding of planning policies and pertinent variables to
assess, while other literature informed the spatial research methods.
2.2.1. Literature Informing the Historic Backdrop and Variable Selection
The disparate treatment of Black population segments in American cities is a topic that
has spawned considerable discussion by other researchers. Many authors have researched the
very tactile repercussions of spatial segregation in the urban landscape, particularly homing in on
and quantifying the disparity between Black communities and other neighborhoods.
Accessibility studies have shown unequal access to quality schools, park space, mass transit,
healthcare, and jobs, while public health literature has documented higher prevalence of obesity
due to insufficient fresh food access and higher rates of asthma due to proximity to highways.
Tracking the perniciousness of housing policies and land use regulations on Black
communities is perhaps more difficult, as it requires sufficient knowledge of their historical arc,
and interactions with local regulations, institutions, and prominent individuals. Because of this,
much of the literature is difficult to communicate sufficiently in a paper; the core texts are often
substantial volumes.
35

A significant amount of literature is devoted to examining the origins of segregation in
the American city and the role various government policies played. This backdrop is important to
understand the complexity of race in the American city, but also to ensure the appropriate
variables are being considered for their role in explaining neighborhood segregation.
Kenneth Jackson, in his seminal book Crabgrass Frontier (1987), was the first scholar to
rediscover the discriminatory HOLC maps, connect this lending risk assessment to FHA home
financing, and assessing the power these tools collectively had in shaping the post-war American
city. Richard Rothstein, in his the equally important Color of Law, explores the cumulative effect
of redlining, zoning ordinances, restrictive covenants, urban renewal designations, and other
racist government practices in shaping the Black experience. These texts are vital to this research
for their tremendous historical sweep and topical range.
Many authors have addressed singular topics concerning the plight of Black Americans in
the 20th-century city. Charles Abrams (1955) and Arnold Hirsh (1998) have written extensively
on mid-20th-century housing policy and the evolution of New Deal era policies into hyper-
segregated federal housing projects. Charles Lamb’s (2005) research extends the arc of this
housing lens from the Fair Housing Act of Johnson’s Great Society into the Nixon era, and the
slow unraveling of Johnson’s Great Society’s laudable goals. David Freund (2007) explores the
sociological shift in suburban mentality that enabled this – the suspension of disbelief where one
could simultaneously embrace civil rights, but oppose integration in their own backyard. Ta-
Nehasi Coates (2014) discusses the rapacious tactics of real estate brokers in blockbusting White
neighborhoods, and reselling homes to Black residents through predatory lending practices like
contract housing. Massey and Denton (1998) tell the historical sweep of these policies, their
36

contribution to segregation, and even ‘hyper-segregation’ and detail the sobering sociological
implications for the Black community.
Many authors have also discussed the racial and exclusionary issues embedded within
zoning regulations. Charles Haar (1989) describes the origin of zoning regulations in the United
States, and the diverse range of issues that advocates of zoning sought to resolve. While some
reformers legitimately sought to protect the interests of the city’s most vulnerable by ensuring
basic rights like access to sunlight and fresh air in the dense immigrant enclaves of New York or
separate validly incompatible uses under the guise of public safety, other influential forces tried
to use zoning as an exclusionary device to protect property values, couching the protection of the
city’s tax base as a valid public purpose (Haar and Kayden 1989). Christopher Silver (1997)
discussed the increasing boldness of abusing zoning in the late 1800s and early 1900s, from
California’s prohibitions on Chinese laundromats as a means of discriminating, to the eventual
explicit prohibition of persons of color from certain sections of cities through racial ordinances.
Interestingly, Silver notes that while racial zoning was struck down by the Supreme Court in
1917, many Southern cities continued to adopt and enforce new legislation afterwards;
Birmingham, Alabama illegally enforced their code as late as 1951. Other literature has made
some of the more covert exclusionary zoning practices, like single-family zoning and minimum
lot sizes, so well-connected with their racist history that some planning departments and local
politicians are summoning the courage to repeal them, and communicate their historical missteps
(DC Office of Planning 2020; Bureau of Planning and Sustainability 2019).
Urban renewal policies relating to highway construction have also been addressed
topically. Eric Avila has written on the disproportionate targeting of low-income neighborhoods
for slum clearance and highway construction and unsuccessful community mobilizations and
37

protests to stop the destruction of communities of color (Avila 2014). Robert Caro (1973) has
chronicled similar histories in New York under the supervision of the powerful urban planner
Robert Moses, and in detailing the construction of the Cross-Bronx Expressway detailed the
singular commitment to removing minority residents, even when less destructive and cheaper
options were available. In evaluating slum clearance designations in five cities, Miles Miller
suspects racist motivations for slum clearance designations for similar reasons; neighborhoods
acquired through eminent domain routinely exceeded slum standards, and often had higher
property values than surrounding areas (Miller 2018). David Karas has similar conclusions as
Miller, that while the Interstate Highway System ushered in tremendous growth, a tepid
approach to civil rights by the Eisenhower administration allowed the blatant targeting of Black
communities for highways (Karas 2015).
Lastly, literature discussing the origins of urban rioting in the late 1960s is useful in
understanding the practices that led to civil unrest. Thomas Sugrue explored the conditions that
led to the 1967 race riot in Detroit, finding that housing prejudice, deindustrialization,
automation, shifting access to capital and job discrimination all played a role (Sugrue 2005). The
Kerner Commission similarly explored the race riots of the late 1960s and found the origins of
urban crises for Black Americans were the economic opportunity, social equity, and public
safety that was systematically deprived to them by White racism (The National Advisory
Commission on Civil Disorders 1968).
While many of this literature will use examples or anecdotes to make a particular point,
or underscore a particular injustice, there are few examples that concretize the cumulative effects
of decades of racist, or otherwise exclusionary policies in a particular location, and examine the
transformation, or persisting segregation, at a neighborhood scale.
38

2.2.2. Literature Informing Spatial Research Methods
2.2.2.1. Related research methods
Notwithstanding this rich literature on racist government land use and lending practices,
there is little research devoted to understanding how their spatial implications continue into the
present. The digitalization of HOLC maps, by the University of Richmond’s Digital Scholarship
Lab, as part of the Mapping Inequality project (Nelson, et al. n.d.), offers a significant
opportunity to analyze the spatial impacts of redlining in cities across the US, but since its
vintage is 2015, it is still a relatively new resource. In an interesting working paper released in
fall of 2020, Fishback et al., digitized 1930 Census maps in ten northern cities and compared
these to the HOLC maps, which were created between 1937 and 1940. They found that 97
percent of Black individuals and 95 percent of Black-owned homes existing in 1930 were
captured in the D designations and argue that the maps were not necessarily the genesis of the
spatial segregation within cities, but rather, they were documenting an already fragmented
metropolis (Fishback, et al. 2020). Amy Hillier (2003), in researching the influence of HOLC
maps on redlining practices in Philadelphia, came to a similar conclusion – that the HOLC maps
in and of themselves were not causing redlining and disinvestment in cities. She noted that bank
lenders were denying loans to Black residents prior to HOLC maps, real estate agents were
acutely aware of the demographic patterns even without the maps and could steer buyers without
them, and that the HOLC maps were not dispersed widely to local real estate agents and lenders

9
. If there was a longer research arc, investigating other primary source materials with spatial

9
In discussing the HOLC maps, Hillier writes:

“…the map provides evidence that ecological and infiltration theories, racial prejudice, and real
estate and appraisal industry codification of all these sentiments in combination with federal
endorsement and promotion of them—not the maps, themselves—caused urban decline. The

39

implications, such as racial zoning ordinances, racial restrictive covenants, and other legal
instruments that would have begun segregating cities prior to HOLC maps, would be an
important research component.
Regression tools have been utilized to connect the role of historic HOLC maps to present-
day disparities in quality-of-life metrics. White, Guikema and Logan (2021) surveyed 13 US
cities and using logistic regression and K-means clustering techniques, tried to discern the
correlation between historic HOLC boundaries and present health, employment, education, and
income measures. They consistently found that the inequities etched in the modern-day urban
fabric align with the historic spatial boundaries of discrimination. Comparing the green, A-rated
areas with the red, D-graded areas, they found that the population presently associated with the
historic A areas in all 13 cities had better health outcomes, and reciprocally, the D areas in all
cities had higher poverty rates and lower high-school graduation rates. The population in D areas
in 12 of 13 cities had lower health insurance coverage rates, while 10 of 13 had higher
unemployment rates.
Regression analyses have also been utilized to discern explanatory variables for
segregation. Yu and Wu (2013) used Landsat, remote-sensing imagery to extract biophysical and
textural information, in particular, vegetation, impervious surface, and soil conditions, which in
turn were used to interpolate an array of land uses. For example, pixels where vegetation is low
and imperious surface most strongly correlate with commercial areas, so that land use might be
assigned depending on the specific mix. The authors then used the derived land use cover to seek
to explain segregation patterns in Milwaukee with OLS and geographically weighted regression

HOLC maps are probably the clearest, most accessible, and most dramatic evidence of this
collusion, but that does not make them the most influential. (Hillier 2003, 413)”

40

(GWR) techniques. The GWR analysis found a large presence of high-density land cover and a
small presence of low-density land cover to be the strongest explanatory variables in predicting
the spatial concentration of Black people. Ogneva-Himmelberger, Pearsall, and Rakshit (2009)
similarly explored the relationship between land cover and socio-economic status in
Massachusetts using GWR and found higher amounts of impervious surfaces were a strong
predictor of higher percentages of both minority population and households living in poverty;
they also found, somewhat reciprocally, that smaller amounts of imperviousness predicted higher
home values.
Several authors have explored the correlation between housing supply, land use
constraints, and housing affordability, both in New York and across the US more broadly.
Glaeser and Gyouko (2003), for instance, studied housing markets throughout the US, and found
that the majority of regional markets behaved as one would expect, with housing sale prices only
slightly exceeding construction costs. In a few urban markets, however, disproportionately
clustered on the east and west coasts, sale prices vastly exceeded construction costs, and in
probing further they found a high correlation between housing prices and overly burdensome
land use regulations (Glaeser and Gyourko 2003).
Glaeser, Gyouko, and Saks (2005) further explored affordability in the context of
Manhattan, where severe land constraints necessitated high-density towers long ago. While this
makes construction much more expensive than a single-family home, they note an important
distinction: land preparation costs in multi-family construction, including excavation and
foundation costs, only need to be done once, so in a perfect economic model, a developer would
add the marginal cost of adding an additional floor until the building rose to a height where the
sale price per square foot slightly outpaced costs, to ensure profitability. However, paradoxically,
41

they found that the percentage of residential buildings over 20 stories that have been constructed
in Manhattan has steadily dropped since 1980, suggesting another factor artificially reducing
building height, and indirectly, housing supply. The authors place this blame squarely on the rise
of community activism in clamoring for shorter buildings and more restrictive land use
regulations (Glaeser, Gyourko, and Saks 2005).
While correlations between land cover and uses have been correlated with race, and
zoning correlated with price, there seems to be research gaps in directly exploring the
relationship between zoning designations and segregation. Moreover, there also seems to be a
literature gap exploring, through regression analyses, the cumulative extent to which government
policies – directly or indirectly – explain the current spatial distribution of minority communities
within select cities. Similarly, there is little literature exploring, at a quantitative level whether
there are other interlocking factors, like economic status, social preferences, urban structure, and
mobility patterns, at play in determining residential housing patters, as Clark (1986) contended.
Perhaps more importantly, there are few studies that have embraced the scope of evaluating how
spatial patterns have ebbed and flowed through time. Lastly, while progressive leaders have
made attempts at undoing the more infamous types of government interventions – removing
highways (Popovich, Williams, and LuMay 2021), removing single-family zoning ordinances
(Bui and Badger 2019), or demolishing public housing – there is little literature on the
effectiveness of these restitutionary interventions in reducing segregation. These gaps all present
promising pathways to contribute novel research and analysis.
2.2.2.2. Studies of population patterns
Spatial autocorrelation analyses, at their core, invoke Tobler’s first law of geography, that
nearer things are more closely related than further things (O'Sullivan and Unwin 2010). These
42

analyses can be conducted in a global sense – where values in a given feature class are compared
across an entire study area, or in a local sense, comparing values against those of neighboring
polygons. Each has their own value, as a global study helps benchmark values of concentration,
which helps understand the degree of localized deviation higher or lower than that level
(O'Sullivan and Unwin 2010).The global Moran’s I test (Spatial Autocorrelation in ArcGIS Pro)
is a global test of spatial autocorrelation, and can be used to determine if the concentrations of
values in polygons are statistically significant. The outputs are entirely numerical and involve
evaluating p-values and Z-scores. P-values indicate the probability that the results are random –
lower resulting values indicate a lower probability that the result is random and greater
probability that the result is statistically significant (values less than 0.1 have a 90% confidence,
values less than 0.05 have 95% confidence and less than 0.01 have 99% confidence). Z-scores
are standard deviations from what result would be predicted under a random distribution. Results
with high or low Z-scores are at the ends of a normal bell curve and would exhibit statistically
significant clustering or dispersion (Esri n.d.). The analysis also outputs a Moran’s Index value,
where a positive value indicates clustering, and a negative value indicates dispersion. Since this
is a global test, there is one value output for the study area.
The Getis-Ord Gi* test (Hot Spot Analysis tool in ArcGIS Pro), is a local measure of
spatial autocorrelation that allows one to reconceptualize the spatial relationship between
neighbors and determinines if an there are ‘hot spots’ – areas of statistically significant clustering
– occuring within the subject study area (Esri n.d.). Based on the p-values of and z-scores of an
areal unit and its neighbors, the tool generates a map showing not only hot spots of clustering,
but also cold spots of dispersion, as well as areas that are not statistically significant and more
randomized. Within the hot and cold spots, there are confidence levels at 90, 95, and 99%
43

thresholds that the results are statistically significant, depending on the p-value. Another local
test, the federal Moran’s I test (Cluster and Outlier Analysis tool in ArcGIS Pro), takes the
premise of a Moran’s scatterplot, and depicts relationships between an areal unit and its
neighbors as high-high, low-low, or clusters and dispersions again, and importantly, also shows
outliers – units that have low-high or high-low values relative to their neighbors (Anselin 1995).
A high-low outlier designation, for example, would mean that the particular areal unit has a high
Moran’s I value relative to the surrounding units.
2.2.2.3. Segregation studies
Measuring segregation has long been a topic of inquiry for sociologists, economists,
geographers, and other social scientists to understand the effects of the striation of people in
different settings, and its intersection with access to quality educational, employment, health
outcomes and other socio-economic opportunities. Indices have been developed over time to try
and grapple with the core issues underpinning segregation, and its deleterious impacts in civil
society, in order to quantify them in localized settings. Most of the segregation research has
grappled with residential segregation by race, but segregation has also been assessed based on
income, age, religion, gender, language or numerous other characteristics, in the context of
schools, workplaces, and other physical locations (Oka and Wong 2019).
Prior to the 1950s, there was no prevailing approach to measuring segregation amongst
social scientists. Duncan and Duncan (1955) changed this, by convincing demonstrating that the
index of dissimilarity captured much of the information offered by other indices, ushering in
over 20 years of relative academic consensus (Massey and Denton 1988). The index of
dissimilarity measures the evenness of population groups across a study area, and the index
values interpreted as the proportion of the population that would need to move to obtain an even
44

distribution (Duncan and Duncan 1955). This was upended, indirectly, in the mid-1970s with a
critique of the dissimilarity index by Cortese, Falk, and Cohen (1976), which, instead of
mustering debate on the index itself, reopened the larger narrative of appropriate measures. The
sociologists Massey and Denton (1988) offered a pivotal framework after over a decade of
spirited debate. They analyzed over 20 different indices and determined that segregation has 5
major dimensions: 1) evenness, the relative distribution of population groups across a space; 2)
exposure between major and minor groups; 3) concentration, the physical space occupied by a
population group; 4) centralization, the degree to which a group is near the geographic center of
an area; and 5) clustering, the degree to which areal units occupied by minority groups adjoin
one another (Massey and Denton 1988). In this framework, evenness and exposure were aspatial
dimensions (even though they invoked areal units with an implied spatiality), and concentration,
centralization and clustering were more explicitly spatial (Reardon and O'Sullivan 2004).
Reardon and O’Sullivan (2004) posited that aspatial dissimilarity indices are often
fraught, as they fail to account for the arrangement of population within an areal unit, and in
relation to its neighborhors, invoking not only the modifiable areal unit problem (MAUP) but
also the “checkerboard problem”. The MAUP is invoked when point data is arbitrarily
aggregated into larger areal units, potentially masquerading more granular patterns. In this case,
a census divison could aribtrarily divide a cluster of a minority, but because it is fragmented, and
diluted with other population groups, the pattern might be concealed (Reardon and O'Sullivan
2004). The checkerboard problem, as shown in Figure 18, refers to a situation wherein a
completely segregated hypothetical grid would generate the same dissimilarity index as a more
integrated checkerboard, because it is failing to account for the pattern of its neighbors (Reardon
and O'Sullivan 2004).
45

Figure 18. Illustration of the checkerboard problem (Katumba, et al. 2021)
Building on the idea that every segregation measure should be spatial, they proposed a
modification to the five Massey and Denton dimensions, simplifying it to two major dimensions.
They argued that aspatial evenness and spatial clustering addressed the same issue and were
better characterized as two ends of a single spectrum instead of two separate indices. To this they
layered in Massey and Denton’s isolation and exposure dichotomy, but in a spatial manner, and
built a conceptual matrix of evenness (clustering) and exposure (isolation) shown in Figure 19.
The top right quadrant, with evenness and exposure, represents an idealized integration while the
bottom left, with clustering and isolation, represents larger segregation (Oka and Wong 2019).

Figure 19. Dimensions of spatial segregation (Oka and Wong 2019)
46

Reardon and O’Sullivan also proposed spatialized versions of several existing measures,
including the spatial dissimilarity index (D), the relative diversity index (R), the spatial
information theory index (H), and an exposure / isolation index (P*). Evaluating these and other
indices through a series of conceptual and mathematic rubrics, they determined the spatial
information theory index (H) and the spatial exposure / isolation index (P*) to be most faithful to
their ideas. The spatial information theory index measures the level of local diversity relative to
the total population of the region, a type of evenness measure; maximum segreation would be
equal to 1, while complete integration would be 0 (Reardon and O'Sullivan 2004). The spatial
exposure / isolation index tabulates the relationship of a particular race to other races (e.g. white
to white, white to black, black to white and black to black) to determine if localized populations
of a population are segregated and spatially isolated from other races, or if they are integrated
and exposed.
O’Sullivan worked with Hong in devising an R package which allows these spatial
segregation measures to be tabulated with relative ease by inputing demographic data and a
spatial geometry (Hong, O'Sullivan, and Sadahiro 2014). This package, in turn, has enabled
greater usage of segregation tools in an array of applications. Katumaba, et. al. (2021) for
instance utilized these spatial segregation measures to evaluate whether policy changes in post-
apartheid South Africa have had a meaningful difference in integreation. They found that since
1996, residential segregation has steadily declined, and exposure of Whites to Black Africans has
increased.
These spatial segregation measures proposed by Reardon and O’Sullivan are all global
measures – they generate a singular set of values for a study area. While this is useful for
comparing the relative segregation of cities or other bounded areas, it does not produce local
47

calculations which might allow intra-urban assessments. Wong (2002) introduced local measures
of spatial segregation, as conceptually similar to local measures of spatial autocorralation
developed by Anselin (1995). Oka and Wong (2014) have practically applied these measures, to
evaluate localized segregation levels in Washington, D.C., St. Louis and Chicago, through a
local spatial dissimilarity index and local spatial isolation index, in order to correlate with public
health indicators. These local indices, however, do not have an associated R package, and despite
the similar (or greater) potential for cross-disciplinary utlization, there has been limited
application.

48

Chapter 3 Methods
This thesis had two principal research questions: 1) to what degree have Black populations
clustered and been segregated from White populations within the City of New York from 1910
to 2020? and 2) what variables might explain any identified patterns of clustering and
segregation in three time-windows: 1960, 1990 and 2020? Intrinsic to each of these two
questions was a spatio-temporal query as to how these patterns have changed over time.
The methods employed to answer these questions are two-fold. The first task was more
investigatory. Spatial autocorrelation tools were deployed to find hot spots of Black residents,
while spatial segregation indices were calculated to determine the level of integration between
racial and ethnic groups. These cluster and segregation analyses were completed decennially
beginning in 1910 to discern overarching spatial trends, pivotal time periods, and outlier
geographies, all of which can validate and further inform regression variables.
The second task was to analyze the factors that predict population clusters using
regression analyses. This included querying the effects of historic boundaries of the HOLC
maps, urban renewal legacies like public housing concentrations and highways, zoning district
typologies, and historic districts. It then compared the influence of these government-oriented
variables in explaining Black and White population patterns with more organic market forces for
population concentration, like proximity to a subway station or distance to a CBD. This second
task was conducted longitudinally, in 30-year intervals, beginning in 1960, to determine how the
influence of different variables has shifted over time. It was conducted using a variety of
different Black and White population metrics as response variables to compare and contrast the
influence of the different planning-related explanatory variables.
49

3.1. Data Description and Preparation
A series of clustering and segregation metrics were chosen as dependent variables to test a series
of independent, planning-related variables. The thought was that each would provide unique
insight into different relationships between explanatory variables that other response variables
didn’t fully elicit. Independent variables were comprised of a series of federal and local
planning-related factors, based on the literature, which evaluate their influence on the relative
clustering and segregation of the New York population. This section describes these variables
and the data that represent them, as well as the necessary preparation of the data for usage in the
research tasks is also described.
3.1.1. Dependent Variables
All of the measures of the dependent variables began with population data from the 1960,
1990, and 2020 census. This historical census data, from the IPUMS National Historic GIS
database, included spatial and tabular census tract information for the entire nation, so
preparation included clipping the datasets to NYC, joining the spatial data with tabular data
(conveniently using the ‘GISJOIN field’), and then changing the projection from a continental
projection – the USA Contiguous Albers Equal Area Conic projection – to a more localized State
Plane projection.
New York’s State Plane coordinate system takes into account its wide east-west
dimension, and, to minimize distortion, subdivides the state into East (3101), Central (3102) and
West (3103) zone, each of which uses a transverse Mercator projection. A fourth zone, the Long
Island zone (3104), which encompasses NYC and two other counties of Long Island (Nassau and
Suffolk), accounts for the fact this geography is largely an appendage at the southern end of the
state, and so uses a different projection method, a Lambert Conformal Conic projection (The
50

Legislature of the State of New York 1995). All data was projected to use this Long Island State
Plane projected coordinate system (PCS), specifically, NAD 1983 StatePlane New York Long
Island FIPS 3104 (US Feet).
The data sources, their original type and projection, are all included in Table 1. Each of
these population-oriented variables is further described after the table, along with the process of
preparing the data.
Table 1. Response variables in regression analyses
Criteria Source

Type Original projection

Black and White
population density, by
census tract
US Census Bureau
National Historic GIS
Data Finder
(IPUMS National
Historical GIS n.d.)
Generated from tabular
demographic data and
joined to polygon
census tract shapefiles
USA Contiguous
Albers Equal Area
Conic
Black and White
population percentage,
by census tract
US Census Bureau
National Historic GIS
Data Finder
(IPUMS National
Historical GIS n.d.)
Generated from tabular
demographic data and
joined to polygon
census tract shapefiles
USA Contiguous
Albers Equal Area
Conic
Black and White
population density
change, by census tract,
in thirty-year intervals
US Census Bureau
National Historic GIS
Data Finder
(IPUMS National
Historical GIS n.d.)
Generated by
intersecting previously
joined tabular and
spatial data.
USA Contiguous
Albers Equal Area
Conic
Black and White local
spatial dissimilarity
index
US Census Bureau
National Historic GIS
Data Finder
(IPUMS National
Historical GIS n.d.)
Generated from
geoprocessing and
mathematical formula,
then joined to polygon
census tract shapefiles
USA Contiguous
Albers Equal Area
Conic
Black and White local
spatial isolation index
US Census Bureau
National Historic GIS
Data Finder
(IPUMS National
Historical GIS n.d.)
Generated from
geoprocessing and
mathematical formula,
then joined to polygon
census tract shapefiles
USA Contiguous
Albers Equal Area
Conic

51

Race has long been reported in the decennial census, but has gotten increasingly complex
over time. In 1960, there were three race categories “White”, “Negro” and “Other races”. In
1990, these had expanded to five categories: “White”, “Black”, “American Indian”, “Asian or
Pacific Islander” and “Some other race”. By 2020, the basic categories had slightly changed to
“White”, “Black or African American”, “American Indian or Alaska Native”, “Asian”, and
“Native Hawaiian or Other Pacific Islander”, and “Some other race” (IPUMS National Historical
GIS n.d.). These 2020 categories were caveatted in a couple important ways, reflective of a more
plurasitic society. First, cohorts like White and Black populations could have Latino or non-
Latino ethnicity, and these basic categories were for individuals that identified as a singular race
alone – multiracial individuals could now identify as “Two or More Races” (IPUMS National
Historical GIS n.d.). To reflect the 1960 limited categories, this analysis used three basic race
categories, “White”, “Black” and “Other”. In 1990 and 2020 data, this meant aggregating several
races into the third category. It also used the White alone and Black and African American alone
categories from the 2020 census, and was agnostic to Latino heritage.
The first response variable evaluates the population density of Black and White cohorts
within the city’s census tracts. A density metric, as opposed to raw population numbers, was
thought to better normalize the data for better comparisons between tracts. It was created first by
divvying the respective Black and White populations by the area of the census tract. However,
since the square footage units associated with the PCS result in high numbers, population per
acre was used as the density metric.
The maps in Figure 20 show that White density in 1960 is largely following the subway
lines, well into the outer Boroughs. In the 1990s the White population held in Manhattan south of
96th Street but largely vacated the South Bronx and central Brooklyn. The pattern in 2020
52

mimics this, but new areas on the Brooklyn waterfront are densifying. Black population density,
in Figure 21, is much more static, expanding and intensifying with White exodus in the 1990s
but de-densifying somewhat by 2020.

Figure 20. White population density per acre: 1960, 1990, and 2020

Figure 21. Black population density per acre: 1960, 1990, and 2020
The relative population density in New York has an extremely wide spectrum, with the
maximum density exceeding 300 persons per acre and the minimum being fewer than one person
per acre. Because of these extremes, a simple percentage metric was deemed worthwhile as a
second response variable, in order to ensure that patterns occurring in areas with high Black
concentrations, but at a lower density, such as Hollis and St. Albans in southeastern Queens,
were not being overlooked. This dataset was generated similar to the density analysis, but the
53

percentage of Black or White population, respectively, as a percentage of the total population per
census tract (inclusive of White, Black and Other populations) was calculated.
The maps in Figure 22 show how New York went from a largely White city in 1960 to a
more a racially diverse one over time. It also shows how dispersed White population is compared
to the clustered Black segments in Figure 23.

Figure 22. White percentage of population of census tract: 1960, 1990, and 2020

Figure 23. Black percentage of population of census tract: 1960, 1990, and 2020
The first two metrics evaluated the influence of the planning-related variables at static
moments in time – 1960, 1990, and 2020. The decades between these benchmarks were full of
widely shifting demographic change however, so it was deemed worthwhile to try and determine
if the variables exerted influence on this change, positive or negative. For instance, were urban
54

renewal and public housing patterns of the 1950s influencing population shifts themselves? Has
the uptick in contextual rezonings spawned housing pressures, fueling gentrification?
To generate this dataset, the subject population dataset for census tracts in each specific
interval was intersected with the same dataset for the period 30-years prior (including one for
1930 for the 1960 dataset), and the resulting comparisons of polygon size differentials for split
census tracts exported to Excel. In this program, the percentages associated with each portion of
the split census tract could be determined, and an estimated population count for the prior period
derived. This was done by assuming an even population distribution over the census tract
(implausibly, but necessarily because of the limited availability of more granular data) and
attributing to each portion of the split the same percentage of population it occupied in the split
(i.e.., a 60 / 40% split in land area would also receive a 60 / 40% split in population). After
generating density changes in the two census tracts, the tabular data was rejoined in ArcGIS Pro.
The maps in Figures 24 and 25 show clear White population losses between the 1960s
and 1990s corresponding largely with Black gains in Harlem, the South Bronx, and Bedford-
Stuyvesant. There are also tracts with population loss and no corresponding gains in certain
portions of these neighborhoods, a result of urban crises, where property abandonment and
municipal service shrinkage led to swathes of neighborhoods being reduced to rubble by fires
(Flood 2011). By 2020 White gains and Black losses were emerging, particularly in central
Brooklyn and Harlem.
55

Figure 24. White population change per acre: 1930-1960, 1960-1990, and 1990-2020

Figure 25. Black population change per acre: 1930-1960, 1960-1990, and 1990-2020
Two local segregation indices were also included as response variable to test, one
measuring evenness of population distribution through the local spatial dissimilarity index, and
the other measuring relative exposure and isolation through the local spatial isolation index. As
mentioned, there is no R package to devise these local spatial segregation measures. These
indices were generated through the following steps.
Both indices account for the population characteristics of their neighboring tracts,
beginning to address the MAUP, by utilizing a ‘composite’ population count. This composite
count is created through a spatial weight matrix, similar to a local spatial autocorrelation
analysis. By using the Generate Spatial Weights Matrix tool in ArcGIS Pro, a ‘Queen’
56

contiguity-based matrix was generated from each subject year’s census tract shapefile. The
Queen case matrix identifies neighbors of an areal unit with a common edge or vertex, as
opposed to a Rook case which only includes neighbors with a common edge. This Queen-based
matrix, and associated field IDs for every adjacency were exported into a table form. In Excel,
using VLOOKUP tools, the field IDs of a principal census tract and each of its neighbors could
be paired with their respective Black, White and Other population counts in separate tallies, and
the summed to generate the composite count for each tract. Unlike a traditional spatial weight
matrix, no ‘weight’ was assigned to populations in neighboring tracts – if they abutted the
subject tract, the full sum of their populations was aggregated. After generating the segregation
indices, the tabular data was joined to the spatial census tract polygons in ArcGIS Pro.
Local spatial dissimilarity indices were generated for both Black and White populations
pursuant to the formula in Figure 26, where cwi and cbi are the composite population counts of
an areal unit i, CW and CB are the composite population counts of Whites and Blacks,
respectively, over the entire study area, cti is the composite count of the total population of areal
unit i and CT is the composite population count of the total population of the entire study area
(Oka and Wong 2014). In these calculations, total population counts were inclusive of White,
Black and all other races.

Figure 26. Formula for local spatial dissimilarity indices (Oka and Wong 2014)
57

The resulting spatial patterns, which became response variables, are shown in Figures 27
and 28, for White – others, and Black – others, respectively. Because the formula uses the
absolute value in determining segregation values, the localized index value is high in instances
where the subject population group constitutes a very high percentage of the composite count
and is also high when that group’s absence is particularly acute. In the Black – other dissimilarity
index, for instance, Bedford-Stuyvesant in Brooklyn and Hollis and St. Albans in southeastern
Queens have high index values because of the high Black population, but the Upper East and
Upper West Sides of Manhattan, along with southern Staten Island are high because of the high
prevalence of other population groups and the absence of Blacks.

Figure 27. White – others dissimilarity, by census tract: 1960, 1990, and 2020

Figure 28. Black – others dissimilarity, by census tract: 1960, 1990, and 2020
58

The local spatial isolation index was calculated pursuant to the formula in Figure 29 for
both the Black and White population groups, where cgi is the composite population count of a
particular segment, group G in census tract i, G is the population count of group G for the entire
study area (not the composite), and cti is the composite population count of the total population
in census tract i (Oka and Wong 2019). As in the prior analysis, the total population counts are
inclusive of White, Black and all other races.

Figure 29. Formula for local spatial isolation index (Oka and Wong 2019)
Figures 30 and 31 show the maps generated from calculating White and Black isolation,
respectively. White isolation was widespread in 1960, while Black isolation limited, because of
the small, clustered populations. Since 1960, White isolation has entrenched in Manhattan south
of the 96th Street, southern Staten Island, and shifting areas of Brooklyn and Queens.

Figure 30. White isolation, by census tract: 1960, 1990, and 2020
59

Figure 31. Black isolation, by census tract: 1960, 1990, and 2020
3.1.2. Independent Variables
The independent variables being evaluated to test their influence on the response variable
are largely comprised of generations of federal and local land use regulations that would
implicate the built environment, but also include a couple of market-driven factors, to compare to
as a baseline. Like the dependent variables, wherever a source dataset had a different PCS (or
lacked one altogether) its projections was changed to NAD 1983 StatePlane New York Long
Island FIPS 3104 (US Feet). National datasets were clipped to the NYC extents.
New York was a thriving center of commerce long before the emergence of HOLC maps
in the 1930s, and already had a strong spatial pattern. Housing markets responded to the
pronounced demand exerted by immigrants and then internal migrants, with the construction of
high-density housing in the most opportune areas. While reform-minded politicians began
regulating this free market with increasing health and safety parameters after the 1850s, to curb
the most scrupulous developers from warehousing immigrant families (Plunz 2016), traditional
market drivers, like transit access and distance to the regions core, often predicted where housing
units would be located.
60

In New York, Manhattan is the epicenter of the city’s (and regions) economic activity. In
earlier eras of the city, density tracked close to job hubs; tenements in the Lower East Side,
housing immigrants, for instance, was in close proximity to sweatshops in SoHo. Within
Manhattan, the majority of jobs, premier institutions and amenities are concentrated in Midtown
and Lower Manhattan, the city’s two CBDs. Even today, unparalleled access remains a
significant factor underpinning Manhattan’s residential neighborhoods high demand and
associated costs. Without market intervention (in the form or government regulations and other
factors), one might expect the highest natural density to agglomerate in the middle of the two,
and so, a dataset was generated that measured the distance from a center point between the two
CBDs to the centroid of each census tract as a means of evaluating if there is disparity in access
between Black and White population segments.
While the subway system is overseen by the state controlled Metropolitan Transportation
Authority (MTA) today, the initial lines were constructed by private entities, and only later
consolidated into one system. Many lines, or segments thereof, predated the development of
outer borough neighborhoods. Early 20th-century housing development largely followed this
infrastructure, facilitating some relief to overcrowding in immigrant enclaves in Manhattan
(Plunz 2016). This variable is thus seen as a proxy for market-driven growth that typified the
city’s development, apart from government influence vis-a-vis urban renewal, zoning or risky
mortgage lending boundaries. It was used to evaluate the percentage of a census tract that is
within a quarter mile of a mass-transit station, a traditional measure of walkability, and evaluate
whether traditional urban growth measures explained Black or White clustering.
61

After these market-driven variables, government policies were evaluated. These can be
largely split into three different epochs – federal-led HOLC designations, federal and local
mixtures of urban renewal era interventions, and local zoning and historic districting.
HOLC area designations have garnered much criticism for the explicit references to
minority groups in assessing higher risks in the red-colored, D areas. However, all categories
were deemed worthy of evaluating, as A and B areas may be positively correlated with White
populations and used as an initial tool in buttressing their privilege. To generate the dataset, the
HOLC boundaries for NYC were obtained from the University of Richmond’s Mapping
Inequality project. The initial shapefiles were initially borough-specific and so combined into
one citywide shapefile. They were then split into A, B, C, and D areas, and intersected with the
census tracts to determine the percentage of each tract, if any, that was allocated to these
designations. Since the HOLC maps were created in the 1930s, which was before the city had
been built out, a fifth category consisting of the percentage of tracts with areas that were not
designated, was also generated. This could be more highly associated with suburban areas on the
periphery, so may have a strong racial association as well.
Urban renewal era variables include the highway proximity, the public housing density,
and the percentage slated for acquisition and disposal by the city.
The first of these, highway proximity, utilized primary and secondary road classifications
from the US Census Bureau Tiger data within New York State. From this, those segments within
NYC were clipped and projected. The percentage of a census tract that is within proximity
(within 250 feet, a standard block width interval) of a primary or secondary highway was
tabulated by constructing a buffer, clipping the buffer to the extents within the shoreline, and
using that revised area in the calculation. This factor helped determine if Black neighborhoods
62

were disproportionately sited for highway construction. Calculating the number of dwelling units
within that proximity was considered, especially as a counterbalance to low-scale suburban areas
but piecing together the number of units in a building at the historic junctures of 1960 and 1990
was deemed infeasible.

Figure 32. HOLC areas, subway adjacent areas, highway adjacent areas
Public housing began as a means for government-sponsored housing to meet a supply
shortage for working and lower-middle class Whites borne of the Depression and World War II.
Only after the Housing Act of 1949, the emergence of the FHA mortgages, and the
suburbanization of middle-class Whites did public housing become associated with housing
lower-income Blacks in segregated silos. The public housing during and after the 1950s was
often the result of Slum Clearance policies associated with urban renewal, and ironically,
integrated neighborhoods were torn down and replaced by segregated silos (Rothstein 2017;
Blumgart 2017).
HUD has a database with all public housing buildings, including the year built and
number of units. This database helped assess the number of total units in a census tract that had
been constructed at a given point in time. The first public housing was associated with the
Housing Act of 1937, but funding and projects expanded dramatically with the Housing Act of
63

1949. While the Nixon administration placed a moratorium on public housing construction in
1974, and programs largely shifted to vouchers, facilities can be altered, and replacement
housing has been constructed in small amounts since then (National Low Income Housing
Coalition 2019). Looking at Figure 33, the distribution of public housing looks plainly
concentrated in historically minority neighborhoods, like Bedford-Stuyvesant, Bushwick,
Brownsville, Harlem, the South Bronx, and the Lower East Side. The majority of Queens and
Staten Island are without public housing.
To generate this variable, the national dataset was first clipped to the NYC boundary, and
then split into three different data layers based on whether they were in existence on the target
dates of 1960, 1990, and 2020 by using the construction year field. Then, for each dataset, the
number of units within public housing buildings was isolated and summarized within each
census tract. Lastly, the number of units of public housing was then normalized by the acres of
land area in the census tract to generate, in effect, a public housing density.

Figure 33. Public housing density: 1960, 1990, and 2020
Urban Renewal Law allows the City to acquire and dispose property within Urban
Renewal Area (URA) boundaries. New York has dozens of plans, enacted over several decades
by the city’s Department of Housing Preservation and Development (HPD), and its agency
64

predecessors. Sometimes plans laid out the usage of eminent domain, the relocation of current
residents, the demolition of building stock, and the preparation of the site. Other plans emerged
when a neighborhood was already under duress, and a cluster of properties were abandoned by
their owners (or were in significant tax arrears), and the building stock unsafe or hazardous. In
all cases, the plans govern the land uses when the site had undergone land disposition for
redevelopment, and covenants typically run with the land (HPD n.d.). The scale and prevalence
of URA designations in minority neighborhoods, and notoriety in the literature makes their study
worthwhile.
This variable evaluated the percentage of parcels in a given census tract that have
historically been slated for renewal. The dataset was somewhat tedious to generate. Many older
plans include an overall project area boundary, but this is something of a misnomer because
oftentimes only select properties within that boundary were acquired (Figure 34). The University
of Richmond Digital Scholarship Lab, Renewing Inequality project is compiling digital
shapefiles for URAs across the country (ed. Nelson and Ayers n.d.), and was the first data
source. In New York, its data sources were compiled by the advocacy group, 596 Acres, through
a Freedom of Information Law (FOIL) request from HPD (596 Acres n.d.). This generated the
project names and property designations throughout the city. Next, this database was paired with
a DCP database that had cruder digitization (overall project boundaries instead of individual
parcels) but that included the dates for enactment. This was cross-validated with a Urban
Renewal project plan database on HPD’s website, which includes scanned copies of most of the
historic plans (HPD n.d.).
65

Figure 34. URA plan examples in Harlem
With the database assembled, with both the digital boundaries and active dates, they
could be cut by those active in 1960, 1990, and 2020, to glean a sense of their influence in
population clustering in each era.

Figure 35. Urban renewal areas: 1960, 1990, and 2020
Several variables were derived from their district categorizations from the NYC Zoning
Resolution, which is administered by the NYC DCP. Since there are hundreds of different
individual zoning designations in New York, they were batched into pertinent representative
66

categories: single-family districts, two-family districts, multi-family ‘contextual’ districts, multi-
family non-contextual districts, districts that permit residential towers, districts emerging from
former industrial areas and purely non-residential areas.
Zoning district designations in New York come in three major district typologies,
residence, commercial and manufacturing, and on maps are designated by an R, C or M.
Appended to each of these letters is a number which implies the associated density or intensity of
the district. Residence districts, for example, range from R1 at the lowest density and R10 at the
highest; manufacturing districts with an M1 are lower intensity than the M3 reserved for the
most noxious uses. Additional letter or number suffixes after the first letter / number
combination denote additional nuance in regulations, with number typically referring to differing
parking regulations and letters signifying ‘contextual’ rules. A diagram showing the general
spectrum of the districts that permit residences is in Figure 36.

Figure 36. NYC zoning district spectrum (NYC DCP 2018)
In the original 1961 NYC Zoning Resolution, areas with zoning designations of R1
(including R1-1 and R1-2) and R2, permitted only single-family homes, while R3-1 districts
permitted single and two-family homes. Over time, the range of districts, has dramatically
67

increased (single family districts like R1-2A, R2A, and R2X were added in the 1980s through
DCP zoning amendments, as well as more two-family districts, including R3A, R3X, R4-1, R4A,
R4B, and R5A districts). The extent of the area mapped with these districts has also dramatically
increased, particularly the latter. Single-family zoning is a well-documented exclusionary zoning
tactic to buttress White privilege, and has come under tremendous scrutiny in the past several
years. In New York, two-family districts have become much more prevalent than their single-
family counterparts over time. While they certainly have an exclusionary undertone, they were
initially created to permit semi-detached homes, or detached homes on much smaller lots,
typologies often associated with the working or lower-middle class (for example, the television
show ‘All in the Family’ depicted the home of Archie Bunker, a blue-collar patriarch prone to
bigotry as a semi-detached home in Queens; the actual home is zoned as a two-family, R4-1
district today, and is shown in Figure 37. Because the 1961 zoning framework made zoning
district distinctions between where single-family homes on large lots and two-family homes on
smaller lots would be appropriate, the districts were evaluated individually. This district split
would help determine if the income striation being accommodated also had racial undertones.

Figure 37. Semi-detached homes in Queens (Google 2022)
68

In 1961, the majority of the original residence district zoning designations allowed multi-
family housing and employed a tool called a height factor devised by DCP to optimize the
balance of open space and building height, and mimic the “tower-in-a-park” ideal of Stuyvesant
Town, under the guise that entire blocks would similarly be razed and recreated in the same
fashion (NYC DCP 2018).
In the years following the 1961 zoning’s enactment, when programs like urban renewal
became controversial, federal funding dried up, and trends like suburbanization went into full
swing in the post-war years, the potential for wholesale urban redevelopment became
increasingly impractical. Beginning in the 1980s, new zoning districts were created by DCP in
medium and high-density areas to respond to the building incompatibility problems produced by
small infill developments using zoning regulations intended for large-scale urban renewal on
multiple blocks (NYC DCP 2018).
These ‘contextual’ multi-family zoning districts are recognizable on a zoning map by
having an A, B, D or X suffix attached to the R5 through R10 district base district (e.g., R7A
instead of R7). Their emergence also led the initial districts to be dubbed, somewhat
contemptuously, as ‘non-contextual’ districts. Unlike the original suite of districts, contextual
districts have strict, predictable height limits, and have been increasingly mapped – at the urging
of local communities – to limit what is viewed as incompatible development: taller buildings
than their older neighbors (NYC DCP 2018). A comparison of the forms derived from buildings
in non-contextual versus contextual districts is shown in Figure 38 (NYC DCP 2018).
69

Figure 38. Comparison of potential built form in non-contextual versus contextual districts
Because these districts required a zoning map amendment to establish though, the
distribution of the contextual districts may be disproportionately allocated to vocal
neighborhoods with political clout and have ripple effects on housing production and racial
dynamics. Glaeser et al. (2003, 2005) have discussed the impact of onerous height limits on
affordability and exclusivity, particularly in central locations like Manhattan.
What complicates contextual districts, however, is that they became the de riguer type of
zoning district to map when undertaking DCP-led rezonings, with intent to facilitate both growth
and preservation – sometimes simultaneously on different streets in the same rezoning area
(Laskow 2014). The Bedford-Stuyvesant rezoning (shown in Figure 39), for example, designated
large swathes of the neighborhood as R5B and R6B districts, with the intent of preserving the
existing built form. These mappings did not increase the zoned capacity so much as set
prescriptive height limits to prevent development that would be taller than existing buildings.
Along a few principal corridors however, like Fulton St., denser districts like R7D and its
commercial district corollary, C4-5D was mapped, planning for substantial redevelopment (NYC
DCP n.d.).
70

Figure 39. Rezoning map for the 2007 Bedford-Stuyvesant rezoning
This example rezoning also implicates the role of rezonings in demographic
transformation, as sweeping development has occurred in the wake of the rezoning. Whether
directly correlated or not, the two Neighborhood Tabulation Areas (NTAs) that the rezoning area
straddles saw a gain of over 30,000 White residents between 2010 and 2020 and a loss of over
20,000 Black residents (NYC DCP 2021).

Figure 40. Demographic change, 2010-2020 in Bedford-Stuyvesant NTAs
(15,000)
(10,000)
(5,000)
-
5,000
10,000
15,000
20,000
White Black Asian Some other Two or more Hispanic
Bedford-Stuyvesant West Bedford-Stuyvesant East
71

Tumultuous trends like this, particularly those that result in the substantial loss of
minority residents, are certainly suggestive of potential displacement from gentrification, and
have connected rezonings as a pejorative term in many neighborhoods (Stremple 2019).
Since contextual districts have been the vanguard tool for these rezonings – every single
new district mapped in Bedford-Stuyvesant was contextual – testing for correlation with
population change seemed a worthwhile venture. It also seemed prudent to evaluate the areas
that remain non-contextual districts to probe at reasons for not having been contextualized – for
instance, is the building stock in those areas really without any prevailing context, or is it that the
designation has remained, in part, as a byproduct of being a neglected minority community.
The predecessor to the New York’s 1961 Zoning Resolution, the 1916 Resolution, was
very liberal in its allowance for residential towers – nearly any lot in the city could allocate up to
25% of its lot and rise in perpetuity (Vorhees, Walker, Smith & Smith 1958). This was a
theoretical allowance, but in practice it was not heavily utilized because other construction codes
dramatically increased costs after 5-6 stories (e.g., because sprinklers and elevators became
required) and the local market dynamics could rarely offset these costs with higher rents (Plunz
2016). In partial acknowledgement to these realities, the 1961 Zoning Resolution only permitted
tower construction in two residential zoning district types, R9 and R10, and their commercial
district equivalents, and limited those geographies exclusively to Manhattan below 96th Street.
Zoning districts that allow residential towers were seen as worthwhile to evaluate for
correlations with demographic and segregation trends for a few reasons. First, in evaluating the
HOLC maps for the most exclusive areas of the city, only one area that was not suburban
warranted the A designation – the high-density apartment towers that march up the east side of
Central Park. Since the 1960s, new crops of towers have come each generation, with new calls
72

for height caps and greater protections to protect the sunlight reaching the park and the
neighborhood (Oser 1989). Despite much consternation from some civic organizations
(Municipal Art Society 2013), since the mid-aughts, a new crop of supertall towers emerging two
blocks below the park has been the local symbol of ostentatious wealth, with W. 57th Street,
being dubbed ‘Billionaires’ Row’ (Hughes 2018). Between the continued exclusivity of areas
around the park, and the expansion of tower districts to other Borough business districts, like
Hudson Yards, Downtown Brooklyn, and Long Island City through high-profile rezonings, the
ability to construct towers seemed a worthy subject to evaluate exclusivity in and of itself.

Figure 41. HOLC A and B areas around Central Park compared with R10 zoning
In the NYC Zoning Resolution, areas zoned C8 or M1, M2, or M3 are semi-industrial or
manufacturing districts that do not permit new residences to be constructed. Unlike the exclusive
residence districts, these districts have generally not been expanded over time, as the reduction in
areas zoned with these designations has largely tracked with the reduction of the manufacturing
sector’s influence in the local economy.
73

Like many other facets of the zoning resolution, there have been policy shifts in this
regard as well. The planners of the 1950s were overzealous in mapping manufacturing districts,
thinking that in order to retain the manufacturing sector, large swathes of the city would need to
be razed and converted to low-slung housing. To effectuate this, they mapped districts not only
over existing industrial and vacant land, but also over nearby worker housing. This created non-
conformances with the zoning, which made lending and re-investment extremely difficult. As the
manufacturing aspirations did not materialize, the city generated special tools to address these
types of situations, first in the form of ‘D’ suffix districts, and then in Special Mixed-Use
Districts. In Manhattan, districts with an ‘A’ and ‘B’ suffix were created as tools to allow artists
to have live-work spaces in older garment loft buildings in and around SoHo (Haughney 2010).
Districts utilizing these new tools were seen as worthwhile to evaluate separately from the
original manufacturing districts as they may represent a skewed demographic privilege in
bringing zoning conformance and neighborhood investment.
The various zoning datasets discussed above were established by taking the range of
districts and extent existing at the time of the analysis (e.g., 1960 versus 2020), and determining
the percentage of each census tract allocated to these district types.
Digital versions of the NYC zoning map designations are updated monthly and extend
back to June 2009 in an archival database available from the NYC DCP. The Department also
has an archival digital shapefile of the initial 1961 zoning map. However, because there is nearly
a 50-year gap between these available datasets, an interstitial map was sought to better
understand trends more fluidly.
74

Figure 42. Zoning designations: 1960, 1990, and 2020
A 1990 zoning map was created that depicted the city's zoning on December 31, 1990. To
do this, the earliest available digital shapefile from June 2009 was used, and historical print
versions of each individual zoning map were assessed to compare the versions, evaluate which
areas had been rezoned between 1990 and 2009, and to essentially "undo" those rezonings
digitally. Usually, this involved recategorizing several smaller contextual districts that had been
created in the early to mid-2000s into the same non-contextual affiliate, and then merging them
into one polygon. A series of R3X and R3A districts, which permit one- and two-family homes
in detached and semi-detached homes for instance, as shown in the 2009 zoning map on the right
in Figure 43, had previously all been R3-2 districts in the 1975 zoning map on the left, which
permits multi-family homes.
75

Figure 43. Example of zoning changes in South Richmond, Staten Island, 1975 and 2006
The pattern of the 1990 zoning map is interesting; if today’s zoning map is associated
with a granularity from multiple rezonings, and the 1961 map is notable for broad swathes of a
singular district, 1990 is a clear juncture between these eras; the first contextual zonings have
occurred and introduced moments of that granularity as islands in a still large sea of non-
contextual districts. The example in Figure 44 highlights this with initial contextual rezonings in
Astoria, Queens (and the Upper East Side) on the left in 1990, and the same stretch in 2020 on
the right. This shows the importance of having this mid-step in the analysis.

Figure 44. Example of zoning changes in Astoria, Queens, 1990 and 2020
76

In the Zoning Resolution, special regulations contained within a special purpose district
set forth a unique plan for a specific neighborhood or geography through modifications to the
underlying use, height or parking regulations. In 1961, there were no special purpose districts –
the first was established by DCP in Lincoln Square in 1969, and since then dozens more have
been established, typically as part of DCP-led rezonings. These special overlays are denoted in
grey tones on the zoning map and have special chapters in the Zoning Resolution. Their
existence in a certain location may equate to disparate levels of political power and privilege, or
simply with representative case studies for significant policy shifts.
Like the zoning districts themselves, a digital shapefile of special purpose districts
extends back to June 2009, and so a similar process was necessary to recreate the range of
Special purpose districts existing at the end of 1990. These have been created, repealed, and
recombined over time, so assessing their geographies also involved cross-validating their time
sequences in the special district time series tables of Appendix B of the Zoning Resolution.

Figure 45. Special purpose districts: 1960, 1990, and 2020
After the demolition of Penn Station in the mid-1960s, the Landmarks Preservation
Commission was established to review design modifications to landmarked buildings and new
construction in historic districts. Since that time, the number of historic districts has grown
77

vastly. Several authors (Glaeser, Gyourko, and Saks 2005) have written on how the added time
and risk associated with an agency’s discretion in making approvals, not to mention the soft
power of a vocal community, adds significant project costs. This squarely layers into discussions
on artificial housing-supply constraints, often in the areas best suited for growth. This variable
aided in evaluating whether historic districts are disproportionately clustered in White
neighborhoods as a means to protect against new development, and if that in turn has had
implications on Black population dynamics.

Figure 46. Historic districts: 1960, 1990, and 2020
The data sources for these independent variables, their original data type and projection,
are all included in Table 2. Also included is the influence anticipated in the various regression
models.
Table 2. Independent variables to investigate in regression analyses
Criteria Data Source

Type Original
projection

Anticipated
influence
Distance to core Utilized census tracts
centroids and the
center of the Lower
Manhattan and
Midtown Special
Districts as a proxy
for CBD boundaries.
Geoprocessing
generated a
distance field in
the attribute
data.
N/A The influence of this
variable may shift over
time, due to
suburbanization in
White segments. In
recent years it may
positively affect White
78

cohorts, as a proxy for
the wealth gap between
the racial groups.
Percent Subway
Proximate
NYC Open Data,
MTA

Point data for
subway
stations, from
which buffers
were drawn.
GCS of WGS
1984; no
projection
This variable may
behave similar to
distance to core, and
disproportionately
explain Black density in
the eras of peak White
suburbanization.
Percent HOLC
A, B, C, D areas

Percent non-
HOLC areas
University of
Richmond Digital
Scholarship Lab,
Mapping Inequality
project
(Nelson, et al. n.d.)
Polygon data GCS of WGS
1984; no
projection
A and B areas were
favorable ratings, so
they may be positively
correlated with White
neighborhoods and
negatively with Black,
particularly the former.
The influence of C and
D areas may be more
balanced between the
racial segments.
Percent Urban
Renewal Areas
596 Acres, through a
Freedom of
Information Law
(FOIL) request with
the HPD (596 Acres
n.d.)
Polygon data NAD 1983
StatePlane
New York
Long Island
FIPS 3104
(US Feet)
Urban renewal areas
may explain Black
population density loss
in some areas, as
neighborhoods were
razed, and gain in
others, as affordable
housing was developed
on abandoned lots.
Public housing
density per acre
HUD data portal
(HUD 2021)
Point data,
summarized by
counts into
census tract
polygons
GCS of WGS
1984; no
projection
This variable may
disproportionately
explain Black
population density, and
the absence of White
cohorts.
Percent highway
proximate
TIGER / line
highway datasets by
state from US Census
Bureau

(US Census Bureau
n.d.)
Line data, from
which buffers
were drawn
GCS of NAD
1983; no
projection
This variable may
verify the
disproportionate
concentration of
highways in Black
neighborhoods but may
also explain vehicular-
oriented suburban
White neighborhoods.
Percent allocated
to different
zoning
designations
(single-family,
NYC DCP

(NYC DCP 2021)
Polygon data
with district
designations for
1960 and 2020.
For the newly
NAD 1983
StatePlane
New York
Long Island
Different zoning
districts likely impact
White and Black
segments differently.
Single-family districts
79

two-family,
contextual multi-
family, non-
contextual multi-
family,
residential
tower, former
manufacturing,
and non-
residential
districts)

created 1990
map, digitized
archival zoning
maps were
utilized.
FIPS 3104
(US Feet)
and contextual districts
may disproportionately
explain White
population patterns by
virtue of their restrictive
nature and / or the
political capital exerted
to generate zoning
changes. Other districts
may have a more even
impact.
Percent special
purpose district
NYC DCP

(NYC DCP 2021)
Polygon data.
For the newly
created 1990
map, digitized
archival zoning
maps were
utilized.
NAD 1983
StatePlane
New York
Long Island
FIPS 3104
(US Feet)
Special purpose districts
may benefit White
populations more by
virtue of the political
capital involved in
undertaking the
rezoning and
establishment of special
rules.
Percent historic
district
Landmarks
Preservation
Commission
(NYC Landmarks
Preservation
Commission 2021)
Polygons GCS of WGS
1984; no
projection
Like special districts,
this variable may speak
to political privilege and
thus explain White
populations more than
Black

3.2. Research Design
The research in this thesis can be broken into two larger parts: exploratory spatial
analyses and comparative regression analyses. Exploratory spatial analyses consisted of studies
of population patterns and global segregation analyses while regression analyses involved
ordinary least squares, and then additional models to address spatial dependence and
heterogeneity. The results of each informed the findings and conclusions. A summary diagram
for the methods in this thesis is included in Figure 47.
80

Figure 47. Process diagram for thesis methodology
3.2.1. Exploratory Spatial Analyses
Exploratory spatial analyses in this study included population pattern analyses, and
segregation analyses.
3.2.1.1. Population pattern analyses
For this study, data from the US Census Bureau’s decennial census counting the number
of Black and White residents within each census tract of the city was used to determine
significant concentrations – both at the citywide level, and in localized areas. This was done not
only for the 2020 census, but also for historical decennial census data, going back to 1910 (the
earliest date that spatial and tabular data are available for NYC (IPUMS National Historical GIS
n.d.)).
81

Both global and local spatial autocorrelation measures were undertaken. The Spatial
Autocorrelation (Global Moran’s I) tool in ArcGIS Pro outputs a single index value evaluating
the strength of autocorrelation of a phenomenon amongst areal units in a study area. In this case,
the tool was evaluating the density of Black population amongst census tracts to determine if the
pattern of distribution was random or exhibited a spatial pattern of clustering or dispersion such
that the null hypothesis of randomness could be rejected.
Two local measures evaluating spatial autocorrelation were also undertaken. The Getis-
Ord Gi* test (Hot Spot Analysis in ArcGIS Pro) was used in the same decennial time period as
the Moran’s I, and was used to visualize localized areas of statistically significant Black
population concentration or dispersion, through hot spots and cold spots color-coded on a census
tract map. The local Moran’s I evaluation was also conducted, mainly to inspect outlier
relationships in the local relationships between the Black population densities of neighborhing
units – the low-high and high-low relationships, as this was the most novel information not
already captured by other analyses.
In both global and local measures of spatial autocorrelation, the spatial weights matrix
chosen to assess relationships between neighboring polygons matters considerably and has a
large influence over the results. Essentially, the choice of relationship between neighbors allows
the algorithm to compute the influence it will exert.
The relationship between neighboring units generally falls into two categories –
contiguity-based, or distance-based (Chi and Zhu 2019). Contiguity-based matrices evaluates
neighbors based on whether they physically touch each other, and variations in methods involve
the type of abutment; the “Rook’s case” (or ‘Contiguity edges only’ in ArcGIS Pro) for instance,
includes neighbors with a shared boundary, while the “Queen’s case” (or ‘Contiguity edges
82

corners’ in ArcGIS Pro) includes both shared boundaries and vertices (Chi and Zhu 2019). In
NYC, the irregular census boundaries generated by the various river courses and the coastline
limits the viability of these continuity-based approaches; many islands are orphaned without
neighbors, and several river-abutting tracts have limited neighbors, even if the distance over the
river is marginal, and readily traversable via bridges. Even if there were no shoreline conditions
to contend with, the irregularity of certain polygon features, like census tracts, and the vast range
of shapes and sizes, can make continuity-based approaches problematic.
To overcome some of these issues and have a more rational outcome, distance-based
approaches can be utilized. These generally evaluate neighbors based on a distance from a
polygon centroid, and similarly have a few variants in approaches. ‘Fixed distance’ approaches,
as the name suggests, capture all the areal units whose centroid falls within a particular distance.
The abruptness of this approach – which does not consider the influence of neighbors outside the
band, even those immediately outside – leads many to consider the ‘zone of indifference’
approach, which accounts for neighbors’ values beyond the distance band, but just with a
lessening influence (Esri 2009). Distance bands can be chosen based on numerous methods,
such as the distance where the z-score is the highest in the global analysis. The Incremental
Spatial Autocorrelation tool, which calculates the global Moran’s I iteratively over a range of
distances, was used to evaluate the peaks of z-scores in the 1930, 1960, 1990, and 2020 Black
population per acre clusters. As Figure 48 shows, the peak z-scores for each year interval occur
at different distances – 6,000 feet in 1930; 12,000 feet in 1960; 14,000 feet in 1990; and 16,000
feet in 2020. This variance makes this approach problematic.
83

Figure 48. Z-scores resulting from Incremental Spatial Autocorrelation tool
The Manhattan Grid was established in 1811 by the Commissioner’s Plan, and at its core
parcelized the island in the north-south direction into modules consisting of 200-foot-deep blocks
(for 100-foot-deep tax lots) divided by streets of 60-foot width. Roughly every tenth street a
wider, cross-town avenue was mapped, which, at 100 feet, added an additional 40 feet to the
standard width, so that every 20 blocks was an even mile (Ballon 2012). The cross-town avenues
also coincide with major transfer points between express and local subway lines. Given the
regularity of the 1-mile interval , this was chosen as the distance band, and the ‘zone of
indifference' was chosen as the evaluation method.
3.2.1.2. Segregation analyses
Segregation indices have long been utilized to evaluate population patterns in cities and
regions with more nuance. A spatial autocorrelation analysis might show high concentrations of
Black residents in a neighborhood of NYC, for example, but until it is compared against other
racial and ethnic groups and their total respective populations, it is difficult to discern if the
population pattern is segregated from or integrated with other groups.
0
20
40
60
80
100
120
140
160
180
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
11000
12000
13000
14000
15000
16000
17000
18000
19000
20000
21000
22000
23000
24000
25000
26000
27000
28000
29000
30000
1930 1960 1990 2020
84

To address this knowledge gap, global spatial segregation measures like the spatial
dissimilarity (D), relative diversity (R), spatial information theory (H) index, and exposure /
isolation index (P) proposed by Reardon and O’Sullivan (2004), and available through ‘seg’
package of R Studio, were conducted. Evaluating the segregation in census tracts through the
same decennial time series gave a baseline window of whether segregation increased or
decreased through time.
In addition to plotting the raw global values, the ‘seg’ package of R Studio has the ability
to plot resulting population density maps, smoothing the values from census tract centroids with
a kernel density estimator (Hong, O'Sullivan, and Sadahiro 2014). This analysis compares the
population densities for different population cohorts being evaluated, in a side-by-side manner to
visualize their spatial differences (and begin to understand the spatial segregation). These were
output for the years 1930, 1960, 1990, and 2020, to obtain a cursory understanding of the spatial
pattern which could be probed deeeper with local segregation tools in the regression analyses.
3.2.2. Comparative Regression Analyses
The spatial autocorrelation and segregation tests help answer the “where” question of the
analysis – where Black households are inordinately segregated within NYC. The next round of
analyses seek to answer “why” this might be occurring. Regression analyses take the topic issue,
the spatial pattern of various Black and White population metrics, and try to understand their
spatial phenomena through other, independent variables. Like the prior analyses, these are done
at the census-tract level and conducted through different time intervals to have a spatio-temporal
understanding of the persisting or waning influence, if any, of these variables on segregating the
city.
85

Running the linear regression analyses twice, with Black and White population metrics as
the dependent variable, was critical to this analysis. It was thought that severable variables may
exert difference influences, or even opposite influences over their respective population
distributions. The literature notes a propensity for urban renewal to displace Black populations,
and cluster them in public housing. Separately, literature notes the exclusionary nature of some
zoning tools, like single-family housing, and its historical usage in the bulwarking of White
privilege. Understanding the influence of each of these variables simultaneously, with the
opposite population cohort, was compelling. Comparing the regression coefficients between the
two groups identified disparities in the concentrations of their respective population sets.
3.2.2.1. Time intervals: 1960, 1990, and 2020
This evaluation was be done in 30-year intervals beginning in 1960. 1960 was chosen as
it represented the oldest decennial year that would reasonably correlate with the 1961 Zoning
Resolution. While the 1961 Zoning Resolution was adopted by the Board of Estimate on
December 15, 1961, its vestiges trace back earlier. The district types and preliminary mappings
were published in the 1958 Zoning New York City by Vorhees, Walker, Smith and Smith
(Vorhees, Walker, Smith & Smith 1958). The year 1990 was chosen as the intervening year
because it represents the stabilizing period after decades of sustained population loss, and it also
corresponds with a substantial policy shift in the DCP towards mapping contextual zoning
districts, with fixed height caps. This policy shift, and subsequent contextual rezonings began in
earnest after citywide text amendments in 1987 and 1989 for medium / high-density and low-
density districts, respectively (NYC DCP 2018). The choice of a 30-year gap, and three overall
time windows was largely a produce of time availability; a narrower gap, of 20 or even 10 years
86

may have been more ideal but would have necessitated the digital creation of zoning maps for
each intervening window.
3.2.2.2. Ordinary least squares and diagnostics
Ordinary least squares (OLS) regression analysis is a common tool for helping
understand the explanatory variables for a particular phenomenon in an area. One of the principal
outputs are coefficients, positive or negative, for each individual explanatory variable. These
coefficient values represent the expected change in the response variable for an additional unit of
the explanatory variable (Chi and Zhu 019); for example, if evaluating the influence of the
percentage of a census tract allocated to an HOLC D area on the Black population density, a
coefficient of 2.5 would mean that for every additional percentage of HOLC D area, one would
expect 2.5 more Black residents per acre. The model also outputs the p-value for the individual
variable, denoting if it is statistically significant in explaining the response variable. The overall
adjusted R-squared

value of the model helps explain the degree to which a set of variables might
influence the dependent variable and are expressed as a decimal from 0 to 1. An R-squared

value
of .5 for instance, would explain 50% of the model’s behavior (Esri n.d.). Much of the labor of
running linear regression models, however, involves preparing the data, and evaluating the
residuals, to ensure the model is properly fitting the data.
In order to prepare data for the regression model, one first needs to evaluate if there is a
linear relationship between the x and y variables. One method is to plot them on a scatter plot,
and visualize if the variables have a relationship, and if so, whether they fall roughly into a
straight line. One can also employ more technical measures, such as evaluating the skew or
kurtosis in each of the variables to ensure they are normally distributed. Where nonlinear
87

relationships occur, and variables are not normally distributed, one should employ a
transformation to the variables, such as a logarithmic or cube root transformation.
A linear regression should avoid independent variables that are highly correlated, the
phenomenon of multicollinearity. This can often occur with demographic or socioeconomic
factors; if trying to explain why food deserts occur in certain urban areas, for example, race and
income might be highly correlated with one another. A common test for multicollinearity is the
Variance inflation factor (VIF) which indicates a level of multicollinearity, the higher the value
the greater the cause for concern. Values of 1 suggest no influence while values between 1 and 5
suggest increasing levels of moderate influence, and those between 5 and 7.5 begin to be highly
correlated, and could exert influence on the model (Esri n.d.).
These data preparation and checks allowed a first run of the OLS model, but other typical
linear regression assumptions also needed to be met to ensure the model’s viability. One standard
assumption in a regression model is that the residual values are characterized by
homoscedasticity – meaning they have a similar range of variance from each other throughout
the regression plot. If there is heteroscedasticity, the residuals will often become more
pronounced in a cone pattern at one end of the plot. More rigorous tests to evaluate this include
the Breusch-Pagan (BP) test, which evaluates the variance of the residuals and determines their
significance. If the p-value is higher than 0.05, the residuals are random and equally varied, and
thus homoscedasticity is present. However, if the p-value is lower, the null hypothesis can be
rejected, and heteroscedasticity is present with unequal variation (Statology 2020).
Another assumption in a regression model is that residuals will be normally distributed.
This can be checked relatively easily by plotting the residuals through a Q-Q plot and ensuring
that they roughly follow a straight line. If one end of the distribution curves upward, this would
88

indicate outliers, and a log transformation of the independent or dependent variable may be
necessary. More rigorous statistical tests like the Jarque-Bera test evaluate a dataset for skew and
kurtosis, signs that a dataset is not normally distributed. Here, one can run it against the residuals
of the linear regression models to determine if they violate the normality check – the higher the
x-squared value and lower the p-value, the more likely the residuals are not normally distributed
(Statology 2019).
A third assumption is that residuals should be independent of one another, and not
autocorrelated. Spatial autocorrelation can be tested by running a global Moran’s I on the
residuals – if the p-value is statistically significant, there is autocorrelation occurring. A more
sophisticated test is the Lagrange multiplier (LM) test. This test can determine whether there is
spatial lag (SLM) or spatial error (SEM) in the model, and whether it is statistically significant; if
so, then it foreshadows the appropriate type of spatial regression model to run next.
R Studio was used to conduct this linear regression work, and three OLS runs were
employed, to incrementally improve the model’s fit. Model fitness is generally measured by an
Akaike information criteria (AIC) or Bayesian information criteria score; the lower the score, the
better the model fit. Step functions were added that evaluate each variable’s contribution to
model fitness and drop any variable that, when removed, lowers the score were utilized.
Similarly, removing any statistically insignificant variables incrementally bolstered the model’s
fit and gives confidence that the chosen variables are properly explaining the phenomenon being
queried (Esri n.d.).
3.2.2.3. Spatial dependence
Where the OLS model does not pass diagnostic tests for homoscedasticity, normality and
autocorrelation in the residuals, there is likely spatial dependence occurring in the model. If this
89

occurs, more sophisticated models that account for spatial dependence in the data, including the
spatial lag and spatial error models need to be employed.
The SLM helps assess how neighboring response variables might positively or negatively
be influencing each other. It is similar to a linear regression model, but in addition to evaluating
the relationship between explanatory variables, and residuals, it evaluates the spatial lag in the
response variable through a new coefficient, Rho, and by applying a spatial weights matrix to the
model. The SEM is used to account for spatial dependence in the residuals, or the error units,
which occurs when there is a spatial autocorrelation occurring between unidentified explanatory
variables (Chi and Zhu 2019) . Like the SLM, a spatial weight matrix is applied, but this time it
is applied to the error term, and a new coefficient, Lambda, is added to measure the degree this
influence exerts in the model.
This analysis’ regression models all used population data as a response variable, which,
particularly in an urban setting, is prone to spillover effects and spatially dependent variables
(Chi and Zhu 2019). In fact, every OLS model run resulted spatially autocorrelated residuals,
necessitating the spatial regression models to be an added component to each model.
Like the OLS regression, these spatial regression models were constructed and run in R
Studio. In constructing the spatial weights matrix, a Queens-based approach was utilized,
capturing the influence of every abutting census tract. The best model fit was derived by
comparing the resulting AIC values. Residuals were also be tested for autocorrelation, as viable
models will not have statistically significant p-values after running a Moran’s I or LM test.
3.2.2.4. Spatial heterogeneity
While the previous models have tested for spatial dependence within the dataset, other
models can test for spatial heterogeneity. This emerges in datasets when X and Y values vary
90

across a study area, and therefore have localized fluctuations between their influence. For
example, if HOLC redlining boundaries are significant to explaining clusters of segregated Black
communities, the degree to which it explains the clustering may be stronger in some areas of the
city than others – while some neighborhood clusters may, hypothetically, coincide perfectly with
former D zones, others may not, as the D zone may have been designated over dilapidated
building stock near nuisances at the time, for instance, like railroads and waterfronts, that are no
longer perceived as such, and could even be highly desirable areas. The nature of the population
datasets being used lends itself to heterogeneity, as the population density varies dramatically in
the city as it transitions from high-density Manhattan to mid-density inner core neighborhoods in
outer boroughs to suburban areas in the peripheries.
To evaluate spatial heterogeneity in a model and determine how the variables that have
been contemplated and isolated through the linear regression analyses might interact with the
dependent variable differently through space, a geographically weighted regression (GWR) can
be utilized. This analysis did not utilize GWR for every variable in every time period, but instead
isolated specific variables for spatial heterogeneity analysis based on interesting, or potentially
counter-intuitive behavior in other regression models.
This analysis was conducted in R Studio. Since this is a local analysis, there is a spatial
weights matrix to consider here as well. The outputs of this tool were exported to Excel, brought
into ArcGIS and joined with the shapefile. The visualizations of different coefficients can display
the spatial differences in the influences they exert throughout the study area. Collectively they
can show the complex relationships between different variables.

91

Chapter 4 Results
The first component of this section discusses the results of exploratory spatial analyses and the
second details the comparative regression analyses. Collectively they show how the city is
segregated, and the impact of government policies on spatial segregation.
4.1. Exploratory Spatial analyses
Results for the exploratory spatial analyses include population pattern assessments and
global segregation analyses.
4.1.1. Population Pattern Analyses
Conducting Getis-Ord Gi* (Hot Spot) analyses for each decade showcased statistically
significant hot spots and cold spots. Sequencing these analyses in a time series revealed
interesting patterns in densification, de-densification, as well as intra-city migrations and shifts
over-time.
Figure 49 shows the sequenced results of Black hot spots in the city for the decennial
years of 1910, 1920, and 1930. In the

early 20th century, the Black population in the city was
extremely modest. Small communities were clustered throughout much of Manhattan’s west
side, with the largest community in Harlem in northern Manhattan. Between 1910 and 1930 two
interesting phenomena occurred – the Black cluster south of Central Park, in the modern-day
Garment District, disappeared, and a new cluster in Bedford-Stuyvesant in Brooklyn emerged
rapidly.
92

Figure 49. Black population density (left) and hot spot analysis (right), 1910 – 1930
93

Figure 50 shows the next sequence of hot spot analyses for the decades of 1940, 1950,
and 1960. In the middle

of the 20th century, which coincided with the greatest growth in Black
population, the pre-existing Black population clusters continued to grow. Clusters in Harlem
pushed northward into the South Bronx, while hot spots in Bedford-Stuyvesant pushed eastward
towards Brownsville. In 1960, the continued development of the outer boroughs, and the
suburbanization of the periphery, resulted in the initial cold spots in the spatial pattern of Black
density. The Black suburban enclave in southeast Queens was also beginning to emerge.

94

Figure 50. Black population density (left) and hot spot analysis (right), 1940 – 1960
Figure 51 shows the third sequence of decennial clusters – 1970, 1980, and 1990. This
period largely coincided with “White flight” and fiscal difficulties for the city. The cold spots in
these decades intensified, presumably as White residents fled the perceived crime in minority
neighborhoods for suburban enclaves or Manhattan south of 96th Street. As they abandoned edge
neighborhoods, Black residents moved in; the hot spot in Bedford-Stuyvesant expands westward
and begins to wrap Prospect Park while the Harlem cluster wraps into Morningside Heights and
the Upper East Side.

95

Figure 51. Black population density (left) and hot spot analysis (right), 1970 – 1990
Figure 52 shows the final sequence of hot spot maps for the years 2000, 2010, and 2020.
As the city reversed its population loss, pressure has emerged on where and how the city should
grow. Popular, transit-accessible neighborhoods saw tremendous pressure, and expanded
outwards, encroaching into these very same minority neighborhoods. In the 2000s, the Brooklyn
hot spot has pushed eastward, while the Harlem and Bronx hotspots have pushed northwards,
further from the CBDs, into less accessible areas.

96

Figure 52. Black population density (left) and hot spot analysis (right), 2000 – 2020
97

Global Moran’s I (spatial autocorrelation) measures were conducted over time to
determine the changes in the intensity of clustering. The index value indicates the strength of
clustering; the stronger the cluster, the closer the index gets to a value of one. The values are
shown in Figure 53 and are compared with overall population changes in Figure 54. The changes
in Moran’s I seem to track closely with the overall population of NYC. In periods of growth,
clustering of Black residents increased, while in the aftermath of population loss, there was in
fact de-densification. This pattern did not abate in the 21st century.

Figure 53. Moran's I values, 1910 – 2020

Figure 54. Population of NYC 1910 – 2020, by race
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020
0
2,000,000
4,000,000
6,000,000
8,000,000
10,000,000
1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020
White Black Other
98

Local Moran’s I measures were also run. These are particularly useful as they identify
outliers from otherwise High-High and Low-Low relationships (hot and cold spots). Examining
some of the High-Low outliers, shown in Figure 55, many seem to be the result of public
housing on New York City Housing Authority campuses. These outlier public housing campuses
were a byproduct of urban renewal policies and seem to serve as hyper-concentrated areas of
Black residents in otherwise non-diverse areas. These are instructive in validating variables to
consider in the regression analyses.

Figure 55. Local Moran’s I, 2020, with public housing insets
4.1.2. Segregation Analyses
Global spatial segregation indices for each decade, shown in Figure 56, reaffirmed the
clustering trends. While the city’s population was growing, the measures of segregation
increased, suggesting that new-coming Black residents settled amongst other Black residents. In
the 1960s and 70s, the dissimilarity index dropped, potentially reflecting the flight of white
residents, or tacit integration from urban renewal interventions, but whatever the cause, it began
99

to rise again in the 1980s, and has largely plateaued. Observing the exposure / isolation index
further, an unfortunate pattern of Black migration and settlement emerges. In the early 1900s, at
the beginning of the Great Migration, many Black residents had exposure to White individuals
(Black – White), and few were limited to exclusively Black communities (Black – Black). As the
Black population increased over the course of the century, the Black – White and Black – Black
values reversed, and isolated White and Black communities predominated.

Figure 56. Dissimilarity, diversity, information theory, and exposure / isolation index results
Figure 57 compares the outputs of population density surfaces made through the spatial
segregation code in R Studio by applying a kernel density estimator over the population inputs.
The outputs compare the years 1930, 1960, 1990 and 2020. Assessing the sequence of images,
the emergence and stubborn resilience of two separate and distinct cities – one Black, and one
White – is clear. The Black community began in Harlem, split into Bedford-Stuyvesant and the
South Bronx, and eventually spread outward from those areas, presumably as the core areas of
Harlem and Bedford-Stuyvesant suffered gentrification in the 20th century.
0.00
0.20
0.40
0.60
0.80
1.00
Dissimilarity (D) Relative diversity (R)
Information theory (H)
0.00
0.20
0.40
0.60
0.80
1.00
White / White White / Black
Black / White Black / Black
100

101

Figure 57. White and Black population density surfaces, comparing 1930, 1960, 1990, and 2020
4.2. Comparative Regression Analysis
In developing the model, first, all variables were evaluated for their skewness and
kurtosis. In each case, a logarithmic transformation (or in the case of population change, a cube
root transformation since it had negative values) was applied to address the distribution
irregularities that can be seen in Figure 58 for a representative run of variables in the White
population density model for 1960.
102

Figure 58. Histograms showing the skewness in select explanatory variables
Each set of response and explanatory variables was next passed through three OLS runs,
with step functions eliminating variables that did not contribute to lowering the AIC and were
not statistically significant in correlating with the response variable. However, the residuals from
each model had issues; all had p-values that were auto-correlated and statistically significant, Q-
Q Plots showed heteroscedasticity in the tails of the residuals, and residual scatterplots showed a
clustering pattern instead of randomness (as shown in a representative run of White population
density in 1960 in Figure 59), so spatial regression models were utilized in each. Of these, nearly
universally, the SEM generated residuals that were not statistically significant, and also typically
had the lower AIC.
103

Figure 59. Fitted line and Q-Q plot of a standard OLS 3 run
The results from each OLS and spatial dependence regression run are detailed for the first
two response variables: population density and population percentage. The results from the other
three response variables: population density change, dissimilarity, and isolation, are set forth in
Appendices A through C, and categorized by the response variable.
Additionally, the results in the following subsections detail the major findings for every
different population-oriented response variable. Charts compare the resulting coefficients for
explanatory variables for a Black population measure with those for a White population measure.
These charts include coefficients from the last OLS run, (OLS 3 in the charts), which included
only statistically significant variables, as well as those same variables applied to a SLM and
SEM.
Following the linear regression models, the results of the GWR analysis are set forth.
Here, select explanatory variables were compared in tabular and mapped form, for each 30-year
interval, to understand spatial heterogeneity in the variable across the landscape.
104

4.2.1. Population Density
This section details the regression results for the population density response variable. It
first details the results of the three OLS runs and then subsequent SLM and SEM runs to address
autocorrelations amongst the residuals. The section then details the coefficient results and
interprets their relative correlations.
4.2.1.1. Regression runs
Table 3 shows the results of the OLS regression runs for Black and White population
density in 1960. The process of winnowing down variables can be seen in the diminishing
number of variables between the first and third runs. The first two OLS runs progressively
removed all statistically insignificant variables (where a confidence level of less than 90% was
achieved through p-values) or those whose removal would lower the models AIC. Coefficients
with no asterisk next to them in the table have no statistical significance, those with a dot have
0.1 or 90% confidence level, those with a single asterisk have 0.05 or 95%, those with two have
0.01 or 99%, and those with three have 0.001 or a 99.9% confidence level in their significance.
This process of running step functions reduced the initial number of 20 variables down to 7 in
the Black population density model and 12 in the White density model. The R-squared value for
OLS 3 in the Black population density model was 0.383, meaning the variables explained
roughly 38 percent of the pattern of Black population density in the city, while in the White
population density model it was 0.465, explaining over 46 percent of the pattern.
105

Table 3. OLS Regression results, Black and White population density, 1960
Evaluating the model’s residuals versus a fitted line and on a Q-Q plot, shown in Figure
60, there are indications of heteroscedasticity in each, as revealed by the clustering of points in
the former and the tail in the latter.

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept 1.17779 . 0.47387 *** 0.47979 *** 4.80017 *** 4.80869 *** 5.46038 ***
Distance to core -0.03690 -0.52654 *** -0.52664 *** -0.51841 ***
Percent Subway Proximate 0.79379 *** 0.82033 *** 0.77529 *** 0.78548 *** 0.78000 *** 0.77331 ***
Percent HOLC A areas -0.76425 3.20260 *** 3.25406 *** 2.31396 ***
Percent HOLC B areas -0.42781 3.96747 *** 4.00862 *** 3.02286 ***
Percent HOLC C areas -0.44289 3.46773 *** 3.50972 *** 2.50226 ***
Percent HOLC D areas 1.73510 *** 2.18041 *** 2.16858 *** 2.84398 *** 2.87550 *** 1.86087 ***
Percent Non-HOLC areas -0.75672 -0.34664 *** -0.39097 *** 0.96405 * 1.00680 *
Percent Urban Renewal Areas -0.59036 -1.65940 * -1.64911 *
Public housing density per acre 0.67941 *** 0.68264 *** 0.68765 *** 0.27211 *** 0.27503 *** 0.27480 ***
Percent highway proximate -0.55792 ** -0.58258 ** -0.58117 ** 0.22172
Percent single-family zoning -0.03657 2.17662 *** 2.14863 *** 2.16244 ***
Percent two-family zoning -0.95467 * -0.99049 *** -0.97488 *** 2.41162 *** 2.36708 *** 2.36235 ***
Percent 'contextual' multi-family zoning NA N/A
Percent 'non-contextual' multi-family zoning 0.02214 2.99250 *** 2.96168 *** 2.95670 ***
Percent high-density 'tower' districts -0.49192 -0.51003 * 2.78837 *** 2.75922 *** 2.75712 ***
Percent former M-districts NA N/A
Percent pure non-residential districts -0.56896 * -0.55113 *** -0.50697 *** 1.61823 *** 1.61303 *** 1.66473 ***
Percent Special Purpose District NA N/A
Percent historic district NA N/A
adjusted R-squared 0.38290 0.38390 0.38300 0.46730 0.46710 0.46510
Rho
Lambda
AIC 6849.10 6179.70
p-value of residuals (LM) < 2.2e-16 < 2.22e-16
1960 - Black population per acre 1960 - White population per acre
OLS run 1 OLS run 2 OLS run 3 OLS run 1 OLS run 2 OLS run 3
106

Figure 60. Residual versus Fitted Line and Q-Q Plots for Black (top) and White (bottom)
population density
Considering that logarithmic transformations have already been run to all variables,
spatial dependence is likely occurring amongst the residuals. To confirm this, a BP test was run
as a more formal way to evaluate heteroscedasticity in the residuals, while a Jarque-Bera test was
run to evaluate their distribution. In both analyses, and both models, the p-values had a value of
less than 0.001. This statistically significant level of correlation, with 99.9 percent confidence,
confirms that heteroscedasticity, and a non-random residual distribution, are present in the
models. In the Black population density model, VIF values, which evaluate multicollinearity, did
not exceed 1.36, while in the White population density model, one variable, the multi-family
non-contextual zoning districts, had a value of 7.16, and all other were below 4.5. This suggests
some instability in the latter model. These tests suggest spatial dependence models needed to be
run.
Table 4 shows the results of SLM and SEM runs for 1960 models of Black and White
population density. In the SLM, a new coefficient, Rho, accounts for spatial lag while in the
SEM, Lambda accounts for unknown correlation in the error term. In each case here, the new
107

coefficient had a statistically significant influence on the response variable. Overall model fitness
is evaluated by seeking the lowest AIC score and the absence of statistically significant
clustering amongst the residuals, as evaluated by an LM or Moran’s I test for autocorrelation. In
each of these the SEM had the lowest AIC and was the only model with random residuals.

Table 4. SLM and SEM Regression results, Black and White population density, 1960
The OLS runs for Black and White population density in 1990 are in Table 5. Here the
step functions removed fewer variables than in the 1960 model, as new zoning tools and historic
districts began to emerge. However, the adjusted R-squared was slightly lower, at 0.327 and
0.350 for Black and White population density models, respectively, suggesting more variables
were explaining less of the population density pattern – a potential symptom of an increasingly
complex city.
Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept 0.47979 *** 0.12426 *** 1.01307 *** 5.46038 *** 0.88570 . 5.57631 ***
Distance to core -0.51841 *** -0.23873 *** -0.54724 ***
Percent Subway Proximate 0.77529 *** 0.06807 -0.00179 0.77331 *** 0.39136 *** 0.47512 ***
Percent HOLC A areas 2.31396 *** 1.46401 *** 2.10821 ***
Percent HOLC B areas 3.02286 *** 1.94747 *** 2.91154 ***
Percent HOLC C areas 2.50226 *** 1.64816 *** 2.58057 ***
Percent HOLC D areas 2.16858 *** 0.42336 *** 0.66850 *** 1.86087 *** 1.30775 *** 2.07853 ***
Percent Non-HOLC areas -0.39097 *** -0.26507 *** -0.36160 ***
Percent Urban Renewal Areas
Public housing density per acre 0.68765 *** 0.53158 *** 0.58632 *** 0.27480 *** 0.24635 *** 0.24600 ***
Percent highway proximate -0.58117 ** -0.43547 *** -0.38718 **
Percent single-family zoning 2.16244 *** 2.62028 *** 2.39922 ***
Percent two-family zoning -0.97488 *** -0.23343 -0.21715 2.36235 *** 2.82639 *** 2.46293 ***
Percent 'contextual' multi-family zoning
Percent 'non-contextual' multi-family zoning 2.95670 *** 3.28854 *** 3.38947 ***
Percent high-density 'tower' districts 2.75712 *** 3.17543 *** 3.29752 ***
Percent former M-districts
Percent pure non-residential districts -0.50697 *** -0.30994 *** -0.35887 *** 1.66473 *** 2.03508 *** 1.88096 ***
Percent Special Purpose District
Percent historic district
adjusted R-squared 0.38300 0.46510
Rho 0.83648 *** 0.56526 ***
Lambda 0.90947 *** 0.73899 ***
AIC 6849.10 4221.30 4134.30 6179.70 5453.50 5122.30
p-value of residuals (LM) < 2.2e-16 0.10737 0.99900 < 2.22e-16 < 2.22e-16 0.99600
Spatial Lag Spatial Error
1960 - Black population per acre 1960 - White population per acre
OLS run 3 Spatial Lag Spatial Error OLS run 3
108

BP tests confirmed heteroscedasticity in the residuals, while Jarque-Bera tests confirmed
a non-random residual distribution, with statistically significant p-values of less than 0.001 in
each instance. In the Black population density model, the highest VIF values hovered around 2.3
while in the White population density model, the value of the same variable with a high VIF in
the 1960 model, multi-family non-contextual zoning districts, jumped to 8.0 (all others were less
than 3.8). These tests again suggested spatial dependence models needed to be run.

Table 5. OLS Regression results, Black and White population density, 1990
Table 6 shows the SLM and SEM runs made from the remaining variables from OLS 3.
Again, the new Rho and Lambda coefficients are statistically significant. Like the 1960 results,
the SEM has the lowest AIC, and the LM tests confirmed the p-value of the residuals for both
White and Black density are not random (in the SLM only the p-value of the residuals in the
Black population density model are random), collectively meaning it is the best fit for the data.
Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -1.38718 . -1.61366 ** -1.31899 ** 6.25276 *** 6.62814 *** 6.44806 ***
Distance to core 0.21163 *** 0.22602 *** 0.22110 *** -0.66651 *** -0.66822 *** -0.63378 ***
Percent Subway Proximate 1.05075 *** 1.04453 *** 1.04897 *** 0.49035 *** 0.49255 *** 0.42496 ***
Percent HOLC A areas -1.20536 . -1.27821 ** -1.32505 ** 2.09428 *** 1.59453 *** 1.48902 ***
Percent HOLC B areas 0.08566 1.98614 *** 1.46927 *** 1.43336 ***
Percent HOLC C areas -0.33568 -0.42817 ** -0.40885 ** 1.53382 ** 1.01179 *** 1.00015 ***
Percent HOLC D areas 0.60442 0.50863 *** 0.52967 *** 0.53367
Percent Non-HOLC areas -1.19312 * -1.26701 *** -1.23129 *** 0.80810 . 0.19515 .
Percent Urban Renewal Areas 1.34092 ** 1.37254 ** 1.44218 ** -0.85539 * -0.85606 *
Public housing density per acre 0.55144 *** 0.55362 *** 0.55277 *** -0.00106
Percent highway proximate -0.02798 0.45238 * 0.45792 *
Percent single-family zoning 0.88467 * 1.04569 *** 0.72674 ** 2.59823 *** 2.63285 *** 2.49613 ***
Percent two-family zoning -0.88715 * -0.71019 * -1.03812 ** 4.03738 *** 4.06922 *** 3.89170 ***
Percent 'contextual' multi-family zoning -0.20096 6.21624 *** 6.25313 *** 5.98659 ***
Percent 'non-contextual' multi-family zoning 1.53692 *** 1.73668 *** 1.37973 *** 3.96996 *** 4.00082 *** 3.80412 ***
Percent high-density 'tower' districts 1.45798 ** 1.67193 *** 1.27676 ** 2.90272 *** 2.90452 *** 2.81579 ***
Percent former M-districts -0.79510 3.56969 *** 3.59200 *** 3.51725 ***
Percent pure non-residential districts 0.35123 0.51403 . 1.33798 *** 1.37584 *** 1.35670 ***
Percent Special Purpose District -1.12283 *** -1.13820 *** -1.12085 *** 0.58391 *** 0.59258 *** 0.64629 ***
Percent historic district 0.13917 *** 0.13697 *** 0.12552 ** -0.05868 . -0.05802 .
adjusted R-squared 0.33350 0.32850 0.32770 0.37730 0.37750 0.34980
Rho
Lambda
AIC 7410.50 6454.90
p-value of residuals (LM) < 2.2e-16 < 2.2e-16
1990 - Black population per acre 1990 - White population per acre
OLS run 1 OLS run 2 OLS run 3 OLS run 1 OLS run 2 OLS run 3
109

Table 6. SLM and SEM Regression results, Black and White population density, 1990
The results from the first three OLS runs for the Black and White population density in
2020 are shown in Table 7. The step functions removed fewer variables meaning that more
variables had statistically significant influences on the population density for Black and White
cohorts; in the Black density model six variables were removed and in the White population
density model only three were discarded. The adjusted R-squared values were 0.304 for the OLS
3 in the Black density model, and 0.418 in the White, an uptick in the latter from the 1990
model.
After these OLS runs were conducted, BP tests and Jarque-Bera tests again confirmed
heteroscedasticity and non-random distribution in the residuals, respectively, with statistically
significant p-values (less than 0.001 in most instances; the Jarque-Bera test for the White density
model had a value of 0.012). In the Black population density model, the highest VIF value was a
Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -1.31899 ** -0.90439 ** 4.74090 * 6.44806 *** -0.20183 7.12654 ***
Distance to core 0.22110 *** 0.05221 . -0.35724 * -0.63378 *** -0.18637 *** -0.70846 ***
Percent Subway Proximate 1.04897 *** 0.27268 *** 0.42485 *** 0.42496 *** 0.22387 *** 0.39291 ***
Percent HOLC A areas -1.32505 ** -0.49933 * -0.59229 . 1.48902 *** 0.51621 * 0.74484 **
Percent HOLC B areas 1.43336 *** 0.59355 *** 1.22577 ***
Percent HOLC C areas -0.40885 ** -0.08103 0.05800 1.00015 *** 0.31705 *** 0.75884 ***
Percent HOLC D areas 0.52967 *** 0.04293 0.28549 *
Percent Non-HOLC areas -1.23129 *** -0.43503 *** -0.59353 ***
Percent Urban Renewal Areas 1.44218 ** -0.00264 0.05509
Public housing density per acre 0.55277 *** 0.35916 *** 0.39584 ***
Percent highway proximate
Percent single-family zoning 0.72674 ** 0.65369 *** 0.79032 *** 2.49613 *** 3.17315 *** 3.49620 ***
Percent two-family zoning -1.03812 ** 0.43595 * 0.86625 *** 3.89170 *** 3.85856 *** 3.67542 ***
Percent 'contextual' multi-family zoning 5.98659 *** 4.36680 *** 4.54691 ***
Percent 'non-contextual' multi-family zoning 1.37973 *** 1.16010 *** 1.76019 *** 3.80412 *** 4.16138 *** 4.53763 ***
Percent high-density 'tower' districts 1.27676 ** 1.18282 *** 1.32977 *** 2.81579 *** 3.49074 *** 3.61014 ***
Percent former M-districts 3.51725 *** 3.81506 *** 3.81666 ***
Percent pure non-residential districts 1.35670 *** 2.04660 *** 1.61085 ***
Percent Special Purpose District -1.12085 *** -0.15828 . -0.00271 0.64629 *** 0.11363 -0.03103
Percent historic district 0.12552 ** 0.05090 * 0.00561
adjusted R-squared 0.32770 0.34980
Rho 0.82416 *** 0.74617 ***
Lambda 0.90389 *** 0.85398 ***
AIC 7410.50 5030.60 4810.30 6454.90 4908.50 4445.70
p-value of residuals (LM) < 2.2e-16 0.45859 0.99900 < 2.2e-16 1.16E-13 0.99900
Spatial Lag Spatial Error
1990 - Black population per acre 1990 - Black population per acre
OLS run 3 Spatial Lag Spatial Error OLS run 3
110

safe 2.7 while in the White population density model, significant multicollinearity was present
with the HOLC-related variables, with values between 18 and 33, likely causing model
instability. Once again, these tests again suggested spatial dependence models needed to be run.

Table 7. OLS Regression results, Black and White population density, 2020
Like the 1960 and 1990 models, SLM and SEM runs were made for 2020 Black and
White population density, and the results are shown in Table 8. The new Rho and Lambda
coefficients are again statistically significant, and the SEM achieved the best model fit; the AIC
was lowest in the SEM model, and the LM test indicated random p-values for residuals in both
Black and White population density models in the SEM model (the Black SLM model still had
correlated residuals).
Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -2.00104 *** -2.01078 *** -1.79682 *** 5.15011 *** 5.25991 *** 4.77224 ***
Distance to core 0.35984 *** 0.36196 *** 0.30680 *** -0.71430 *** -0.71747 *** -0.68549 ***
Percent Subway Proximate 0.42820 *** 0.43743 *** 0.45700 *** 0.26655 ** 0.26790 *** 0.32000 ***
Percent HOLC A areas -3.24182 *** -3.30148 *** -2.09209 *** 4.68570 *** 4.78976 *** 4.93482 ***
Percent HOLC B areas -1.35695 ** -1.40963 ** 4.50664 *** 4.62070 *** 4.79186 ***
Percent HOLC C areas -1.88627 *** -1.94243 *** -0.71306 *** 3.67408 *** 3.80341 *** 3.96860 ***
Percent HOLC D areas -1.06340 * -1.10617 * 3.53271 *** 3.65718 *** 3.80621 ***
Percent Non-HOLC areas -2.30216 *** -2.36140 *** -1.23921 *** 3.33818 *** 3.46257 *** 3.41995 ***
Percent Urban Renewal Areas 1.23087 ** 1.24985 ** 1.39343 ** 0.00367
Public housing density per acre 0.42259 *** 0.42104 *** 0.43351 *** -0.16716 *** -0.16642 *** -0.15178 ***
Percent highway proximate -0.27110 0.14336
Percent single-family zoning 0.78644 ** 0.79657 ** 1.95157 *** 1.75107 *** 1.83976 ***
Percent two-family zoning 1.12020 *** 1.14285 *** 0.51836 ** 2.32985 *** 2.12739 *** 2.18306 ***
Percent 'contextual' multi-family zoning 1.78817 *** 1.79488 *** 1.15942 *** 3.68977 *** 3.48343 *** 3.51886 ***
Percent 'non-contextual' multi-family zoning 2.36135 *** 2.36700 *** 1.64524 *** 2.69438 *** 2.48231 *** 2.48144 ***
Percent high-density 'tower' districts 2.61332 *** 2.61952 *** 1.94491 *** 3.24205 *** 3.04351 *** 3.19256 ***
Percent former M-districts 1.33880 ** 1.31560 ** 2.04720 *** 1.88973 *** 1.93760 ***
Percent pure non-residential districts 0.80444 ** 0.78846 ** 0.31529
Percent Special Purpose District -0.80331 *** -0.81402 *** -0.80079 *** 0.69053 *** 0.69850 *** 0.66544 ***
Percent historic district 0.52744 * 0.55354 * 0.74456 *** 0.71482 ** 0.74389 ***
adjusted R-squared 0.31130 0.31110 0.30420 0.42930 0.42940 0.41860
Rho
Lambda
AIC 7240.70 6803.90
p-value of residuals (LM) < 2.2e-16 < 2.2e-16
2020 - Black population per acre 2020 - White population per acre
OLS run 1 OLS run 2 OLS run 3 OLS run 1 OLS run 2 OLS run 3
111

Table 8. SLM and SEM Regression results, Black and White population density, 2020
4.2.1.2. Coefficient results
After the models were run, and reasonable fitness achieved, the coefficients could then be
interpreted.
The 1960 population density regression run shows several interesting correlations in the
chart comparing explanatory variable coefficients in Figure 61. While White density is
associated with all HOLC areas, even D areas, Black density is only correlated with HOLC D
areas (though less so than the White population). The Black population is also negatively
correlated with the areas of the city beyond the 1930s era HOLC mappings, which is likely to be
emerging suburban areas that were farmland in the 1930s. Both Black and White populations
correlate with public housing, supporting the notion it originally housed working-class Whites.
White populations correlated positively with almost every available district type (contextual
zoning, special purpose districts, and tools for transitioning manufacturing districts were only
Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -1.79682 *** -1.26607 *** 0.64368 . 4.77224 *** -0.38755 -0.13557
Distance to core 0.30680 *** 0.09753 *** 0.05740 . -0.68549 *** -0.14165 *** 0.01479
Percent Subway Proximate 0.45700 *** 0.05566 0.23955 ** 0.32000 *** 0.05536 . 0.03682
Percent HOLC A areas -2.09209 *** -0.62174 *** -0.54070 * 4.93482 *** 1.72806 *** 1.67337 ***
Percent HOLC B areas 4.79186 *** 1.85647 *** 2.04863 ***
Percent HOLC C areas -0.71306 *** -0.11465 * -0.15419 . 3.96860 *** 1.52602 *** 1.53152 ***
Percent HOLC D areas 3.80621 *** 1.58929 *** 1.50080 ***
Percent Non-HOLC areas -1.23921 *** -0.29289 *** -0.73918 *** 3.41995 *** 1.33657 *** 0.62168 *
Percent Urban Renewal Areas 1.39343 ** 0.32523 . 0.40399 *
Public housing density per acre 0.43351 *** 0.25547 *** 0.26898 *** -0.15178 *** -0.11415 *** -0.15464 ***
Percent highway proximate
Percent single-family zoning 1.83976 *** 1.52401 *** 2.00515 ***
Percent two-family zoning 0.51836 ** 0.85565 *** 1.11911 *** 2.18306 *** 1.85323 *** 2.32256 ***
Percent 'contextual' multi-family zoning 1.15942 *** 0.98739 *** 1.51757 *** 3.51886 *** 2.16495 *** 2.80181 ***
Percent 'non-contextual' multi-family zoning 1.64524 *** 1.13643 *** 1.69888 *** 2.48144 *** 1.86522 *** 2.16841 ***
Percent high-density 'tower' districts 1.94491 *** 1.17865 *** 1.79980 *** 3.19256 *** 2.02632 *** 2.73624 ***
Percent former M-districts 1.93760 *** 2.08134 *** 2.06971 ***
Percent pure non-residential districts
Percent Special Purpose District -0.80079 *** -0.07826 -0.03486 0.66544 *** 0.15026 . -0.00736
Percent historic district 0.74389 *** 0.26060 . 0.3638 *
adjusted R-squared 0.30420 0.41860
Rho 0.82800 *** 0.79385 ***
Lambda 0.89824 *** 0.90607 ***
AIC 7240.70 4640.80 4443.80 6803.90 4875.90 4487.40
p-value of residuals (LM) < 2.2e-16 0.00287 0.99900 < 2.2e-16 0.0922 0.9990
Spatial Lag Spatial Error
2020 - Black population per acre 2020 - White population per acre
OLS run 3 Spatial Lag Spatial Error OLS run 3
112

being developed after 1961), while Black populations negatively correlated with single-family
zoning. Historic districts had also not yet been established in 1961.

Figure 61. Regression coefficients explaining Black versus White population density, 1960
Figure 62 shows the resulting coefficients for population density measures in 1990. The
Black population is correlating negatively with HOLC A areas and continuing its prior HOLC
trends of a positive association with D areas and negative association with the areas beyond the
original HOLC boundaries. White population is no longer correlating with D areas. Urban
renewal areas are exerting an influence now on the Black population, and public housing has
shifted to only influence the Black population segment. New zoning tools, like contextual zoning
districts and tools for transitioning manufacturing districts are strongly correlating with White
population density, while other new tools, like special purpose districts, are negatively
correlating with Black density.
-2.00
-1.00
0.00
1.00
2.00
3.00
4.00
1960 Black pop per acre OLS 3 1960 Black pop per acre SLM 1960 Black pop per acre SEM
1960 White pop per acre OLS 3 1960 White pop per acre SLM 1960 White pop per acre SEM
113

Figure 62. Regression coefficients explaining Black versus White population density, 1990
Figure 63 shows the 2020 regression coefficients for variables explaining Black and
White population density. In 2020, the divergence in Black and White correlation between
HOLC A areas and their respective population densities was strongest. Black populations no
longer correlated with D areas, while White areas did – perhaps signifying some gentrification
emerging. White populations are now negatively correlating with public housing. In terms of
relationships with zoning district typologies, the Black population is now correlating with some
of the newer zoning tools, but not at the same degree as White neighborhoods. Black population
density still is not correlating single-family districts. Lastly, Whites are unique in correlating
their density with historic district locations.
-2.00
-1.00
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
1990 Black pop per acre OLS 3 1990 Black pop per acre SLM 1990 Black pop per acre SEM
1990 White pop per acre OLS 3 1990 White pop per acre SLM 1990 White pop per acre SEM
114

Figure 63. Regression coefficients explaining Black versus White population density, 2020
4.2.2. Population Percentage
This section details the results of the OLS, SLM and SEM regression runs for the
population percentage response variable, and then delves into detailing the coefficient results for
individual explanatory variables.
4.2.2.1. Regression runs
The results of the OLS runs for Black and White population percentage in 1960 are in
Table 9. Like the 1960 model for population density, the step functions removed several
statistically insignificant variables in OLS run 1 and 2, several more in the Black percentage
model than the White. Despite evaluating the influence of the same explanatory variables, the
resulting adjusted R-squared values in these models were much lower than that for population
-3.00
-2.00
-1.00
0.00
1.00
2.00
3.00
4.00
5.00
6.00
2020 Black pop per acre OLS 3 2020 Black pop per acre SLM 2020 Black pop per acre SEM
2020 White pop per acre OLS 3 2020 White pop per acre SLM 2020 White pop per acre SEM
115

density in 1960, with values of only 0.204 and 0.185 for Black and White population percentage
models, respectively.
BP tests confirmed heteroscedasticity in the residuals, while Jarque-Bera tests confirmed
non-random residual distribution. Statistically significant p-values of less than 0.001were present
in each instance. In the Black percentage model, the highest VIF value was safely around 1.7
while in the White population density model, many VIF values associated with HOLC areas
were problematically in the 23 to 38 range, indicating high multi-collinearity; non-contextual
zoning districts were also at around 7. These tests suggested spatial dependence models needed
to be run.

Table 9. OLS Regression results, Black and White population percentage, 1960
The results for the SLM and SEM runs for 1960 Black and White population percentage
are shown in Table 10. The new Rho and Lambda coefficients are statistically significant, and
Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -0.00365 -0.03000 -0.01869 0.21452 * 0.21665 * 0.17856 .
Distance to core 0.02130 ** 0.02304 *** 0.02264 *** -0.02654 *** -0.02674 *** -0.02337 **
Percent Subway Proximate 0.02601 * 0.02447 . -0.02375 . -0.02368 .
Percent HOLC A areas -0.24289 ** -0.25977 *** -0.28995 *** 0.56013 *** 0.55998 *** 0.57140 ***
Percent HOLC B areas -0.24289 *** -0.26858 *** -0.27337 *** 0.58934 *** 0.58938 *** 0.59580 ***
Percent HOLC C areas -0.19666 ** -0.22046 *** -0.22416 *** 0.54967 *** 0.54978 *** 0.55702 ***
Percent HOLC D areas 0.02586 0.34668 *** 0.34701 *** 0.35152 ***
Percent Non-HOLC areas -0.20784 ** -0.23024 *** -0.23890 *** 0.41441 *** 0.41457 *** 0.42886 ***
Percent Urban Renewal Areas -0.03426 0.03704
Public housing density per acre 0.04965 *** 0.05000 *** 0.05055 *** -0.02513 *** -0.02511 *** -0.02441 ***
Percent highway proximate -0.08940 ** -0.08954 ** -0.08649 ** 0.06051 . 0.06060 .
Percent single-family zoning -0.02225 0.43889 *** 0.43892 *** 0.43862 ***
Percent two-family zoning -0.11788 * -0.08470 * 0.50946 *** 0.50937 *** 0.49716 ***
Percent 'contextual' multi-family zoning NA NA
Percent 'non-contextual' multi-family zoning -0.03620 0.45674 *** 0.45663 *** 0.44831 ***
Percent high-density 'tower' districts -0.10084 * -0.06456 . 0.52597 *** 0.52588 *** 0.50977 ***
Percent former M-districts NA NA
Percent pure non-residential districts -0.04624 * 0.43439 *** 0.43399 *** 0.42973 ***
Percent Special Purpose District NA NA
Percent historic district NA NA
adjusted R-squared 0.20550 0.20660 0.20390 0.18680 0.18710 0.18540
Rho
Lambda
AIC -1981.20 -1545.40
p-value of residuals (LM) < 2.2e-16 < 2.2e-16
1960 - Black percentage of population (census tract) 1960 - White percentage of population (census tract)
OLS run 1 OLS run 2 OLS run 3 OLS run 1 OLS run 2 OLS run 3
116

larger than any other coefficient. The SEM was the best model fit as, while the residuals for both
SLM and SEM were random, the AIC was the lowest in the SEM.

Table 10. SLM and SEM Regression results, Black and White population percentage, 1960
Moving to 1990, the OLS runs for Black and White population density are shown in
Table 11. Step functions removed seven insignificant variables from the Black population
density model and only three from the White. The adjusted R-squared remained comparatively
lower than the population density, at 0.193 and 0.250 for Black and White population percentage
models, respectively.
Both the BP and Jarque-Bera tests confirmed heteroscedasticity and non-randomness in
the residuals, respectively, with statistically significant p-values of less than 0.001. The highest
VIF value in the Black percentage model was a safe 1.8 and was a comfortable 2.5 in the White
Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -0.01869 -0.03303 0.18816 0.17856 . -0.23123 ** 0.30983 .
Distance to core 0.02264 *** 0.00538 * -0.00438 -0.02337 ** -0.00645 -0.02206
Percent Subway Proximate
Percent HOLC A areas -0.28995 *** -0.03107 -0.05301 . 0.57140 *** 0.33131 *** 0.36162 ***
Percent HOLC B areas -0.27337 *** -0.04463 *** -0.09165 *** 0.59580 *** 0.35056 *** 0.43743 ***
Percent HOLC C areas -0.22416 *** -0.02907 *** -0.04755 *** 0.55702 *** 0.32032 *** 0.39254 ***
Percent HOLC D areas 0.35152 *** 0.23981 *** 0.26309 ***
Percent Non-HOLC areas -0.23890 *** -0.02796 *** -0.03996 *** 0.42886 *** 0.24085 *** 0.20145 ***
Percent Urban Renewal Areas
Public housing density per acre 0.05055 *** 0.04111 *** 0.04822 *** -0.02441 *** -0.01947 *** -0.02264 ***
Percent highway proximate -0.08649 ** -0.02300 . 0.01877
Percent single-family zoning 0.43862 *** 0.37491 *** 0.38280 ***
Percent two-family zoning 0.49716 *** 0.39477 *** 0.32304 ***
Percent 'contextual' multi-family zoning
Percent 'non-contextual' multi-family zoning 0.44831 *** 0.41312 *** 0.41075 ***
Percent high-density 'tower' districts 0.50977 *** 0.44704 *** 0.53506 ***
Percent former M-districts
Percent pure non-residential districts 0.42973 *** 0.37623 *** 0.36993 ***
Percent Special Purpose District
Percent historic district
adjusted R-squared 0.20390 0.18540
Rho 0.89799 *** 0.68071 ***
Lambda 0.92927 *** 0.76183 ***
AIC -1981.20 -5072.30 -5170.50 -1545.40 -2668.20 -2800.50
p-value of residuals (LM) < 2.2e-16 0.16984 0.99900 < 2.2e-16 0.7156 0.99900
Spatial Lag Spatial Error
1960 - Black percentage of population (census tract) 1960 - White percentage of population (census tract)
OLS run 3 Spatial Lag Spatial Error OLS run 3
117

percentage model. The autocorrelation amongst the residuals suggested that SLM and SEM
models needed to be run to address spatial dependence.

Table 11. OLS Regression results, Black and White population percentage, 1990
Results for the 1990 Black and White population percentage SLM and SEM runs are in
Table 12. Like 1960 models, the new Rho and Lambda coefficients are statistically significant,
and by far the largest coefficients, suggesting large spillovers and other unexplained phenomena.
The SLM model continued to have correlated residuals, as determined through an LM test, while
the SEM were not statistically significant and random. The SEM had the lowest AIC and so was
the best model fit again.

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -0.59461 *** -0.53565 ** -0.63921 *** 0.92199 *** 0.94851 *** 1.11546 ***
Distance to core 0.09041 *** 0.08549 *** 0.09203 *** -0.07812 *** -0.07771 *** -0.08653 ***
Percent Subway Proximate 0.02171 -0.10898 *** -0.11008 *** -0.10595 ***
Percent HOLC A areas -0.28410 * -0.30532 *** -0.29690 *** 0.37558 *** 0.33653 *** 0.36204 ***
Percent HOLC B areas -0.15521 . -0.16978 *** -0.16775 *** 0.25252 ** 0.21253 *** 0.21048 ***
Percent HOLC C areas -0.16538 . -0.17904 *** -0.18168 *** 0.21222 ** 0.17274 *** 0.17499 ***
Percent HOLC D areas 0.01204 0.04124
Percent Non-HOLC areas -0.21987 * -0.23306 *** -0.22935 *** 0.29576 ** 0.25913 *** 0.25578 ***
Percent Urban Renewal Areas 0.21694 ** 0.22573 ** 0.24766 *** -0.21180 ** -0.21411 ** -0.21011 **
Public housing density per acre 0.05509 *** 0.05504 *** 0.05469 *** -0.05579 *** -0.05552 *** -0.05589 ***
Percent highway proximate -0.11289 ** -0.11151 ** 0.02875
Percent single-family zoning -0.03174 -0.11151 ** 0.22025 *** 0.22037 *** 0.13119 ***
Percent two-family zoning -0.36363 *** -0.35490 *** -0.31233 *** 0.58530 *** 0.58360 *** 0.48682 ***
Percent 'contextual' multi-family zoning -0.50467 *** -0.48802 *** -0.44685 *** 0.54930 *** 0.54651 *** 0.47532 ***
Percent 'non-contextual' multi-family zoning -0.05784 -0.04489 0.26706 *** 0.26669 *** 0.15807 ***
Percent high-density 'tower' districts 0.10184 0.11298 . 0.16556 ** 0.22571 ** 0.22295 **
Percent former M-districts -0.27533 * -0.26294 ** 0.43594 *** 0.43428 *** 0.31746 ***
Percent pure non-residential districts 0.00771 0.08724 0.09204 .
Percent Special Purpose District -0.18659 *** -0.18565 *** -0.19384 *** 0.21202 *** 0.21361 *** 0.23362 ***
Percent historic district 0.02684 *** 0.02707 *** 0.02696 *** -0.00165
adjusted R-squared 0.19620 0.19670 0.19250 0.25200 0.25270 0.25030
Rho
Lambda
AIC -455.95 -655.54
p-value of residuals (LM) < 2.2e-16 < 2.2e-16
1990 - Black percentage of population (census tract) 1990 - White percentage of population (census tract)
OLS run 1 OLS run 2 OLS run 3 OLS run 1 OLS run 2 OLS run 3
118

Table 12. SLM and SEM Regression results, Black and White population percentage, 1990
Finally, the OLS runs for Black and White population percentage in 2020 are in Table 13.
Step functions removed a similar number of variables as the 1990 model (six from the Black
percentage and four from the White). The adjusted R-squared was slightly lower in the Black
percentage model, at 0.167 and higher for the White percentage model, at 0.304, when compared
to 1990.
Like other models, the BP tests confirmed heteroscedasticity in the residuals, while
Jarque-Bera tests confirmed the residuals were not random distributed. Each had statistically
significant p-values of less than 0.001. In both Black and White percentage models, unstable VIF
values emerged with some the HOLC area variables, ranging between 18 and 34 in each. Other
variables did not exceed a VIF of 4 in either model. These tests again suggested spatial
dependence models, in the form of SLM and SEM, needed to be run.
Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -0.63921 *** -0.10431 ** 0.16810 1.11546 *** 0.20923 *** 0.69884 *
Distance to core 0.09203 *** 0.01199 ** 0.00581 -0.08653 *** -0.02053 *** -0.03796
Percent Subway Proximate -0.10595 *** -0.01731 * -0.01895
Percent HOLC A areas -0.29690 *** -0.04094 -0.09567 * 0.36204 *** 0.09195 ** 0.12484 *
Percent HOLC B areas -0.16775 *** -0.01883 . -0.06869 *** 0.21048 *** 0.04067 *** 0.10776 ***
Percent HOLC C areas -0.18168 *** -0.00835 -0.01653 0.17499 *** 0.01630 . 0.03268 *
Percent HOLC D areas
Percent Non-HOLC areas -0.22935 *** -0.01519 -0.01157 0.25578 *** 0.03135 ** 0.00483
Percent Urban Renewal Areas 0.24766 *** -0.00668 0.00552 -0.21011 ** 0.02110 0.01453
Public housing density per acre 0.05469 *** 0.03261 *** 0.03908 *** -0.05589 *** -0.02664 *** -0.02945 ***
Percent highway proximate
Percent single-family zoning 0.13119 *** 0.07869 *** 0.14582 ***
Percent two-family zoning -0.31233 *** -0.03572 . -0.00449 0.48682 *** 0.11622 *** 0.12165 ***
Percent 'contextual' multi-family zoning -0.44685 *** -0.07084 ** -0.12389 * 0.47532 *** 0.10329 *** 0.10284 *
Percent 'non-contextual' multi-family zoning 0.15807 *** 0.07413 *** 0.10827 ***
Percent high-density 'tower' districts 0.16556 ** 0.02359 0.02304
Percent former M-districts 0.31746 *** 0.07452 . 0.09580 .
Percent pure non-residential districts
Percent Special Purpose District -0.19384 *** -0.01501 -0.00773 0.23362 *** 0.03221 ** 0.04760 *
Percent historic district 0.02696 *** 0.00138 -0.00316
adjusted R-squared 0.19250 0.25030
Rho 0.90815 *** 0.88803 ***
Lambda 0.93163 *** 0.92480 ***
AIC -455.95 -3870.00 -3923.70 -655.54 -3724.60 -3811.00
p-value of residuals (LM) < 2.2e-16 < 2.2e-16 0.99900 < 2.2e-16 < 2.2e-16 0.99900
Spatial Lag Spatial Error
1990 - Black percentage of population (census tract) 1990 - White percentage of population (census tract)
OLS run 3 Spatial Lag Spatial Error OLS run 3
119

Table 13. OLS Regression results, Black and White population percentage, 2020
Table 14 shows the results for SLM, and SEM runs for 2020 Black and White population
percentages. As with the prior two models, the new Rho and Lambda coefficients are statistically
significant, and the largest coefficients in each model. The SEM was again the best model fit; it
had AIC and, unlike the SLM, the p-values for residuals were not statistically significant.

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -0.41317 *** -0.41742 *** -0.37649 *** 0.60549 *** 0.61597 *** 0.62439 ***
Distance to core 0.07605 *** 0.07646 *** 0.07202 *** -0.08882 *** -0.08933 *** -0.09028 ***
Percent Subway Proximate -0.04414 ** -0.04621 *** -0.03970 ** -0.05890 *** -0.05830 *** -0.06030 ***
Percent HOLC A areas -0.45547 *** -0.44759 *** -0.42229 *** 0.91064 *** 0.91317 *** 0.87727 ***
Percent HOLC B areas -0.31118 *** -0.30500 *** -0.29142 *** 0.72982 *** 0.73304 *** 0.69597 ***
Percent HOLC C areas -0.34756 *** -0.34191 *** -0.33276 *** 0.61230 *** 0.61635 *** 0.57614 ***
Percent HOLC D areas -0.22647 *** -0.22158 *** -0.20607 *** 0.62405 *** 0.62762 *** 0.59068 ***
Percent Non-HOLC areas -0.32455 *** -0.31725 *** -0.30347 *** 0.68068 *** 0.68306 *** 0.65164 ***
Percent Urban Renewal Areas 0.13355 * 0.13367 * -0.02440
Public housing density per acre 0.03432 *** 0.03433 *** 0.03494 *** -0.04061 *** -0.04092 *** -0.04186 ***
Percent highway proximate -0.13459 *** -0.13356 *** -0.14196 *** -0.05355 . -0.05223 .
Percent single-family zoning 0.02605 0.19388 *** 0.18420 *** 0.21651 ***
Percent two-family zoning 0.00312 0.17340 *** 0.16358 *** 0.19816 ***
Percent 'contextual' multi-family zoning -0.00186 0.20899 *** 0.19765 *** 0.23144 ***
Percent 'non-contextual' multi-family zoning 0.08401 * 0.07850 *** 0.07875 *** 0.08601 * 0.07391 * 0.10961 ***
Percent high-density 'tower' districts 0.12863 * 0.12155 ** 0.12076 ** 0.12084 * 0.10601 * 0.13311 **
Percent former M-districts 0.04461 0.05199
Percent pure non-residential districts 0.08867 * 0.08312 ** 0.08280 ** -0.06515 -0.07333 .
Percent Special Purpose District -0.14860 *** -0.14670 *** -0.14876 *** 0.23426 *** 0.23707 *** 0.23785 ***
Percent historic district 0.08351 * 0.08396 * 0.15871 *** 0.16035 *** 0.16935 ***
adjusted R-squared 0.16910 0.17010 0.16720 0.30420 0.30450 0.30430
Rho
Lambda
AIC -1472.40 -1600.40
p-value of residuals (LM) < 2.2e-16 < 2.2e-16
2020 - Black percentage of population (census tract) 2020 - White percentage of population (census tract)
OLS run 1 OLS run 2 OLS run 3 OLS run 1 OLS run 2 OLS run 3
120

Table 14. SLM and SEM Regression results, Black and White population percentage, 2020
4.2.2.2. Coefficient results
After reasonable fitness was achieved, the coefficients for explanatory variables could
then be interpreted.
Switching the response variable to the percentage of Black and White population,
respectively, there were a few notable modifications in each era. In 1960, shown in Figure 64,
the positive and negative correlations in nearly all of the HOLC areas between Black and White
are pronounced. Black populations percentages do not negatively correlate with D areas, but they
do not positively correlate with them either. White population groups continued to positively
correlate with each zoning district type – even areas within, or adjoining manufacturing districts.
There was likely not enough Black population in 1960 to have statistically significant
correlations with zoning districts yet.
Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -0.37649 *** -0.08496 ** 0.01426 0.62439 *** -0.09408 * -0.22017 ***
Distance to core 0.07202 *** 0.00957 *** 0.01187 ** -0.09028 *** -0.00899 ** 0.02091 ***
Percent Subway Proximate -0.03970 ** -0.01299 * -0.01233 -0.06030 *** -0.01068 -0.02149 .
Percent HOLC A areas -0.42229 *** -0.05064 -0.07847 . 0.87727 *** 0.31737 *** 0.34925 ***
Percent HOLC B areas -0.29142 *** -0.02818 -0.06383 . 0.69597 *** 0.25270 *** 0.26164 ***
Percent HOLC C areas -0.33276 *** -0.01699 -0.02043 0.57614 *** 0.21724 *** 0.20501 ***
Percent HOLC D areas -0.20607 *** -0.00895 -0.00045 0.59068 *** 0.24317 *** 0.21954 ***
Percent Non-HOLC areas -0.30347 *** -0.01128 -0.00852 0.65164 *** 0.22560 *** 0.17473 **
Percent Urban Renewal Areas
Public housing density per acre 0.03494 *** 0.01869 *** 0.02174 *** -0.04186 *** -0.02859 *** -0.03406 ***
Percent highway proximate -0.14196 *** -0.00539 0.03691 *
Percent single-family zoning 0.21651 *** 0.12680 *** 0.23449 ***
Percent two-family zoning 0.19816 *** 0.08935 *** 0.13746 ***
Percent 'contextual' multi-family zoning 0.23144 *** 0.09993 *** 0.15123 ***
Percent 'non-contextual' multi-family zoning 0.07875 *** 0.02377 *** 0.04943 *** 0.10961 *** 0.07036 *** 0.08678 ***
Percent high-density 'tower' districts 0.12076 ** 0.03001 . 0.03152 0.13311 ** 0.06429 ** 0.10734 **
Percent former M-districts
Percent pure non-residential districts 0.08280 ** 0.02354 * 0.04063 **
Percent Special Purpose District -0.14876 *** -0.01468 . -0.01934 0.23785 *** 0.05270 *** 0.08636 ***
Percent historic district 0.16935 *** 0.07038 *** 0.0874 ***
adjusted R-squared 0.16720 0.30430
Rho 0.92025 *** 0.86649 ***
Lambda 0.93846 *** 0.90226 ***
AIC -1472.40 -5364.30 -5391.40 -1600.40 -4229.10 -4421.00
p-value of residuals (LM) < 2.2e-16 < 2.22e-16 0.99900 < 2.2e-16 0.0012 0.9990
Spatial Lag Spatial Error
2020 - Black percentage of population (census tract) 2020 - White percentage of population (census tract)
OLS run 3 Spatial Lag Spatial Error OLS run 3
121

Figure 64. Regression coefficients explaining Black versus White population percentage, 1960
In 1990, as shown in Figure 65, the Black population is positively correlating with
distance to the core, while White population percentage are negatively associated with it. HOLC
schisms remain strong, but neither group correlates with D areas. The percentage of an area
subjected to urban renewal is exerting strong influence over the percentage of a census tract that
is Black and diminishing the percentage that is White. The same is true for the density of public
housing, but to a lesser extent. Zoning tools also show a schism – the mapping of two-family
zoning districts, contextual zoning districts and special purpose districts negatively influence
Black populations while they simultaneously positively influence White population percentages.
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
1960 Black percentage OLS 3 1960 Black percentage SLM 1960 Black percentage SEM
1960 White percentage OLS 3 1960 White percentage SLM 1960 White percentage SEM
122

Figure 65. Regression coefficients explaining Black versus White population percentage, 1990
In 2020, shown in the chart in Figure 66, similar trends continue, however, the Black
populations is negatively correlating with all HOLC areas, even D, contextual districts are now
pervasive enough that they are now positively correlating with Black percentages of population,
and the percentage of historic districts in a given area is now influencing the White population
percentage.

-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
1990 Black percentage OLS 3 1990 Black percentage SLM 1990 Black percentage SEM
1990 White percentage OLS 3 1990 White percentage SLM 1990 White percentage SEM
123

Figure 66. Regression coefficients explaining Black versus White population percentage, 2020
4.2.3. Population Density Change
The results of the regression runs for the population density change response variable are
detailed in Appendix A. The explanatory coefficient results are discussed below.
In the chart in Figure 67, denoting population density change between 1930 and 1960,
subway proximity has bifurcated correlations, positively with Black population change and
negatively with White, perhaps suggesting suburbanization with the latter. In terms of HOLC
areas, Black population change is only correlated with D areas, while White population change is
associated with A, B and C, as well as the areas beyond. What is most striking in the chart is the
very strong negative correlation between the percentage urban renewal areas and the change in
Black population density. The percentage of a tract adjacent to a highway also manifests itself
differently for Black and White population change – positively for White and negatively for
Black. This potentially speaks more to suburbanization than deliberate harmful practices in
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
2020 Black percentage OLS 3 2020 Black percentage SLM 2020 Black percentage SEM
2020 White percentage OLS 3 2020 White percentage SLM 2020 White percentage SEM
124

highway siting. In terms of zoning correlations, the Black and White relationship to two-family
districts is split, negatively and positively, respectively. All available zoning district typologies
that permitted residential units, in fact, positively correlated with White population change.
There was no positive correlation with a zoning district and Black population change; besides
two-family zoning, there was also a negative correlation between the Black population change
and manufacturing areas (which could be positive, unless being displaced for the razing of land
for new industrial uses).

Figure 67. Regression coefficients explaining Black versus White density change, 1960
The population change between 1960 and 1990 coincided with White flight and a large
growth in the Black population. In the chart in Figure 68, the prior trend in negative correlations
between White populations and subway access continues. Meanwhile, Black populations have
made positive correlations with HOLC B and C areas, potentially moving into neighborhoods
abandoned by Whites. Both population segments have positive correlation with nearly every
-8.00
-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
1960 Black density change OLS 3 1960 Black density change SLM 1960 Black density change SEM
1960 White density change OLS 3 1960 White density change SLM 1960 White density change SEM
125

zoning tool, although tools like contextual zoning and two-family districts exerts far more
influence on White population change.

Figure 68. Regression coefficients explaining Black versus White density change, 1990
In the time period between 1990 and 2020, as the city’s population stabilized, and began
to outpace its previous historic high, some variables invert their statistically significant
correlation from positive to negative, and vice-versa, from the prior 30-year period. This can be
seen in the chart in Figure 69. White population change is negatively correlating with the
percentage of HOLC A areas, while positively correlating with D areas. The percentage highway
adjacent is now negatively influencing White population change. Urban renewal areas are
positively correlating with White population change, while public housing density is negatively
correlating with Black population change. In terms of zoning, most tools are positively
associating with Black population change, while the percentage of two-family zoning districts
and contextual districts are now correlating negatively with White population change.
-2.00
-1.00
0.00
1.00
2.00
3.00
4.00
5.00
6.00
1990 Black density change OLS 3 1990 Black density change SLM 1990 Black density change SEM
1990 White density change OLS 3 1990 White density change SLM 1990 White density change SEM
126

Cumulatively, these could suggest a re-densification of more urban areas by the White
population and a suburbanization of the Black population. The question on the latter would be
whether this is by housing preference, or displacement as a byproduct of increased housing
demand and slow-growing housing supply. Lastly, special purpose districts were positively
associated with White population change, and historic districts negatively correlated with Black
population change.

Figure 69. Regression coefficients explaining Black versus White density change, 2020
4.2.4. Dissimilarity
The results from each OLS as well as the SLM, and SEM regression runs are detailed in
Appendix B for the dissimilarity response variable. The explanatory coefficient results are
discussed below.
Figure 70 compares the coefficients for explanatory variables for White-other and Black-
other dissimilarity in 1960, 1990, and 2020 from the SEM analyses. The dissimilarity index
-3.00
-2.00
-1.00
0.00
1.00
2.00
3.00
4.00
5.00
2020 Black density change OLS 3 2020 Black density change SLM 2020 Black density change SEM
2020 White density change OLS 3 2020 White density change SLM 2020 White density change SEM
127

results in 1960 nearly mirror each other for Black and White population groups. This is likely
because the absolute value measure, and lack of other population cohorts, meant that at the time
it was effectively a Black-White dissimilarity index. By 1990, a significant number of Whites
had left the city, the Black population had increased, as had the number of other population
cohorts that identified as another race, making some divergence in Black-other and White-other
dissimilarity correlations. Interestingly, White population segments in any HOLC area, and
beyond, had a propensity for dissimilarity. Amongst zoning districts, the percent of a census tract
that was two-family zoning districts reduced White-other dissimilarity while increasing Black-
other dissimilarity. Contextual districts initially prompted large White-other dissimilarity, and
transitional manufacturing district tools lessened it, but both of these correlations have dissipated
over time.

Figure 70. Comparison of SEM dissimilarity coefficients for 1960, 1990, and 2020
-0.10
-0.05
0.00
0.05
0.10
0.15
0.20
Black - other dissimilarity 1960 Black - other dissimilarity 1990 Black - other dissimilarity 2020
White - other dissimilarity 1960 White - other dissimilarity 1990 White - other dissimilarity 2020
128

4.2.5. Isolation
The results from each regression run for the isolation response variable are detailed in
Appendix C. The explanatory coefficient results are discussed below.
The results for the isolation index mimic some of the earlier patterns. Increases in the
percentage of HOLC A, B, and C areas in 1960, shown in Figure 71, correlate negatively with
Black isolation, while correlating positively with White isolation. Highway proximity and single-
family zoning both were associated with White isolation, potentially as trends of
suburbanization. White cohorts also were isolated by non-contextual districts and tower districts,
while Black populations were negatively correlated with tower districts.

Figure 71. Regression coefficients explaining Black versus White isolation, 1960
By 1990, the degree of White isolation associated with each HOLC area had diminished,
perhaps because of the contemporaneous population loss, while the negative correlation for
Black population segments had been largely eliminated. This can be seen in Figure 72. Urban
-0.20
-0.10
0.00
0.10
0.20
0.30
0.40
1960 Black isolation OLS 3 1960 Black isolation SLM 1960 Black isolation SEM
1960 White isolation OLS 3 1960 White isolation SLM 1960 White isolation SEM
129

renewal areas were strong explanatory variables for Black isolation, while highway adjacency
was not. Many zoning districts had split impacts. Two-family zoning districts, contextual
districts and special purpose districts all had strong White isolation correlations, particularly
contextual districts, while all had negative Black isolation correlations. This suggests that
mapping these districts in an area with Black population segments would expose them to other
population groups (exposure being the opposite of isolation on the exposure / isolation index),
while mapping these districts in an area with a higher White population might further isolation.

Figure 72. Regression coefficients explaining Black versus White isolation, 1990
The 2020 results in Figure 73 show a sharp increase in White isolation associated with
HOLC areas again; potentially signaling the city’s renaissance and a tightening housing market.
Within this, the isolation associated with A areas is highest, and is higher than it has ever been.
Black population groups continue to be isolated by urban renewal policies. In terms of zoning
associations, White populations are again being isolated by tower provisions, and continue to be
-0.20
-0.10
0.00
0.10
0.20
0.30
0.40
0.50
1990 Black isolation OLS 3 1990 Black isolation SLM 1990 Black isolation SEM
1990 White isolation OLS 3 1990 White isolation SLM 1990 White isolation SEM
130

isolated by special purpose districts. These White populations have a negative correlation with
manufacturing districts or other transitional formally industrial areas, so would be potentially
exposed to other population segments by their increase. Black segments negatively correlate with
single-, two-family, contextual districts, as well as special purpose districts. Lastly, historic
districts are correlating with White isolation.

Figure 73. Regression coefficients explaining Black versus White isolation, 2020
4.2.6. Spatial Heterogeneity
GWR calculations were generated from the percentage Black and percentage White
response variable to compare a subset of variables for each era. The comparisons were made on
those variables where a statistically significant coefficient was present for both Black and White
populations; however, the original HOLC areas were not included to focus on post-war policies.
The results of these are shown in Table 15, and include the minimum, 1st quartile, median, 3rd
-0.20
-0.10
0.00
0.10
0.20
0.30
0.40
0.50
2020 Black isolation OLS 3 2020 Black isolation SLM 2020 Black isolation SEM
2020 White isolation OLS 3 2020 White isolation SLM 2020 White isolation SEM
131

quartile, and maximum coefficient values for these select variables. The global value is also
included, as a benchmark.
Notably, several variables that have positive operators as a global value also have
minimum local values that are negative, and reciprocally, several variables with negative global
values have maximum local values that are positive. So, while variables like the percentage of
urban renewal are positively associated with the percentage of Black residents and negatively
with White residents globally, there are local instances with strong negative correlations amongst
Black residents (-2481 in 1990) and strong positive amongst Whites (645 in the same year).

Table 15. Range of GWR coefficients for select variables
While several variables have extreme values at the minimums and maximum, the values
within one standard deviation of the median show much lower range. Still, there are several
Variables Minimum 1st Qu. Median 3rd Qu. Maximum Global
Value
1960
Percent Non-HOLC areas, White
-36.40300 -0.01381 0.21113 0.54827 47.41100 0.42830
Percent Non-HOLC areas, Black
-9.49980 -0.23944 -0.05228 0.00115 2.72900 -0.23890
Public housing density per acre, White
-0.80544 -0.04488 -0.02004 0.00099 0.20584 -0.02430
Public housing density per acre, Black
-0.30148 0.02262 0.05055 0.08198 0.79339 0.05050
1990
Percent Urban Renewal Areas, White -5.42872 -0.48206 -0.14935 0.10624 645.17640 -0.21010
Percent Urban Renewal Areas, Black -2481.20000 -0.07957 0.17939 0.62153 40.48500 0.24840
Public housing density per acre, White -7.20200 -0.06665 -0.04425 -0.01748 30.52668 -0.05590
Public housing density per acre, Black -49.19700 0.02946 0.05343 0.08056 13.25800 0.05470
Percent two-family zoning, White -1.52790 0.04310 0.21440 0.41065 1.88170 0.48680
Percent two-family zoning, Black -2.02770 -0.30330 -0.04505 0.06981 2.55680 -0.31220
Percent 'contextual' multi-family zoning, White -33.82921 -0.08695 0.41663 2.18859 1344.62186 0.47530
Percent 'contextual' multi-family zoning, Black -1915.90000 -2.34270 -0.31949 0.04979 51.33300 -0.44120
Percent Special Purpose District, White -0.94534 -0.11930 0.04464 0.28747 76.99546 0.23360
Percent Special Purpose District, Black -117.12000 -0.26473 -0.01084 0.12348 5.77360 -0.19350
2020
Public housing density per acre, White -0.11583 -0.06194 -0.03397 -0.01863 0.00121 -0.04190
Public housing density per acre, Black -0.02456 0.01840 0.03881 0.05832 0.26276 0.03490
Percent 'non-contextual' multi-family zoning, White -0.11333 0.02154 0.05709 0.17166 0.69265 0.10960
Percent 'non-contextual' multi-family zoning, Black -0.09627 0.01160 0.05889 0.11928 0.40506 0.07870
Percent Special Purpose District, White -0.38080 -0.06483 0.06500 0.20371 0.67459 0.23780
Percent Special Purpose District, Black -0.97182 -0.17496 -0.02644 0.06551 0.24890 -0.14880
132

variables with operator shifts between the 1st and 3rd quartile. While the range of coefficients for
particular variables is interesting in tabular form, visualizing them in map form is often more
instructive to intuit the spatial pattern.
When values were exported from R Studio, they were joined with the census shapefile
and visualized with a quantile classification, relative to the specific dataset. The same standard
deviation output as the chart above was not used because many variables had significant outliers,
and a majority of the census tracts would be within a single standard deviation, meaning no real
spatial pattern would be visible. For the same reason, the coefficients were not equalized
between Black and White GWR models. GWR coefficients explaining Black population
percentages are shown in a red-yellow gradient and White population percentage is shown in a
green-yellow gradient. In every model, significant spatial heterogeneity was occurring, and in
many instances the spatial patterns between variable coefficients were somewhat inverted.
Figure 74 shows the salient coefficient differentials for 1960, with percentage of a census
tract that was not within an HOLC area on top and the public housing density on the bottom.
The coefficients for non-HOLC areas associated with a Black percentage of population
were nearly always negative, except in the darkest red areas. These can be seen in core areas of
Manhattan, Western Queens, downtown Brooklyn, and in the select peripheral areas of St.
Albans in southeast Queens, Throgs Neck and Soundview in the Bronx, and various portions of
Staten Island. White coefficients are positive in the darkest green, but this covers the majority of
the city. A few portions of the city are both negatively correlated, like Washington Heights in
upper Manhattan, and Flushing, Queens, and potentially showing the emergence of Dominican
and Asian enclaves, respectively.
133

Public housing density has a strong positive association with a greater Black population
percentage in a census tract (all but the yellow), while the White population is largely negative
(all but the darkest green).

Figure 74. Select GWR coefficients influence on Black and White population percentage, 1960
Figure 75 shows the Black and White comparisons for the GWR coefficients for several
explanatory variables in 1990: urban renewal, public housing density, percentage two-family
zoning district, percentage contextual multi-family housing district, and percentage special
district, from top to bottom respectively.
134

Urban renewal areas are largely positively correlated with Black populations, and
negatively with White populations. Areas where more urban renewal might have increased the
White population percentage and lowered the Black percentage, the darkest green and yellow,
are around Hell’s Kitchen in Manhattan, Greenpoint and Sunset Park in Brooklyn, Long Island
City, and northeast Queens, the northwest Bronx, and the north shore of Staten Island.
Coefficients for public housing per acre remain strongly associated with explaining
greater Black population percentages and negative White population percentages. The strongest
positive coefficients for Black percentages are in areas without a lot of existing public housing
campuses, locations like southern Brooklyn, northeastern Queens, or southern Staten Island.
The two-family coefficients suggest increasing these districts in Jamaica, Queens and
northern Bronx would be conducive to increasing the Black population. In many other locations,
a two-family district increase would disproportionately benefit White population segments.
The coefficients for percentage contextual districts show strong vulnerability for the
Black population in a central Brooklyn belt that includes Bedford-Stuyvesant, Crown Heights,
Brownsville, and Flatlands. Percentage increases here would decrease Black populations and
increase White populations. This finding is troubling as it aligns with population losses in those
geographies and increases in contextual districts.
Lastly, coefficients for the increase in special purpose districts show the strongest
vulnerability for the loss of Black population in a swathe from East New York, in Brooklyn, to
Jamaica, Queens, as well as South Richmond in Staten Island. In these areas, based on the pre-
existing correlations between population segments, new special districts would likely increase
White populations and decrease Black populations. This is also an interesting finding
considering subsequent public policy measures; East New York was the location for the
135

DeBlasio administration’s first mandatory inclusionary housing area and included special
purpose districts. Land speculation stemming from the upzoned residential capacity has led to
claims of gentrification and displacement from the community (Hogan 2021).

136

Figure 75. Select GWR coefficients influence on Black and White population percentage, 1990
137

Figure 76 shows the GWR coefficient comparisons for three variables in 2020 on Black
and White population percentages: public housing density, percentage non-contextual multi-
family housing district, and percentage special district, from top to bottom. The impact of public
housing remains positively correlated with Black populations and negatively with White – even
more than prior eras. Areas that would receive the largest Black population gains from more
public housing density largely remain those areas without public housing campuses. The
relationship between population and the percentage of non-contextual multi-family district is one
of the only variables assessed where the impacts are largely positively correlated with both races.
The negative correlation between these districts and Black populations along the Brooklyn and
Lower East Side waterfronts should give pause, as waterfront parcels often utilize these districts
and developments have spurred tremendous transformation. Finally, the relationship between
increases in the number or size of special purpose districts continue to show a beneficial tilt
towards White communities. Special districts in central Bronx may increase Black populations,
whereas those in central Brooklyn may decrease them and lease to a tremendous uptick in White
populations.

138

Figure 76. Select GWR coefficients influence on Black and White population percentage, 2020

139

Chapter 5 Discussion
This section begins with overall findings in the analysis, along with some policy discussions. The
discussion then moves to data limitations and concludes with other potential research topics
stimulated by these findings.
5.1. Overall Findings
The results culminated in several overall findings, including the persistence of white
privilege, and a spatial path dependency between different eras of policies.
5.1.1. Persistence of White Privilege
Most broadly, there are strong correlations between the spatial pattern of numerous
government policies and disparate impacts on Black and White population segments. This
disparity is often so pronounced that Black and White population groups had inverse correlations
for the same variable. This can be seen in Figures 77 and 78, which visualize the SEM
coefficients for Black and White density and population percentage response variables,
respectively, directly comparing coefficient results for 1960, 1990, and 2020. Whether it be
urban renewal and public housing density positively correlating with Black population density
and percent population and negatively with Whites, or inversely, HOLC A areas, contextual
zoning districts, special purpose districts and historic district designations positively correlating
with White segments while negatively with Black, the impacts are clear. Even in instances where
both positively correlate, like recent trends in two-family zoning districts, or zoning districts that
permit towers, the degree of correlation is always higher in White areas.
Figures 79 through 81 detail these two initial figures, isolating specific variables to
observe the trends more granularly.
140

Figure 77. Comparison of SEM population density coefficients for 1960, 1990, and 2020

Figure 78. Comparison of SEM population percentage coefficients for 1960, 1990, and 2020
-2.00
-1.00
0.00
1.00
2.00
3.00
4.00
5.00
Black pop per acre 1960 Black pop per acre 1990 Black pop per acre 2020
White pop per acre 1960 White pop per acre 1990 White pop per acre 2020
-0.25
0.00
0.25
0.50
0.75
Black percentage 1960 Black percentage 1990 Black percentage 2020
White percentage 1960 White percentage 1990 White percentage 2020
141

Figure 79 shows a select range of variables from the prior charts: HOLC A and D areas,
non-HOLC areas, urban renewal and public housing. For certain variables the disparity between
different coefficients in explaining Black and White density and population percentages not only
persists, but in the worst instances, has even grown. While positively, the impact of HOLC D,
urban renewal and public housing densities has slowly diminished in predicting Black population
densities, there are still statistically significant correlations. Interestingly, HOLC D areas are
stronger predictors of White populations today, a potential indicator of gentrification, or more
complex market dynamics at play. In 1960, White population density was positively correlated
with public housing density, but over time, has become negatively correlated. Tools like HOLC
A areas, which represented the safest areas for investment nearly 100 years ago, still
resoundingly predict White population cohort, and over time, have grown to predict the absence
the absence of Black residents to a statistically significant degree. Areas that were outside of the
original HOLC designations follow a similar trend.

Figure 79. Select SEM population density (left) and percentage (right) coefficients
-1.00
-0.50
0.00
0.50
1.00
1.50
2.00
2.50
Black pop per acre 1960 Black pop per acre 1990
Black pop per acre 2020 White pop per acre 1960
White pop per acre 1990 White pop per acre 2020
-0.20
-0.10
0.00
0.10
0.20
0.30
0.40
Black percentage 1960 Black percentage 1990
Black percentage 2020 White percentage 1960
White percentage 1990 White percentage 2020
142

Figure 80 also shows a subset of the variables from the 1960, 1990 and 2020 SEM
coefficients, detailing the influence of original zoning districts on population density and
population percentage. Single-family zoning districts disproportionately benefited White
populations in 1960 and have continued on that trajectory with Black populations rarely even
registering enough population in those areas to register a statistically significant correlation.
Non-contextual zoning districts and districts that permit towers disproportionately explained
White populations in 1960, but in each, White association has diminished, and Black correlation
increased. In each of these district typologies, it is important to bear in mind that the lack of fixed
height controls in each has made both district typologies somewhat reviled amongst different
communities for their lack of predictability and potential incompatibility with the surrounding
context (Laskow 2014; Adams 2022). In 1960, White population groups were positively
correlated with the percentage of a tract allocated to manufacturing districts. This could have
been worker housing proximate to industrial areas, or land rendered non-conforming and
aspirationally slated for redevelopment. Over time, this correlation has dissipated and
disappeared, potentially as a result of global shifts in manufacturing sector, and a long-term pivot
in the local economy towards other traded sectors as an economic base (Moretti 2013). In 1960,
Black populations were negatively correlated with the percent of an area slated for
manufacturing, but troublingly, over time, have become more positively associated with these
areas, a potential indicator of rising costs, and being pushed towards increasingly fringe locations
of the city. Even if these are no longer manufacturing adjacent (by virtue of larger de-
industrialization trends), they may be adjacent to contaminants from former industrial byproducts
or to noxious emissions from trucks accessing warehousing and wholesaling facilities, which are
often relegated to these areas, raising a host of environmental justice concerns (Kilani 2019).
143

Figure 80. Select SEM population density (left) and percentage (right) coefficients
When new zoning or other planning tools were created, White population groups often
seem to have been the chief beneficiaries. This can be seen in Figure 81, which continues to
isolates select variables from the SEM coefficients of Figures 77 and 78. The year 1990 is
particularly instructive here; new contextual districts, new two-family districts, novel transitional
tools for former industrial areas, and a whole host of new special purpose districts all correlated
with White communities (and, at least in some instances, negatively correlate with Black
communities). To situate this finding in its historical arc, it means that in the era where Black
and Latino communities in the South Bronx and central Brooklyn were at the apex of urban crisis
– with nearly a full generation of disinvestment, White flight, degrading municipal services,
rising crime, deteriorating schools, and a drug epidemic (Flood 2011) – tools that
disproportionately benefited White neighborhoods were seemingly at the forefront of the
-1.00
0.00
1.00
2.00
3.00
4.00
5.00
Black pop per acre 1960 Black pop per acre 1990
Black pop per acre 2020 White pop per acre 1960
White pop per acre 1990 White pop per acre 2020
0
0.1
0.2
0.3
0.4
0.5
0.6
Black percentage 1960 Black percentage 1990
Black percentage 2020 White percentage 1960
White percentage 1990 White percentage 2020
144

planning agenda. While the disparity in some of these new tools has dissipated since 1990,
special purpose districts and historic districts have continued to disproportionately benefit White
communities into the 2020s.

Figure 81. Select SEM population density (left) and percentage (right) coefficients
5.1.2. Path Dependency Between Policies
The spatial pattern for many policies and the manner in which their constituent variable
interplay with Black and White neighborhoods often mimics that of previous policies. There
appears to be a spatial path dependency (David 1985) wherein one generation’s policy
manifestations for an area can entrench the subsequent generations in privilege or hardship.
If an area was designated as an HOLC A area, for instance, as the Fieldston and the
Riverdale-on-Hudson neighborhoods in the Bronx were, there is a stronger likelihood that it
avoided public housing and urban renewal, and similarly, a greater likelihood that it would have
-1.00
0.00
1.00
2.00
3.00
4.00
5.00
Black pop per acre 1960 Black pop per acre 1990
Black pop per acre 2020 White pop per acre 1960
White pop per acre 1990 White pop per acre 2020
-0.20
-0.10
0.00
0.10
0.20
0.30
0.40
Black percentage 1960 Black percentage 1990
Black percentage 2020 White percentage 1960
White percentage 1990 White percentage 2020
145

single-family designations, not to mention other tools like special purpose districts and historic
districts. In this case, all these exclusionary tools layer on each other to buttress an enclave that is
nearly 70% White, in a Borough that has no majority population group, and is roughly 44%
Black and White (US Census Bureau 2021). Figure 82 shows this pattern in Riverdale. At the far
left are the original A and B HOLC designations, next is the percentage of the census tract that is
white, in the middle are single-family zoning designations, followed by special purpose district
and historic districts.

Figure 82. Patterns of layered exclusion in Riverdale, the Bronx
Inversely, HOLC D areas mapped in the 1930s not only predicted where Black
populations would reside in subsequent decades, but also where urban renewal areas and public
housing campuses would come to be placed. The charts in Figure 83 show the degree to which
each of these were disproportionately located within HOLC D areas, and the increasingly fewer
amounts located in HOLC C, B, and A areas.
146

Figure 83. Public housing density and percent urban renewal area by HOLC area
As zoning districts have evolved over time, areas with those former HOLC D
designations continue to have disproportionately lower percentages of the aggregate of single-,
two-family and contextual zoning regulations, while they have higher percentages of non-
contextual zoning districts, higher percentages of manufacturing districts and other specialized
rules for transitional areas. This can be seen in the chart in Figure 84. The aggregated
percentages allocated to single-, two-family, and contextual districts dramatically increase for
every incrementally “better” HOLC area. They also increased dramatically within each subset
between 1960 and 2020.
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
1960
1990
2020
1960
1990
2020
1960
1990
2020
1960
1990
2020
A B C D
Public housing units / acre
0.0%
1.0%
2.0%
3.0%
4.0%
5.0%
1960
1990
2020
1960
1990
2020
1960
1990
2020
1960
1990
2020
A B C D
Percentage urban renewal area
147

Figure 84. Percentage of zoning districts within each historic HOLC area
Similar patterns are apparent for the percentages of historic districts ascribed to each
former HOLC boundary – C and D areas have proportionally fewer areas than A and B, as
shown by Figure 85.

Figure 85. Percentage of historic districts within each historic HOLC area
To test the statistical significance of certain correlations, a final regression model was
made to formally test the correlation between urban renewal, zoning typologies and historic
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
1960 1990 2020 1960 1990 2020 1960 1990 2020 1960 1990 2020
A B C D
Single-family districts Two-family districts Contextual districts
Non-contextual districts Tower districts Former M districts
Pure non-residential
0%
2%
4%
6%
8%
10%
12%
14%
16%
1960 1990 2020 1960 1990 2020 1960 1990 2020 1960 1990 2020
A B C D
Perctentage Historic District
148

districts with the old HOLC areas. The resulting coefficients are in Figure 86, and show the third
OLS run, as well as SLM and SEM runs. It confirms many of these relationships. Urban renewal
areas are positively correlated with D areas and negatively with B and C. Areas with single-
family zoning correlate with A, B and C areas, but not D, while two-family areas, which were
more working-class than their single-family counterparts, correlates more strongly with C areas.
The multi-family districts that permit the majority of the city’s housing correlate with C and D
areas while tower districts strongly correlate with A. Special purpose districts and historic
districts are positively associated with A and B areas, while negatively correlated with C and D.
The results from each OLS and spatial regression run are set forth in Appendix D.

Figure 86. Correlation between post-war policies and HOLC areas
While in some instances the influence of prior generations patterns has dissipated, the
original institutional racism seems to persist in the landscape and simultaneously, seems to have
triggered several successive waves of segregation.
-0.20
-0.10
0.00
0.10
0.20
0.30
0.40
0.50
HOLC A 1960 HOLC A 1990 HOLC A 2020 HOLC B 1960 HOLC B 1990 HOLC B 2020
HOLC C 1960 HOLC C 1990 HOLC C 2020 HOLC D 1960 HOLC D 1990 HOLC D 2020
149

5.2. Policy Recommendations
Based on these results and overall findings, several policy recommendations were
generated. Future work could incorporate these recommendations into a model and evaluate how
they might be expected to work in specific locations, and reduce segregation, given some of the
findings outlined above.
5.2.1. Remove Exclusionary Zoning Tools
Cities and states around the nation have been trying to eliminate exclusionary zoning
tools, principally single-family zoning, in order address historic inequities. In New York, the fact
that nearly every permutation of regression model showed strong correlations between single-
family zoning and White populations, as well as correlations with the historic HOLC A areas,
should make policymakers consider modification. Many legislative precedents have simply
raised the floor on the minimum number of units permitted. Minneapolis, for example, overlaid
their zoning with a mandate that no fewer than four units be permitted in any district. However,
because they only explored unit numbers, and not modifications to other secondary exclusionary
features of those districts, like high parking requirements, larger minimum lot sizes, not to
mention onerous yard and building form rules geared towards single-family homes and not
quadplexes, uptake has been limited (Fox 2022). New York should evaluate its single-family
district regulations comprehensively to permit a greater range of housing typologies in workable
configurations.
While it may be an impulse to associate two-family zoning districts with the same
exclusion as single-family districts, the history here shows they are clearly more complex. They
correlated more strongly with HOLC C areas and have had increasing correlations with Black
populations over time. Changes to zoned capacity should be considered more carefully here, as
150

drastic increases could lead to land speculation, squeezing renters. At the same, modest increases
could give struggling middle-class homeowners an opportunity for rental units that provide a
viable income stream, helping keep minority families in their homes, and safeguarding the ability
to build generational wealth.
Since these single- and two-family areas have fewer amenities, are less transit accessible
and more reliant on vehicle ownership, removing density limits altogether would be ill-advised.
Many cities have explored “missing middle” models to target the creation of 3 – 8 units in a
building (Opticos Design n.d.); these could be appropriate targets to explore in the areas with
best access and services.
5.2.2. Reconsider Contextuality
From this analysis, it is clear that contextual districts have an origin that was closely
correlated with White communities and negatively with Black. While initially under the guise of
preserving the existing scale of neighborhoods, these tools, despite a relatively recent origin of
the 1980s (at least in terms of planning tools), were also preserving the existing segregation in
the landscape. That these districts have now been mapped in neighborhoods throughout the city
does not change the fact that they were dramatically correlated with White neighborhoods, and
preserved their character first and foremost. DCP should question the merits of using existing
context as a planning guidepost to the zoned capacity of an area, instead of data-driven metrics
like access to transit, jobs, amenities, open space, and risk from climate change.
5.2.3. Allow Infill Opportunities on Campuses
One of the most persistent racial imbalances in the city stems from the disproportionate
association with Black communities and public housing campuses. The zoning on these
campuses is often frozen in time, still favoring the tower-on-a-park type of developments they
151

were often created under. Decoupling permitted zoning capacity from the amount of open space
provided on these campuses could free up sizable infill development opportunities. If reasonably
balanced with the preservation of key open space amenities, new infill could provide desperately
needed revenue to offset deferred maintenance on the campus, while also helping to integrate
those hyper-segregated areas.
5.2.4. Equitably Distribute Zoning and Planning Resources
It has been clear that the distribution of special zoning rules, like special purpose districts,
or new districts, disproportionately favors White neighborhoods, as does the mapping of historic
districts. While the impact is different, the commonality in all of this is that agencies seemingly
spend more resources planning for White neighborhoods than Black. City resources are finite,
and for every staff member working on a rezoning action in a White neighborhood, or evaluating
contributing buildings to a proposed historic district, there is an opportunity cost that often
means neglecting the Black neighborhood. Creative strategies that expand the attention paid to
minority communities – without it only being to accommodate more affordable housing for the
city – would be far more equitable.
5.3. Data and Other Research Limitations
While much of the data used in this analysis was rich and authoritative, there were a few
instances where additional or more nuanced data would have been desirable.
5.3.1. Subway and Highway Datasets
Some datasets for this project had limited availability of information that would
authoritatively state how the dataset changed over time, whether expanding, or contracting in its
geography. For some data sources, like urban renewal designations, two datasets could be
152

combined to show the nuance of growing adoption areas over time. For other datasets however,
no authoritative source for year of completion or decommission existed, making the dataset static
in all runs. Specifically, this was true for the subway access and highway proximity datasets.
The subway system began construction in earnest in the early 1900s and largely reached
the system in place today by the 1940s after the city consolidated the prior Interborough Rapid
Transit (IRT) Company and Brooklyn-Manhattan Transit (BMT) Corporation contracts with
their Independent Subway System (IND) lines. Since then, changes have been slow and largely
de minimis. The large moves include the deactivation of the 3rd Ave. El, the opening of the 2nd
Ave. subway and the extension of the Flushing line to Hudson Yards. However, these may not
have had tremendous impacts on the datasets.
The closure of the 3rd Ave. El had begun in the 1950s in Manhattan and by the 70s all
that remained was a vestigial piece in the Bronx. Since this line necessitated a transfer to
continue downtown into Manhattan, it is easy to imagine ridership diminishing as its
connectedness waned.
New subway lines / extensions since 2000 have brought new connectivity to areas with
historic density from prior elevated service, but are only the first phases of decades-long planned
enhancements; the Second Avenue line is planned to extend into Harlem (MTA 2022) while the
MTA has been exploring the extension of the Flushing line to New Jersey under the Hudson
River (McGeehan 2018). While the recent enhancements are tremendous, they do not
revolutionize subway access in their service boundaries; the true impacts will be yielded in future
phases which will bring greater connectivity to areas with limited service.
Beyond these larger moves, several smaller changes have been made, including closing
the Culver El, and extending the E train to Jamaica (nycsubway.org n.d.). While painstakingly
153

comparing the subway maps (which graphically modify the underlying geography to better
visualize the lines) to reconstruct the time applicable geography would have been interesting, it
was not feasible in the time window for this project.
The parkway, expressway and highway system in NYC is one of the oldest in the nation
and is largely the product of the controversial vision of Robert Moses. Many of the initial
planning and construction began as early as the 1930s, using federal funding to stimulate the
economy from the crippling depression. As such, the majority of highways that now crisscross
the city were already constructed by 1960. Like the subway issues, there is no central repository
denoting when highway components were constructed. Complicating this, many highways were
constructed in phases, and several segments progressively upgraded from major roads to
highway class roadways in the same geographies. So, cumulatively, while being infeasible to
partition the dataset into different eras, it may not have had tremendous impacts.
One other aspect of the highway analysis that was limited involved assessing its impacts
on adjacent areas. While this analysis used buffers to capture the land area impacted, it would
have been more meaningful to capture the housing units abutting them. However, while the DCP
has databases that denote the number of units on a zoning lot, it does not extend back past 2002.
Using this to predict 1960 and 1990 housing densities seemed a very flawed assumption.
Beyond the data itself, the buffers being drawn around each in ArcGIS Pro in both the
subway and highway analyses as a means for establishing proximity could also be improved.
Using the Network Analyst tools in ArcGIS Pro, for example, could have incorporated the actual
street network, and devised a more accurate walkshed for subway station proximity. In the
highway analysis, having a more scientifically backed radius connected to the impacts of
airborne particulate matter from vehicular emissions, understanding the passenger vehicle versus
154

freight splits (and associated differentials in emissions) of different highways, and accounting for
the presence of on and off-ramps would all have contributed to a more robust dataset.
5.3.2. Zoning Datasets
Initially this project was seeking to also query the influence of zoning changes in
particular areas and determine if they may correlate with population clustering or segregation.
For example, would a series of rezonings to reduce the residential capacity in one neighborhood
have ripple effects across others by limiting future supply and potentially squeezing current and
would-be residents into other areas? Would the residential racial makeup change as a result?
While much of the zoning district analysis is capturing the effects of those rezonings, as
many tools like contextual zoning districts did not exist in 1961 so any occurrence represents a
rezoning. What it does not indicate, however, is its relation to the previous district and whether it
is increasing or decreasing capacity. Integrating this nuance would have been very compelling
but was unfortunately not in the scope of this project. While DCP has shapefiles representing the
boundaries of rezoned areas, they do not show every individual district change within.
Historically, the Department has done large neighborhood-level rezonings which combine
different moments of increased and decreased capacity, a sort of grand bargain strategy to
effectuate change. Without the individual district changes it is impossible to capture the impact
of those rezonings, however.
While the district designations themselves convey some very interesting narratives, there
are certainly additional layers to the complexity of the zoning that are likely cloaking some
additional nuance and spatial heterogeneity. For example, in 2009 the Department created a
Voluntary Inclusionary Housing (VIH) program that was often, but not always mapped in
conjunction with medium and high-density zoning changes, and in 2016, a Mandatory
155

Inclusionary Housing (MIH) program was established for the next tranche of rezonings. These
boundaries change the FARs, heights and, by virtue of affordability requirements, the residents.
Teasing this intersectionality out, which only really impacted the last decade, seemed daunting
and better suited for a more granular study.
Another limitation is that there are many locations in the city where a textual
modification in the Zoning Resolution modifies the typical district form. For example, along the
waterfront, the Department enacted citywide waterfront public access requirements in 1993 and
generated smaller footprint, tower-like forms for many districts to help enhance sight lines from
upland areas. Many non-contextual districts have subsequently been mapped along the
waterfront but because of the geographically specific modifications, the districts here do not have
much similarity to the same district elsewhere, as they are height capped and have restricted
footprints. Similarly, in the tower districts along the avenues of Manhattan, stricter tower
controls in 'tower-on-a-base' regulations alter the permitted building form to necessitate a squat
building base to match the tenement building context and restrict tower coverage to create a de-
facto height limit. The portion of these districts impacted by these amendments may
disproportionately correlate to White population change or existing White populations,
respectively.
Another example of significantly divergent regulations not being captured by zoning
maps can be seen in some of the most suburban areas of the city – Staten Island and portions of
the northeastern Bronx – which are located within Lower Density Growth Management Areas.
Special textual modifications increase parking requirements, minimum lot sizes, and a series of
other building controls within the Zoning Resolution, to limit development in such a way that
156

housing production and associated neighborhood socioeconomics and demographics have likely
been disproportionately impacted.
5.3.3. Market-based Counterfactuals
While this thesis attempted to include a couple of market-based variables as a
counterpoint to the influence of more overt government policies in codifying or perpetuating
segregation in the city, those attempts had their limitations, and there were other measures that
would have added nuance to the model.
Access is a core aspect to housing delivery in a market-based system, as price premiums
are commonplace for shorter commutes, a greater range of services, and a larger diversity of
amenities. This analysis was simplistic in using a distance from the region’s core as a proxy for
access in several ways. First, distance was measured in a Euclidian fashion; the evaluation could
have been richer if using the ArcGIS Network Analyst tools to generate street grid-derived
travel-sheds and using public transit modes to derive comparative transit times. Next, the
evaluation uses the center of the two CBDs as the apex of access; a more rigorous evaluation
could have combined the travel-shed work with a measure of jobs available. For instance, using
data from the Bureau of Labor Statistics, a job access index could have been generated to
comparatively evaluate the number of jobs available in one hour of commute time from the
centroid of each census tract. Yet, even this would still be simplistic as it doesn’t account for
disparity in access at the origin census tract in the range of neighborhood retail and services. A
census tract that is a half an hour from a half of million jobs in the South Bronx but is in a food
desert should not be evaluated as having the same level of access as a tract in an affluent
neighborhood of Brooklyn with a range of restaurants and specialty grocers.
157

Apart from the shortcomings of the variables that were evaluated, this thesis was limited
by a general difficulty in teasing out the roles of the government versus the free-market in
creating the racial wealth gap, and disentangling one’s influence versus the other in perpetuating
pre-existing divisions. The chief instrument for wealth creation for most middle-class families in
the United States is derived from their home, and its increase in value over the term of their
mortgage (Rothstein 2017). The federal government played a role in blocking access to this
wealth creation for countless Black families by deeming investments in Black neighborhoods as
risky investments, and channeling FHA-backed mortgages – the linchpin to most home
purchases – to suburban areas where local government controls often filtered out the housing
typologies affordable to most minority families (Rothstein 2017). Private sector real estate
brokerage firms doubled down on redlining tactics by discriminating in sales in predominantly
White areas (Coates 2014). Today, the combination of this public and private sector racism has
generated a profound schism in wealth between Black and White families; the average Black
family has just 13 cents in wealth for every dollar held by White families (The White House
2021). This racial wealth divide spawns differences in the ability to afford college tuition, start
businesses, and a host of quality-of-life metrics that only perpetuate the divide generationally
(Perry, Rothwell, and Harshbarger 2018). Shifts in Black populations in the city may certainly be
attributable, in part, to zoning or other policy changes, and the associated influence they exert on
housing supply and costs, but displacement pressures are certainly intensified by the pre-existing
racial wealth gap that began generations ago.

158

5.4. Additional Avenues of Inquiry
This thesis spurred additional thoughts on several ways to either expand this analysis, or
develop interrelated stand-alone research tracks.
5.4.1. Gentrification
Between 2010 and 2020, the Black population in NYC dropped by 4.5 percent while the
city’s overall population grew by 7.7 percent (NYC DCP 2021). Trends like this are incredibly
troubling, and it is paramount to evaluate them, but extremely difficult to analyze through 30-
year time windows, where entire generations of change have occurred.
In order to more fully evaluate trends like gentrification, more granular spatio-temporal
analyses could be conducted. For example, generating a year 2000 zoning map, and coupling this
with the generated 1990 and pre-existing 2010 and 2020 maps would give 10-year increments
instead of 30, and dial in to see the change in influence of certain variables, like two-family
districts, and contextual districts. For this type of analysis, targeting zoning map changes, instead
of relying on contextual districts and other new zoning tools as proxies for them, and being able
to tease out whether these changes constituted upzonings or downzonings, with added or
diminished localized zoning capacity, could layer substantial nuance onto the conversation.
5.4.2. 1916 Zoning
As noted previously, the 1961 Zoning Resolution was a complete overhaul of the 1916
Zoning Resolution. The 1916 Resolution had no singular districts; instead, there were three,
derived from use, height, and area maps (NYC DCP 2018). The genesis of this Resolution was,
in part, discriminatory – the continued encroachment of garment factories, and their immigrant
workers, into the vicinity of department stores along Fifth avenue and was an affront to the
White affluent neighbors who patronized those stores (The Skyscraper Museum 1997-2020).
159

City officials soon called for distinct zones for residences, businesses, and industrial uses, to
address the land use conflict (which may well have been a class conflict, as garment factories
were not belching noxious emissions). The 1916 zoning maps were still quite liberal, with large
swaths of the city zoned 'unrestricted', where any use was permitted, but it did have designated
areas where only residential uses were allowed (Harrison, Ballard & Allen 1950).
Over time, the 1916 Resolution also was amended to address prevailing trends and
planning best-practices. In 1944, a substantial amendment cut heights, increased mandated open
space, and introduced new area district types that only permitted single-family homes (NYC
DCP 2018). Extending the spatio-temporal spectrum to include some of these early designations
would be fascinating. It would be particularly interesting to determine if some of these
restrictions correlate closely with HOLC ratings, and therefore might be a truer original source of
etching segregation into the landscape.
5.4.3. Comparative Measures Between Cities
This analysis has focused exclusively on assessing the persistence of racialized policies in
New York, but the cocktail of federal and local planning policies used to divide Black and White
population groups is hardly unique to New York. Comparing a series of cities would by
fascinating. Cities in differing regions of the US with divergent cultural histories, development
timelines and sizes would be interesting to compare to see if they might uniquely influence the
degree to which these discriminatory policies explain the distribution of Black residents in each
respective city. For example, adding a city like St. Louis, which has the dubious legacy of having
two landmark Supreme Court cases striking down their racist housing practices (Rothstein 2014)
would be compelling, as would Sun Belt cities like Phoenix or Las Vegas which experienced the
160

majority of their growth after HOLC designations, and may therefore have experienced fewer
overlapping layers of segregatory policies.
5.4.4. Other Segregation Studies
This analysis focuses exclusively on race, specifically the impact of government policies
on Black and White segregation. New York, the melting pot that it is, certainly has other races
and ethnicities equally worth studying to evaluate whether other minorities have historically
received disparate treatment, similar to Black populations.
Besides race, socio-economic status divides the spatial landscape of many US cities, and
in high-cost cities like New York, the disparity is heightened. Determining correlations between
the same planning variables and historic median household incomes may be very illuminating.
Historically, a prevalence of low-income households may explain HOLC D area designations
more than race.
5.5. Conclusion
The perniciousness of redlining in New York for the Black community is augmented both
by the cascade of other deleterious policies in a spatially duplicative boundary and by the
reciprocating bulwark of overlapping White privilege. Future planning policies must be mindful
of their historic impact, and should redress their role in segregating communities, and the
disparate opportunities and access that emerge as byproducts of this second-class citizen status.
Ideally this thesis contributes fresh insight into salient issues and elucidates pragmatic solutions
borne of the analysis.

161

References
596 Acres. Urban Reviewer: Plans. n.d. https://www.urbanreviewer.org/#map=12/40.7400/-
74.0080&sidebar=plans.

Abrams, C. Forbidden Neighbors: A Study of Prejudice in Housing. New York: Harper, 1955.

Adams, E. "Eric Adams on how to end America’s housing crisis: YIMBYism, not NIMBYism,
is required." The Economist, November 18, 2022.

Avila, E. "The Master’s Plan: The Rise and Fall of the Modernist City." In Folklore of the
Freeway: Race and Revolt in the Modernist City, by Eric Avila, 17-52. Minneapolis:
University of Minnesota Press, 2014.

Ayers, E. L., J. Madron, and N. Ayers. "Southern Journey: The Migrations of the American
South, 1790-2020." University of Richmond, Digital Scholorship Lab. 2021.
https://dsl.richmond.edu/southernjourney/.

Ballon, H. The Greatest Grid: The Master Plan of Manhattan, 1811-2011. New York: Columbia
University Press, 2012.

Blumgart, J. "Housing Is Shamefully Segregated. Who Segregated It?" Slate. June 2, 2017.
https://slate.com/business/2017/06/an-interview-with-richard-rothstein-on-the-color-of-
law.html.

Boeing, G. "We Live in a Motorized Civilization: Robert Moses Replies to Robert Caro."
Research Gate. March 2021.
https://www.researchgate.net/publication/317570292_We_Live_in_a_Motorized_Civiliza
tion_Robert_Moses_Replies_to_Robert_Caro.

Bui, Q., and E. Badger. "Cities Start to Question an American Ideal: A House with a Yard on
Every Lot." The New York Times. June 18, 2019.
https://www.nytimes.com/interactive/2019/06/18/upshot/cities-across-america-question-
single-family-zoning.html.

Bureau of Planning and Sustainability. "Historical Context of Racist Planning." City of Portland,
Oregon. September 2019. https://www.portland.gov/sites/default/files/2019-
12/portlandracistplanninghistoryreport.pdf.

Caro, R. The Power Broker. New York: Alfred A. Knopf, 1973.

Chatelain, P. M. The NYC That Never Was: Robert Moses' Lower Manhattan Expressway
(LOMEX). n.d. https://untappedcities.com/2013/09/11/nyc-that-never-was-robert-moses-
lower-manhattan-expressway-lomex/.

Chi, G., and J. Zhu. Spatial Regression Models for the Social Sciences. Sage Publications, 2019.
162

Chronopoulos, T. "“What’s Happened to the People?” Gentrification and Racial Segregation in
Brooklyn." Journal of African American Studies, No. 24, 2020: 549–572.

Clark, W. A.V. "Residential segregation in American cities: a review and interpretation."
Population Research and Policy Review, No. 5, 1986: 95-127.

Closson, T. "‘Nobody Cares About Us Here’: Anguish and Anger on Buffalo’s East Side." The
New York Times. May 15, 2022. https://www.nytimes.com/2022/05/15/nyregion/east-
side-buffalo-neighborhood.html.

Coates, T. "The Case for Reparations." The Atlantic. June 2014.
https://www.theatlantic.com/magazine/archive/2014/06/the-case-for-reparations/361631/.

Cohen, J. "New exhibit examines the historical impact of the 7 train on Queens." Jackson
Heights post. July 25, 2017. https://jacksonheightspost.com/new-exhibit-examines-the-
historical-impact-of-the-7-train-on-queens.

Committee on Banking and Currency, US Senate. Summary of Provisions of the National
Housing Act of 1949. Senate testimony, Washington, DC: US Government Printing
Office, 1949.

David, P. "Clio and the Economics of QWERTY." The American Economic Review Vol. 75, No.
2, Papers and Proceedings of the Ninety-Seventh Annual Meeting of the American
Economic Association, 1985: 332-337.

DC Office of Planning. "Single-Family Zoning in the District of Columbia." DC Office of
Planning. April 2020.
https://plandc.dc.gov/sites/default/files/dc/sites/Comprehensiveplan/007_Single%20Fami
ly%20Housing%20Report.pdf.

Department of Housing and Urban Development. Public Housing Buildings. January 8, 2021.
https://hudgis-hud.opendata.arcgis.com/datasets/52a6a3a2ef1e4489837f97dcedaf8e27_0.

Department of Housing Preservation and Development. Urban Renewal. n.d.
https://www.nyc.gov/site/hpd/services-and-information/urban-renewal.page.

Detroit News. "Detroit News Photograph Collection (UAV002691)." Walter P. Reuther Library,
Archives of Labor and Urban Affairs, Wayne State University. July 27, 1970.
https://digital.library.wayne.edu/item/wayne:vmc25493_4.

Duncan, O. D., and B. Duncan. "A Methodological Analysis of Segregation Indexes." American
Sociological Review Vol. 20, No. 2, 1955: 210-217 .

163

ed. Nelson, R. K., and E. L. Ayers. "Renewing Inequality: Urban Renewal, Family
Displacements, and Race 1955-1966." University of Richmond, Digital Scholarship Lab.
n.d. https://dsl.richmond.edu/panorama/renewal/#view=0/0/1&viz=cartogram.

Ejdemyr, S. Segregation Measures in R. n.d. https://sejdemyr.github.io/r-
tutorials/statistics/measuring-segregation.html.

Elliott, M., and J. Hughes. "Four hundred years after enslaved Africans were first brought to
Virginia, most Americans still don’t know the full story of slavery." The New York Times
Magazine, the 1619 Project. August 19, 2019.
https://www.nytimes.com/interactive/2019/08/19/magazine/history-slavery-
smithsonian.html.

Esri. "Exploratory Regression (Spatial Statistics)." ArcGIS Pro. n.d.
https://pro.arcgis.com/en/pro-app/2.9/tool-reference/spatial-statistics/exploratory-
regression.htm.

—. Hot Spot Analysis Part 1: Conceptualization of Spatial Relationships. December 3, 2009.
https://www.youtube.com/watch?v=QJtPmnFgZbs.

—. How Spatial Autocorrelation (Global Moran's I) works. n.d.
https://desktop.arcgis.com/en/arcmap/10.3/tools/spatial-statistics-toolbox/h-how-spatial-
autocorrelation-moran-s-i-spatial-st.htm.

—. Introduction to Regression Analysis Using ArcGIS Pro. n.d.
https://www.esri.com/training/catalog/57630430851d31e02a43ee0c/introduction-to-
regression-analysis-using-arcgis-pro/ (accessed October 25, 2020).

—. Mapping Clusters: Hot Spot and Cluster and Outlier Analysis. n.d.
https://www.esri.com/training/catalog/5d39b7916b9c17664bb65a3b/mapping-
clusters%3A-hot-spot-and-cluster-and-outlier-analysis/ (accessed October 24, 2020).

—. Modeling Spatial Relationships. n.d. https://pro.arcgis.com/en/pro-app/tool-reference/spatial-
statistics/modeling-spatial-relationships.htm#GUID-729B3B01-6911-41E9-AA99-
8A4CF74EEE27.

Fishback, P. V. , J. LaVoice, A. Shertzer, and R. Walsh. Race, Risk and the Emergence of
Federal Redlining. Working Paper, Cambridge, MA: National Bureau of Economic
Research, 2020.

Flood, J. The Fires: How a Computer Formula, Big Ideas, and the Best of Intentions Burned
Down New York City--and Determined the Future of Cities. New York: Riverhead
Books, 2011.

Foner, E. Reconstruction: America’s unfinished revolution, 1863-1877. New York: Harper
Collins, 1988.
164

Fox, J. What Happened When Minneapolis Ended Single-Family Zoning. August 20, 2022.
https://www.bloomberg.com/opinion/articles/2022-08-20/what-happened-when-
minneapolis-ended-single-family-zoning?leadSource=uverify%20wall.

Freeman, L. There Goes the 'Hood: Views of Gentrification from the Ground Up. Philadelphia:
Temple University Press, 2006.

Freund, D. M. P. Colored Property: State Policy & White Racial Politics in Suburban America.
Chicago: The University of Chicago Press, 2007.

Gannett, H. "Proportion of Negro to Total Population at the Twelfth Census, 1900." Frontispiece
of Negroes in the United States. United States Bureau of the Census, 1904.

Gibson, C., and K. Jung. "Historical Census Statistics On Population Totals by Race, 1790 to
1990, For Large Cities and Other Urban Places in the United States." US Census Bureau.
February 2005. https://www.census.gov/library/working-papers/2005/demo/POP-
twps0076.html.

Glaeser, E. "How Biden Can Free America From Its Zoning Straitjacket." The New York Times.
April 12, 2021. https://www.nytimes.com/2021/04/12/opinion/biden-infrastructure-
zoning.html.

Glaeser, E. L., and J. Gyourko. "The Impact of Zoning on Housing Affordability." NBER
Working Paper Series, 2003.

Glaeser, E. Triumph of the City: How Our Greatest Invention Makes Us Richer, Smarter,
Greener, Healthier, and Happier. Penguin Books, 2012.

Glaeser, E., and J. Gyourko. "The Economic Implications of Housing Supply." Journal of
Economic Perspectives—Vol. 32, No. 1, 2018: 3-30.

Glaeser, E., J. Gyourko, and R. Saks. "Why is Manhattan So Expensive? Regulation and the Rise
in Housing Prices." Journal of Law and Economics—Volume XLVIII, 2005: 331-368.

Google. Google Maps. 2022. http://maps.google.com (accessed December 8, 2022).

Gratz, R. B.. The Battle for Gotham: New York in the Shadow of Robert Moses and Jane Jacobs.
New York: Nation Books, 2010.

Haar, C. M., and J. S. Kayden. Zoning and the American Dream: Promises Still to Keep.
Routledge, 1989.

Hannah-Jones, N. "It Was Never About Busing: Court-ordered desegregation worked. But white
racism made it hard to accept." The New York Times Sunday Review. July 12, 2019.
https://www.nytimes.com/2019/07/12/opinion/sunday/it-was-never-about-busing.html.
165

Harrison, Ballard & Allen. Plan for Rezoning the City of New York. New York, 1950.

Haughney, C. "Suddenly, SoHo Heeds Law on Artists’ Lofts." The New York Times. November
11, 2010. https://www.nytimes.com/2010/11/12/nyregion/12soho.html.

Hillier, A. "Redlining and the Homeowners' Loan Corporation." University of Pennsylvania
Scholarly Commons, 2003.

Hirsh, A. R. Making the Second Ghetto: Race and Housing in Chicago 1940-1960 . Chicago:
University of Chicago Press, 1998.

Hogan, G. "Grading De Blasio: In East New York, Mayor’s First Rezoned Neighborhood,
Promises Fall Short." Gothamist. December 22, 2021.
https://gothamist.com/news/grading-de-blasio-in-east-new-york-mayors-first-rezoned-
neighborhood-promises-fall-short.

Hong, S., D. O'Sullivan, and Y. Sadahiro. "Implementing spatial segregation measures in R."
PLOS ONE 9(11): e113767, 2014.

Hughes, C. J. "A New Tower Rises Above Billionaires’ Row." The New York Times. September
6, 2018. https://www.nytimes.com/2018/09/06/realestate/a-new-tower-rises-above-
billionaires-row.html.

Investopedia. Variance Inflation Factor (VIF). n.d.
https://www.investopedia.com/terms/v/variance-inflation-
factor.asp#:~:text=Variance%20inflation%20factor%20(VIF)%20is,only%20that%20sin
gle%20independent%20variable.

IPUMS National Historical GIS. Data Finder. n.d. https://data2.nhgis.org/main.

Jackson, K. T. Crabgrass Frontier: The Suburbanization of the United States. Oxford University
Press, 1987.

Jacobs, J. The Death and Life of Great American Cities. New York: Vintage Books, 1961.

Jonnes, J. Conquering Gotham: Building Penn Station and Its Tunnels. New York: Penguin
Books, 2008.

Justice Sutherland, US Supreme Court. "Village of Euclid v. Ambler Realty Co." Justia: US
Supreme Court. January 26, 1926. https://supreme.justia.com/cases/federal/us/272/365/.

Kahlenberg, R. D. "How Minneapolis Ended Single-Family Zoning." The Century Foundation.
October 24, 2019. https://tcf.org/content/report/minneapolis-ended-single-family-
zoning/?agreed=1.

166

Karas, D. "Highway to Inequity: The Disparate Impact of the Interstate Highway System on Poor
and Minority Communities in American Cities." New Visions for Public Affairs, Vol. 7,
2015: 9-21.

Katumba, S., S. Coetzee, A. Stein, and & I. Fabris-Rotelli. "Using spatial indices to measure
dynamic racial residential segregation in Gauteng province (South Africa)." South
African Geographical Journal, 2021.

Kaul, G. "With covenants, racism was written into Minneapolis housing. The scars are still
visible." MinnPost. February 22, 2019. https://www.minnpost.com/metro/2019/02/with-
covenants-racism-was-written-into-minneapolis-housing-the-scars-are-still-
visible/#:~:text=Whereas%20cities%20like%20Chicago%20and,half%20of%20the%202
0th%20century.

Keller Williams Realty. "Northridge Cooperative East Elmhurst/Jackson Heights." KW
Landmark II. May 20, 2017. https://www.queenshometeam.com/buying-a-home-coop-
condo-in-queens-ny/northridge-east-elmhurstjackson-heights/.

Kilani, H. "'Asthma Alley': why minorities bear bear burden of pollution inequity caused by
white people." The Guardian. April 4, 2019. https://www.theguardian.com/us-
news/2019/apr/04/new-york-south-bronx-minorities-pollution-inequity.

Lamb, C. Housing Segregation in Suburban America since 1960: Presidential and Judicial
Politics. Cambridge, UK: Cambridge University Press, 2005.

Laskow, S. "The quiet, massive rezoning of New York." Politico. February 24, 2014.
https://www.politico.com/states/new-york/city-hall/story/2014/02/the-quiet-massive-
rezoning-of-new-york-078398.

Laurence, P. L. "The Death and Life of Urban Design: Jane Jacobs, The Rockefeller Foundation
and the New Research in Urbanism, 1955–1965." Journal of Urban Design , 2006: 145-
172.

lbennett, user:. The Cross Bronx Expressway and the Ruination of the Bronx. November 10,
2019. The Cross Bronx Expressway and the Ruination of the Bronx.

Lemann, N. The Promised Land: The Great Black Migration and How it Changed America. New
York: Alfred A Knopf, Inc., 1991.

Logan, J. R., and B. J. Stults. "The Persistence of Segregation in the Metropolis: New Findings
from the 2020 Census." Diversity and Disparities Project, Brown University. 2021.
https://s4.ad.brown.edu/Projects/Diversity.

Madison.com. More than 100 years of Navigating the NYC Transit System, in photos. n.d.
https://madison.com/gallery/lifestyles/more-than-100-years-of-navigating-the-new-york-
transit-system-in-photos/collection_d4c72452-a0de-5bd5-83f3-4bab0b7e4522.html.
167

Massey, D. S., and N. A. Denton. "The Dimensions of Residential Segregation." Social Forces
Vol. 67, No. 2, December 1988: 281-315.

Massey, D. S., and N. A. Denton. American Apartheid: Segregation and the Making of the
American Underclass. Cambridge: Harvard University Press, 1998.

McGeehan, P. "A Subway Ride to New Jersey? It Could Happen, Officials Say." The New York
Times. February 27, 2018. https://www.nytimes.com/2018/02/27/nyregion/subway-
expand-new-jersey.html.

Metropolitan Transportation Authority. Second Avenue Subway, phase 2. October 17, 2022.
https://new.mta.info/project/second-avenue-subway-phase-2.

Miller, M. The Interstates and the City: An Examination of Interstate Highways and the
Perception of Slum Clearance in American Cities. Master's Thesis, Wilmington:
University of Delaware, 2018.

Moretti, E. The New Geography of Jobs. New York: Harper Business, 2013.

Municipal Art Society. "Accidental Skyline: A blueprint for a more intentional city." MAS NYC.
2013. https://www.mas.org/initiatives/accidental-skyline/.

—. "The Accidental Skyline." Reports. 2017. https://www.mas.org/wp-
content/uploads/2017/10/accidental-skyline-report-2017.pdf.

National Low Income Housing Coalition. Public Housing History . October 17, 2019.
https://nlihc.org/resource/public-housing-
history#:~:text=The%20federal%20public%20housing%20program,result%20of%20pow
erful%20grassroots%20organizing.

Nelson, R. K., L. Winling, R. Marciano, and N. Connolly. Mapping Inequality: Redlining in New
Deal America. n.d. https://dsl.richmond.edu/panorama/redlining/#loc=12/44.974/-
93.332&text=downloads (accessed July 11, 2020).

Newman, O. "Creating Defensible Space ." U.S. Department of Housing and Urban
Development, Office of Policy Development and Research. April 1996.
https://www.huduser.gov/publications/pdf/def.pdf.

—. Defensible Space: Crime Prevention Through Urban Design. New York: Maximillian Books,
1976.

NYC Department of City Planning. 2020 Census Results for New York City: Key Population and
Housing Statistics. August 2021.
https://www1.nyc.gov/assets/planning/download/pdf/planning-level/nyc-
population/census2020/dcp_2020-census-briefing-booklet-1.pdf?r=3.
168

—. "C 070447 ZMK - Sketch Map." n.d.
https://www.nyc.gov/assets/planning/download/pdf/zoning/zoning-
maps/sketchmaps/skz070447zmk.pdf.

—. Dynamics of Racial/Hispanic Composition in NYC Neighborhoods. November 2021.
https://storymaps.arcgis.com/stories/46a91a58447d4024afd00771eec1dd23.

—. Message from the NYC Department of City Planning. 2020.
https://www1.nyc.gov/site/planning/about/eeo-diversity-statement.page.

—. NYC GIS Zoning Features. July 2021. https://www1.nyc.gov/site/planning/data-maps/open-
data/dwn-gis-zoning.page#metadata.

—. Population. n.d. https://www1.nyc.gov/site/planning/planning-level/nyc-population/nyc-
population.page.

—. Zoning Handbook, 2018 Edition. New York: NYC Department of City Planning, 2018.

NYC Landmarks Preservation Commission. Historic Districts. 2021.
https://data.cityofnewyork.us/Housing-Development/Historic-Districts/xbvj-gfnw.

nycsubway.org. Historical Maps. n.d. https://www.nycsubway.org/wiki/Historical_Maps.

Ogneva-Himmelberger, Y., H. Pearsall, and R. Rakshit. "Concrete evidence & geographically
weighted regression: A regionalanalysis of wealth and the land cover in Massachusetts."
Applied Geography, 2009: 478–487.

Oka, M., and D. W. S. Wong. "Capturing the two dimensions of residential segregation at the
neighborhood level for health research." Frontiers in Public Health, 2014.

Oka, M., and D. W. S. Wong. "Segregation: a multi-contextual and multi-faceted phenomenon in
stratified societies." In Handbook of Urban Geography, by Tim Schwanen (Editor), &
Ronald van Kempen (Editor), 255-280. Cheltanham, UK: Edward Elgar Publishing,
2019.

Opticos Design. Missing Middle Housing. n.d. https://missingmiddlehousing.com/about.

Oser, A. S. "PERSPECTIVES: East Side Zoning; Adjusting the Rules for Avenue Buildings."
The New York Times. December 17, 1989.
https://www.nytimes.com/1989/12/17/realestate/perspectives-east-side-zoning-adjusting-
the-rules-for-avenue-buildings.html.

O'Sullivan, D., and D. J. Unwin. Geographic Information Analysis. Hoboken, NJ: Wiley, 2010.

169

Perry, A. M., J. Rothwell, and D. Harshbarger. "The devaluation of assets in Black
neighborhoods: The case of residential property." The Brookings Institute. November 27,
2018. https://www.brookings.edu/research/devaluation-of-assets-in-black-
neighborhoods/.

Plunz, R. A history of housing in New York City. Columbia University Press, 2016.

Popovich, N., J. Williams, and D. LuMay. "Can Removing Highways Fix America’s Cities?"
The New York Times. May 27, 2021.
https://www.nytimes.com/interactive/2021/05/27/climate/us-cities-highway-
removal.html.

Reardon, S. F., and D. O'Sullivan. "Measures of Spatial Segregation." Sociological Methodology,
2004: 121-162.

Rothstein, R. From Ferguson to Baltimore, the Fruits of Government-Sponsored Segregation.
Working Economics Blog, Economic Policy Institute, 2015.

—. The Color of Law: A Forgotten History of How Our Government Segregated America. New
York: Liveright, 2017.

Rothstein, R. "The Making of Ferguson." Journal of Affordable Housing & Community
Development Law Vol. 24, No. 2, 2015: 165-204.

Rothstein, R. The Making of Ferguson: Public Policies at the Root of its Troubles. Economic
Policy Institute, 2014.

Sagalyn, L. B. "The Cross Manhattan Expressway." Museum of the City of New York. November
14, 2016. https://www.mcny.org/story/cross-manhattan-expressway.

Seamon, D. "A Phenomenological Reading of Jane Jacobs' Death and Life of Great American
Cities." Mediapolis - a journal of cities and culture. October 4, 2019.
https://www.mediapolisjournal.com/2019/10/phenomenological-reading-of-jane-jacobs/.

Sheidlower, N. "The Controversial History of Levittown, America's First Suburb." Untapped
New York. July 31, 2020. https://untappedcities.com/2020/07/31/the-controversial-
history-of-levittown-americas-first-suburb/.

Statology. How to Conduct a Jarque-Bera Test in R. March 10, 2019.
https://www.statology.org/how-to-conduct-a-jarque-bera-test-in-r/.

—. The Breusch-Pagan Test: Definition & Example. December 21, 2020.
https://www.statology.org/breusch-pagan-test/.

170

Stremple, P. "Brooklyn rezoning opponents look beyond Brooklyn for strategies." Brooklyn
Eagle. July 29, 2019. https://brooklyneagle.com/articles/2019/07/29/bushwick-rezoning-
opponents-look-beyond-brooklyn-for-strategies/.

Sugrue, T. J. The Origins of the Urban Crisis: Race and Inequality in Postwar Detroit.
Princeton: Princeton University Press, 2005.

The Central Park Conservancy. "Before Central Park: The Story of Seneca Village." Central
Park Conservancy Magazine. January 18, 2018.
https://www.centralparknyc.org/articles/seneca-village.

The Committee on Slum Clearance Plans. "South Village: Slum Clearance Plan Under TItle I of
the Housing Act of 1949." Department of Housing and Urban Development (HUD)
Office of Policy Development. January 1951.
https://www.huduser.gov/portal/sites/default/files/pdf/South-Village-SCP.pdf.

The Legislature of the State of New York. State Plane Coordinate System. 1995.
https://cdn.ymaws.com/www.nysapls.org/resource/resmgr/mts-legal-ref-page/State-
Plane-Coordinate-Syste.pdf.

The Museum of the City of New York. "The Greatest Grid: The Master Plan of Manhattan 1811
- Now." 2015. https://thegreatestgrid.mcny.org/greatest-grid/.

The National Advisory Commission on Civil Disorders. The Kerner Report. 1968.

The Skyscraper Museum. Urban Fabric exhibition. 1997-2020.
https://old.skyscraper.org/EXHIBITIONS/URBAN_FABRIC/nw02.php.

The Washington Post. Fatal Force. August 5, 2021.
https://www.washingtonpost.com/graphics/investigations/police-shootings-database/.

The White House. FACT SHEET: Biden-⁠Harris Administration Announces New Actions to Build
Black Wealth and Narrow the Racial Wealth Gap. June 1, 2021.
https://www.whitehouse.gov/briefing-room/statements-releases/2021/06/01/fact-sheet-
biden-harris-administration-announces-new-actions-to-build-black-wealth-and-narrow-
the-racial-wealth-gap/.

University of Minnesota Libraries. Mapping Prejudice: Visualizing the hidden histories of race
and privilege in the built environment. n.d. https://mappingprejudice.umn.edu/index.html.

US Census Bureau. Bronx County, New York. July 2021.
https://www.census.gov/quickfacts/fact/table/bronxcountynewyork/PST045221.

—. Explore Census Data. n.d. https://data.census.gov/.

—. TIGER/Line® Shapefiles. n.d. https://www.census.gov/cgi-bin/geo/shapefiles/index.php.
171

Vorhees, Walker, Smith and Smith. "Zoning New York City." NYC Zoning History. 1958.
https://www1.nyc.gov/assets/planning/download/pdf/about/city-planning-
history/zoning_nyc.pdf.

White, A., S. D. Guikema, and T. M. Logan. "Urban population characteristics and their
correlation with historic discriminatory housing practices." Applied Geography, Vol. 132,
2021.

Williams, K. "How Lincoln Center Was Built (It Wasn't Pretty)." The New York Times.
December 21, 2017. https://www.nytimes.com/2017/12/21/nyregion/how-lincoln-center-
was-built-it-wasnt-pretty.html.

Yakas, B. See Incredible Photos Of The NYC Subway Getting Built In The Early 20th Century.
February 13, 2020. https://gothamist.com/arts-entertainment/photos/streetscapes-early-
20th-century-photos?image=9.

Yu, D., and C. Wu. "Understanding Population Segregation from Landsat ETM+ Imagery: A
Geographically Weighted Regression Approach." GIScience & Remote Sensing, 2013:
187-206.

172

Appendix A: Detailed Regression Results – Population density change

Table 16. Regression results, Black population density change, 1930-1960

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.68973 0.51482 *** 0.43903 *** 0.13873 *** 0.65320 ***
Distance to core -0.02403 .
Percent Subway Proximate 0.54708 *** 0.56205 *** 0.62530 *** 0.16820 * 0.10134
Percent HOLC A areas -0.85135 -0.68431 *
Percent HOLC B areas -0.28873
Percent HOLC C areas -0.21368
Percent HOLC D areas 1.33500 * 1.58014 *** 1.68020 *** 0.52538 *** 0.76532 ***
Percent Non-HOLC areas -0.52877 -0.36201 **
Percent Urban Renewal Areas -7.27472 *** -7.23638 *** -7.27599 *** -6.87654 *** -5.91391 ***
Public housing density per acre 0.63290 *** 0.63721 *** 0.62282 *** 0.55838 *** 0.60526 ***
Percent highway proximate -0.60783 ** -0.64030 ** -0.69368 ** -0.41485 ** -0.28828 ***
Percent single-family zoning 0.25477
Percent two-family zoning -0.68430 . -1.02301 *** -0.99969 *** -0.33904 -0.31384
Percent 'contextual' multi-family zoning NA
Percent 'non-contextual' multi-family zoning 0.37132
Percent high-density 'tower' districts -0.55396 -0.80437 ** -1.09870 *** -0.02379 1.01583 *
Percent former M-districts NA
Percent pure non-residential districts -0.52268 . -0.78005 *** -0.96313 *** -0.65357 *** -0.72672 ***
Percent Special Purpose District NA
Percent historic district NA
adjusted R-squared 0.27340 0.27420 0.27120
Rho 0.71690 ***
Lambda 0.76834 ***
AIC 7139.20 5978.70 5984.50
p-value of residuals (LM) < 2.2e-16 3.13E-07 0.99900
1960 - Black change in population per acre (1930-1960)
Spatial Error OLS run 1 OLS run 2 OLS run 3 Spatial Lag
173

Table 17. Regression results, White population density change, 1930-1960

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
-9.72209 *** -9.16892 *** -9.38374 *** -3.38867 *** -11.53386 ***
Distance to core 0.51306 *** 0.50213 *** 0.52246 *** 0.11649 0.83692 ***
Percent Subway Proximate -1.37229 *** -1.36417 *** -1.37031 *** -0.64861 *** -0.96184 ***
Percent HOLC A areas 4.52746 *** 4.29280 *** 4.33818 *** 2.59505 *** 4.28754 ***
Percent HOLC B areas 5.39658 *** 5.16029 *** 5.18765 *** 2.64237 *** 4.07054 ***
Percent HOLC C areas 3.77046 *** 3.54456 *** 3.56592 *** 1.76474 *** 2.37676 ***
Percent HOLC D areas 0.22557
Percent Non-HOLC areas 4.74883 *** 4.51873 *** 4.53170 *** 2.26994 *** 2.48920 ***
Percent Urban Renewal Areas -3.87781 * -3.92396 *
Public housing density per acre 0.37556 *** 0.37314 *** 0.37080 *** 0.45220 *** 0.47825 ***
Percent highway proximate 1.18095 ** 1.18207 ** 1.17127 ** 0.62388 . 0.50436
Percent single-family zoning 3.36005 *** 3.03007 *** 2.99461 *** 1.08984 *** 1.59418 **
Percent two-family zoning 2.67582 *** 2.31928 *** 2.29520 *** 1.61943 ** 1.94913 **
Percent 'contextual' multi-family zoning NA
Percent 'non-contextual' multi-family zoning 3.09600 *** 2.71351 *** 2.68843 *** 1.71360 *** 1.98810 ***
Percent high-density 'tower' districts 3.54160 *** 3.16403 *** 3.13901 *** 1.85364 *** 2.93222 ***
Percent former M-districts NA
Percent pure non-residential districts 0.43188
Percent Special Purpose District NA
Percent historic district NA
adjusted R-squared 0.32770 0.32810 0.32690
Rho 0.63951 ***
Lambda 0.68595 ***
AIC 9925.90 9186.50 9204.40
p-value of residuals (LM) < 2.22e-16 3.90E-05 0.99600
1960 - White change in population per acre (1930-1960)
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
174

Table 18. Regression results, Black population density change, 1960-1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
-2.79972 ** -2.02030 *** -1.17805 *** -0.91091 *** -0.06280
Distance to core 0.05838
Percent Subway Proximate 0.14939
Percent HOLC A areas 1.45346 . 1.49021 .
Percent HOLC B areas 3.44919 *** 3.49805 *** 2.38824 *** 0.87896 *** 1.00621 ***
Percent HOLC C areas 3.07646 *** 3.12884 *** 2.01843 *** 0.72639 *** 0.84762 ***
Percent HOLC D areas 1.09548 . 1.13527 .
Percent Non-HOLC areas 1.90476 * 1.89180 * 0.86307 *** 0.25606 . -0.05422
Percent Urban Renewal Areas 0.33527
Public housing density per acre 0.13974 ** 0.14210 ** 0.14528 ** 0.17682 *** 0.18397
Percent highway proximate 0.72896 * 0.68071 *
Percent single-family zoning 1.39744 ** 1.20651 ** 1.27025 ** 0.73473 ** 0.76361 *
Percent two-family zoning 0.33007
Percent 'contextual' multi-family zoning 1.83681 ** 1.66098 ** 1.75624 ** 1.20967 ** 1.39691 *
Percent 'non-contextual' multi-family zoning 2.40994 *** 2.20678 *** 2.20902 *** 1.42135 *** 1.73020 ***
Percent high-density 'tower' districts 3.13736 *** 2.91547 *** 2.91944 *** 1.68727 *** 1.92694 **
Percent former M-districts 3.58491 *** 3.34546 *** 3.34009 *** 1.91644 *** 1.50865 *
Percent pure non-residential districts 1.39672 ** 1.24620 *** 1.35010 *** 0.39185 . -0.08553
Percent Special Purpose District -0.73135 ** -0.71385 ** -0.63544 ** -0.14124 0.03142
Percent historic district -0.17294 **' -0.17591 *** -0.17612 *** -0.08988 * -0.15088 **
adjusted R-squared 0.12010 0.12090 0.11920
Rho 0.76871 ***
Lambda 0.79692 ***
AIC 8723.00 7407.70 7358.70
p-value of residuals (LM) < 2.2e-16 1.04E-07 0.99900
1990 - Black change in population per acre (1960-1990)
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
175

Table 19. Regression results, White population density change, 1960-1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
-3.63611 ** -3.11487 *** -3.12820 *** -1.25075 *** -2.51354 ***
Distance to core 0.09584
Percent Subway Proximate -1.41780 *** -1.44782 *** -1.43010 *** -0.65590 *** -1.02922 ***
Percent HOLC A areas 0.67763 1.24493 *
Percent HOLC B areas -0.88262 -0.34912 .
Percent HOLC C areas -0.64098
Percent HOLC D areas -0.38469
Percent Non-HOLC areas 2.45076 ** 2.95598 *** 2.99110 *** 1.71135 *** 2.03755 ***
Percent Urban Renewal Areas 0.69721
Public housing density per acre -0.43067 *** -0.40818 *** -0.40410 *** -0.27615 *** -0.29928 ***
Percent highway proximate 0.23560
Percent single-family zoning 0.89639 . 1.05812 ** 1.11730 ** 0.24135 0.66282
Percent two-family zoning 3.08476 *** 3.19061 *** 3.22170 *** 1.20514 ** 1.30827 **
Percent 'contextual' multi-family zoning 4.85376 *** 5.00832 *** 5.30710 *** 2.55529 *** 3.49561 ***
Percent 'non-contextual' multi-family zoning 1.23253 * 1.40598 *** 1.39590 *** 0.55255 * 0.61295 *
Percent high-density 'tower' districts 3.07360 *** 3.11375 *** 3.69780 *** 1.18711 ** 1.63365 *
Percent former M-districts 3.30282 *** 3.30771 *** 3.55840 *** 1.83040 * 2.27839 **
Percent pure non-residential districts -0.27119
Percent Special Purpose District 0.38007 0.37448
Percent historic district 0.09782 . 0.08148 .
adjusted R-squared 0.21910 0.21940 0.21590
Rho 0.60824 ***
Lambda 0.63461 ***
AIC 9096.50 8487.20 8533.10
p-value of residuals (LM) < 2.2e-16 4.77E-15 0.99900
Spatial Lag Spatial Error
1990 - White change in population per acre (1960-1990)
OLS run 1 OLS run 2 OLS run 3
176

Table 20. Regression results, Black population density change, 1990-2020

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept -2.14893 ** -2.17412 ** -0.64463 *** -0.71377 *** -0.74784 ***
Distance to core 0.15580 * 0.15548 *
Percent Subway Proximate -0.49276 *** -0.50203 *** -0.50951 *** -0.15670 -0.26084
Percent HOLC A areas -0.18736
Percent HOLC B areas -1.55114 * -1.49995 *** -1.46560 *** -0.60524 *** -0.71300 **
Percent HOLC C areas -1.39771 * -1.34632 *** -1.31968 *** -0.61254 *** -0.75103 ***
Percent HOLC D areas -1.07908 -1.05083 *** -1.04147 *** -0.41287 ** -0.58863 **
Percent Non-HOLC areas -0.03101
Percent Urban Renewal Areas -0.37159
Public housing density per acre -0.40148 *** -0.40360 *** -0.41407 *** -0.31568 *** -0.34229 ***
Percent highway proximate -0.58911 . -0.58405 .
Percent single-family zoning 1.50621 *** 1.48988 *** 1.67491 *** 1.28986 *** 1.66435 ***
Percent two-family zoning 3.08178 *** 3.07739 *** 3.27389 *** 1.94609 *** 2.30846 ***
Percent 'contextual' multi-family zoning 1.37326 *** 1.36959 *** 1.37042 *** 1.34775 *** 1.52013 ***
Percent 'non-contextual' multi-family zoning 2.02031 *** 2.00866 *** 2.10484 *** 1.52524 *** 1.42245 ***
Percent high-density 'tower' districts 2.93886 *** 2.89182 *** 2.63514 *** 1.77797 *** 1.89396 ***
Percent former M-districts 4.05261 *** 4.07515 *** 3.82214 *** 2.73067 *** 3.10566 ***
Percent pure non-residential districts 1.88504 *** 1.87016 *** 1.82094 *** 1.52461 *** 1.73296 ***
Percent Special Purpose District -0.00686
Percent historic district -2.49059 *** -2.50145 *** -2.63954 *** -1.34023 *** -1.57972 ***
adjusted R-squared 0.15550 0.15680 0.15400
Rho 0.70160 ***
Lambda 0.72687 ***
AIC 9053.20 8051.80 8067.20
p-value of residuals (LM) < 2.2e-16 1.40E-13 0.99900
2020 - Black change in population per acre (1990-2020)
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
177

Table 21. Regression results, White population density change, 1990-2020

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept 2.19378 ** 2.39521 *** 2.88188 *** 0.53389 0.40695
Distance to core -0.29502 *** -0.29666 *** -0.35830 *** -0.05976 -0.13512 *
Percent Subway Proximate 0.07700
Percent HOLC A areas -2.20727 * -2.40063 *** -2.25826 *** -0.95910 * -0.25598
Percent HOLC B areas 0.37632
Percent HOLC C areas 0.25649
Percent HOLC D areas 2.42726 ** 2.09902 *** 2.10609 *** 0.69045 *** 1.00516 ***
Percent Non-HOLC areas 1.09373 0.78679 *** 0.85536 *** 0.47804 *** 1.27606 ***
Percent Urban Renewal Areas 1.80685 ** 1.80395 ** 1.80183 ** 0.15927 0.13658
Public housing density per acre -0.03472
Percent highway proximate -0.98730 ** -1.01797 ** -1.04139 ** -0.22324 0.01288
Percent single-family zoning -0.59402 -0.53971
Percent two-family zoning -1.50494 *** -1.48083 *** -1.28492 *** -0.61967 *** -0.53106 *
Percent 'contextual' multi-family zoning -1.10738 * -1.01753 *** -0.75491 ** -0.41050 * -0.34243
Percent 'non-contextual' multi-family zoning -1.63042 *** -1.57266 *** -1.33750 *** -0.75676 *** -1.13377 ***
Percent high-density 'tower' districts 0.12007
Percent former M-districts -0.33462
Percent pure non-residential districts -0.15744
Percent Special Purpose District 0.83967 *** 0.89449 *** 0.92849 *** 0.34087 * 0.69657 **
Percent historic district 0.60807 0.66740 .
adjusted R-squared 0.17780 0.17960 0.17850
Rho 0.76708 ***
Lambda 0.79560 ***
AIC 9485.30 8108.00 8099.70
p-value of residuals (LM) < 2.2e-16 < 2.2e-16 0.9990
Spatial Lag Spatial Error
2020 - White change in population per acre (1990-2020)
OLS run 1 OLS run 2 OLS run 3
178

Appendix B: Detailed Regression Results – Dissimilarity

Table 22. Regression results, Black-other dissimilarity, 1960

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.14060 ** 0.11293 *** 0.11159 *** 0.00870 *** 0.01631 *
Distance to core -0.00338
Percent Subway Proximate 0.05199 *** 0.05319 *** 0.05380 *** 0.00280 0.00547
Percent HOLC A areas -0.06997 . -0.07452 *** -0.07942 *** -0.01514 * -0.02757 *
Percent HOLC B areas -0.09419 ** -0.10001 *** -0.10218 *** -0.01549 *** -0.02102 ***
Percent HOLC C areas -0.09670 ** -0.10305 *** -0.10518 *** -0.01154 *** -0.01145 **
Percent HOLC D areas 0.00454
Percent Non-HOLC areas -0.09962 ** -0.10485 *** -0.10743 *** -0.01920 *** -0.02264 ***
Percent Urban Renewal Areas 0.01605
Public housing density per acre 0.00118
Percent highway proximate -0.03028 * -0.03011 *
Percent single-family zoning 0.01320
Percent two-family zoning -0.02287 -0.02870
Percent 'contextual' multi-family zoning NA
Percent 'non-contextual' multi-family zoning 0.00534
Percent high-density 'tower' districts -0.05877 * -0.05700 *** -0.05550 ** 0.00437 0.02479 .
Percent former M-districts NA
Percent pure non-residential districts -0.05012 ** -0.05350 *** -0.05552 *** -0.00719 * 0.00428
Percent Special Purpose District NA
Percent historic district NA
adjusted R-squared 0.18490 0.18640 0.18450
Rho 0.95806 ***
Lambda 0.96758 ***
AIC -5260.40 -9625.10 -9588.60
p-value of residuals (LM) < 2.2e-16 4.33E-13 0.00100
1960 - Black - other dissimilarity
Spatial Error OLS run 1 OLS run 2 OLS run 3 Spatial Lag
179

Table 23. Regression results, White-other dissimilarity, 1960

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.12634 ** 0.11286 * 0.11152 *** 0.00873 *** 0.01624
Distance to core -0.00208
Percent Subway Proximate 0.05272 *** 0.05341 *** 0.05402 *** 0.00279 0.00518
Percent HOLC A areas -0.06833 . -0.07393 *** -0.07881 *** -0.01482 * -0.02713 *
Percent HOLC B areas -0.09110 ** -0.09817 *** -0.10034 *** -0.01533 *** -0.02038 ***
Percent HOLC C areas -0.09381 ** -0.10126 *** -0.10338 *** -0.01139 *** -0.01078 *
Percent HOLC D areas 0.00600 -0.10126 ***
Percent Non-HOLC areas -0.09772 ** -0.10405 *** -0.10663 *** -0.01909 *** -0.02213 ***
Percent Urban Renewal Areas 0.02213
Public housing density per acre 0.00159
Percent highway proximate -0.03060 * -0.03019 *
Percent single-family zoning 0.01115
Percent two-family zoning -0.02354 -0.02834
Percent 'contextual' multi-family zoning NA
Percent 'non-contextual' multi-family zoning 0.00440
Percent high-density 'tower' districts -0.05838 * -0.05872 *** -0.05722 *** 0.00377 0.02406 .
Percent former M-districts NA
Percent pure non-residential districts -0.04986 ** -0.05335 *** -0.05539 *** -0.00733 * 0.00412
Percent Special Purpose District NA
Percent historic district NA
adjusted R-squared 0.18150 0.18320 0.18140
Rho 0.95673 ***
Lambda 0.96640 ***
AIC -5250.70 -9573.80 -9537.20
p-value of residuals (LM) < 2.2e-16 4.89E-11 0.00100
1960 - White - other dissimilarity
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
180

Table 24. Regression results, Black-other dissimilarity, 1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.20801 *** 0.23303 *** 0.30573 *** 0.04416 * 0.28262 *
Distance to core -0.01678 *** -0.01790 *** -0.01533 ** -0.00334 . -0.02058
Percent Subway Proximate 0.00869
Percent HOLC A areas 0.15543 ** 0.15606 **
Percent HOLC B areas 0.14866 ** 0.15288 ***
Percent HOLC C areas 0.14061 ** 0.14508 **
Percent HOLC D areas 0.16824 *** 0.17173 *** 0.03532 *** 0.00861 * 0.00266
Percent Non-HOLC areas 0.12755 ** 0.12643 **
Percent Urban Renewal Areas 0.05687 . 0.05857 .
Public housing density per acre -0.00084
Percent highway proximate -0.03379 . -0.03604 .
Percent single-family zoning 0.01528
Percent two-family zoning -0.02661 -0.04486 *
Percent 'contextual' multi-family zoning 0.14893 *** 0.13415 *** 0.15145 *** 0.03549 * 0.01674
Percent 'non-contextual' multi-family zoning 0.01995
Percent high-density 'tower' districts -0.03339 -0.04718 .
Percent former M-districts -0.10501 * -0.11772 *
Percent pure non-residential districts -0.08987 *** -0.10326 *** -0.10770 *** -0.01931 ** 0.00718
Percent Special Purpose District -0.02283 . -0.02161
Percent historic district 0.00693 * 0.00726 *
adjusted R-squared 0.07501 0.07597 0.06127
Rho 0.89104 ***
Lambda 0.90559 ***
AIC -3857.80 -6589.90 -6563.90
p-value of residuals (LM) < 2.2e-16 2.58E-08 0.99900
1990 - Black - other dissimilarity
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
181

Table 25. Regression results, White-other dissimilarity, 1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.03243 -0.03093 -0.02791 -0.06256 *** 0.01912
Distance to core -0.00534
Percent Subway Proximate 0.02398 ** 0.02887 *** 0.03262 *** 0.00988 * 0.00895
Percent HOLC A areas 0.28085 *** 0.26790 *** 0.28931 *** 0.10412 *** 0.10825 **
Percent HOLC B areas 0.26886 *** 0.25910 *** 0.25854 *** 0.10383 *** 0.10338 ***
Percent HOLC C areas 0.25189 *** 0.24208 *** 0.23890 *** 0.10841 *** 0.11848 ***
Percent HOLC D areas 0.29470 *** 0.28942 *** 0.29687 *** 0.12056 *** 0.12395 ***
Percent Non-HOLC areas 0.25520 *** 0.24820 *** 0.25327 *** 0.10118 *** 0.11327 ***
Percent Urban Renewal Areas 0.06887 * 0.07271 *
Public housing density per acre 0.00628 * 0.00640 *
Percent highway proximate -0.02451
Percent single-family zoning -0.02700 .
Percent two-family zoning -0.01214
Percent 'contextual' multi-family zoning 0.23466 *** 0.25041 *** 0.23812 *** 0.06716 *** 0.08241 **
Percent 'non-contextual' multi-family zoning -0.01336
Percent high-density 'tower' districts -0.01307
Percent former M-districts -0.20846 *** -0.18848 *** -0.19754 *** -0.06556 * -0.07642 *
Percent pure non-residential districts -0.12525 *** -0.11374 *** -0.11701 *** -0.02995 *** -0.00871
Percent Special Purpose District 0.03598 ** 0.03654 **
Percent historic district 0.01398 *** 0.01534 *** 0.01520 *** 0.00416 * 0.00141
adjusted R-squared 0.14170 0.14230 0.13690
Rho 0.84444 ***
Lambda 0.87253 ***
AIC -3872.20 -5966.10 -5910.80
p-value of residuals (LM) < 2.2e-16 0.00015 0.99000
Spatial Lag Spatial Error
1990 - White - other dissimilarity
OLS run 1 OLS run 2 OLS run 3
182

Table 26. Regression results, White-other dissimilarity, 1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
-0.00360 0.00155 -0.02931 -0.04718 * 0.01570
Distance to core 0.00980 * 0.00983 * 0.01515 *** 0.00454 * 0.00684 *
Percent Subway Proximate -0.01988 * -0.01964 *
Percent HOLC A areas 0.02855
Percent HOLC B areas -0.00355
Percent HOLC C areas 0.00245
Percent HOLC D areas 0.06396 0.06342 *** 0.05762 *** 0.01413 *** 0.01628 *
Percent Non-HOLC areas 0.01956 0.01767
Percent Urban Renewal Areas 0.05016 0.04916
Public housing density per acre -0.00724 * -0.00722 *
Percent highway proximate -0.04679 * -0.04573 * 0.00895 0.04044 ***
Percent single-family zoning 0.04749 . 0.04228 *
Percent two-family zoning 0.05312 * 0.04768 **
Percent 'contextual' multi-family zoning 0.10112 *** 0.09448 *** 0.05086 *** 0.02781 *** 0.02757 **
Percent 'non-contextual' multi-family zoning 0.10954 *** 0.10250 *** 0.05523 *** 0.01210 * 0.00153
Percent high-density 'tower' districts 0.14562 *** 0.14428 *** 0.11650 *** 0.03191 ** 0.01524
Percent former M-districts 0.02060
Percent pure non-residential districts 0.00663
Percent Special Purpose District -0.06055 *** -0.05911 *** -0.06387 *** -0.00546 0.00165
Percent historic district 0.06759 ** 0.06940 ** 0.07022 ** 0.00932 -0.00933
adjusted R-squared 0.06291 0.06443 0.05533
Rho 0.88891 ***
Lambda 0.89819 ***
AIC -3820.10 -6645.70 -6692.40
p-value of residuals (LM) < 2.2e-16 1.32E-09 0.99900
2020 - Black - other dissimilarity
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
183

Table 27. Regression results, White-other dissimilarity, 2020

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.03473 0.03535 -0.03210 -0.08022 *** 0.00088
Distance to core -0.00631 . -0.00497
Percent Subway Proximate -0.00478
Percent HOLC A areas 0.25886 *** 0.22221 *** 0.23421 *** 0.14476 *** 0.16835 ***
Percent HOLC B areas 0.17472 *** 0.14973 *** 0.15208 *** 0.10957 *** 0.11817 ***
Percent HOLC C areas 0.19015 *** 0.16229 *** 0.16552 *** 0.11753 *** 0.13098 ***
Percent HOLC D areas 0.22462 *** 0.19047 *** 0.20153 *** 0.11881 *** 0.13481 ***
Percent Non-HOLC areas 0.17767 *** 0.14643 *** 0.15798 *** 0.10276 *** 0.11755 ***
Percent Urban Renewal Areas 0.06073 . 0.05673 .
Public housing density per acre -0.00483 . -0.00486 .
Percent highway proximate 0.00029
Percent single-family zoning -0.02236
Percent two-family zoning 0.02233 0.04151 ** 0.04017 ** -0.00134 -0.02871 **
Percent 'contextual' multi-family zoning 0.06541 ** 0.08279 *** 0.08879 *** 0.02515 *** 0.01639
Percent 'non-contextual' multi-family zoning 0.05695 ** 0.07617 * 0.07729 *** 0.00870 -0.01190
Percent high-density 'tower' districts 0.16826 *** 0.18393 *** 0.20487 *** 0.03046 ** 0.01426
Percent former M-districts -0.06690 . -0.05356
Percent pure non-residential districts -0.02741
Percent Special Purpose District 0.07389 *** 0.07304 *** 0.06950 *** 0.01122 . 0.01777 .
Percent historic district 0.14575 *** 0.14550 *** 0.15147 *** 0.02753 ** 0.01588
adjusted R-squared 0.15430 0.15500 0.15260
Rho 0.88325 ***
Lambda 0.90295 ***
AIC -4331.70 -6897.40 -6862.40
p-value of residuals (LM) < 2.2e-16 0.0441 0.9460
Spatial Lag Spatial Error
2020 - White - other dissimilarity
OLS run 1 OLS run 2 OLS run 3
184

Appendix C: Detailed Regression Results – Isolation

Table 28. Regression results, Black isolation, 1960

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.05007 0.01936 0.08167 *** 0.00708 *** 0.01222
Distance to core 0.00605 . 0.00594 .
Percent Subway Proximate 0.03931 *** 0.03885 *** 0.03666 *** -0.00108 0.00058
Percent HOLC A areas -0.12369 *** -0.10214 *** -0.10238 *** -0.00850 -0.01108
Percent HOLC B areas -0.13749 *** -0.11636 *** -0.11728 *** -0.01349 *** -0.02215 ***
Percent HOLC C areas -0.13532 *** -0.11437 *** -0.11450 *** -0.01067 *** -0.01555 ***
Percent HOLC D areas -0.02274
Percent Non-HOLC areas -0.12111 *** -0.09664 *** -0.09816 *** -0.01105 *** -0.01554 ***
Percent Urban Renewal Areas 0.07373
Public housing density per acre 0.00403 0.00392
Percent highway proximate -0.03593 ** -0.03382 *
Percent single-family zoning -0.01363
Percent two-family zoning -0.04140 .
Percent 'contextual' multi-family zoning NA
Percent 'non-contextual' multi-family zoning -0.02349
Percent high-density 'tower' districts -0.09341 *** -0.07308 *** -0.08730 *** -0.00189 0.00419
Percent former M-districts NA
Percent pure non-residential districts -0.05598 ** -0.03929 *** -0.04715 *** -0.00500 * 0.00031
Percent Special Purpose District NA
Percent historic district NA
adjusted R-squared 0.18240 0.18230 0.17910
Rho 0.97805 ***
Lambda 0.98389 ***
AIC -5342.30 -10855.00 -10829.00
p-value of residuals (LM) < 2.2e-16 < 2.2e-16 0.00100
1960 - Black isolation
Spatial Error OLS run 1 OLS run 2 OLS run 3 Spatial Lag
185

Table 29. Regression results, White isolation, 1960

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.25989 *** 0.25781 *** -0.26470 *** -0.05032 0.28746 *
Distance to core -0.03751 *** -0.03733 *** -0.03815 *** -0.00556 * -0.02925 *
Percent Subway Proximate 0.07290 *** 0.07284 *** 0.07331 *** 0.02665 *** 0.03537 ***
Percent HOLC A areas 0.31523 * 0.31550 ** 0.32476 ** 0.11150 *** 0.08744 *
Percent HOLC B areas 0.32864 *** 0.32883 *** 0.33892 *** 0.14006 *** 0.14951 ***
Percent HOLC C areas 0.28838 ** 0.28849 *** 0.30065 *** 0.13478 *** 0.14438 ***
Percent HOLC D areas 0.23782 ** 0.23751 *** 0.24848 *** 0.12461 *** 0.12077 ***
Percent Non-HOLC areas 0.21748 0.21734 *** 0.22151 *** 0.10397 *** 0.09759 ***
Percent Urban Renewal Areas -0.02850
Public housing density per acre -0.00051
Percent highway proximate 0.10008 *** 0.09982 *** 0.09539 *** 0.01920 . 0.02933 *
Percent single-family zoning 0.11765 *** 0.11761 *** 0.05943 ** 0.04105 *** 0.03202 *
Percent two-family zoning 0.09692 ** 0.09691 **
Percent 'contextual' multi-family zoning NA
Percent 'non-contextual' multi-family zoning 0.14868 0.14865 *** 0.08133 *** 0.05113 *** 0.04722 ***
Percent high-density 'tower' districts 0.23959 *** 0.23985 *** 0.17830 *** 0.08527 *** 0.14045 ***
Percent former M-districts NA
Percent pure non-residential districts 0.06415 ** 0.06459 **
Percent Special Purpose District NA
Percent historic district NA
adjusted R-squared 0.20830 0.20900 0.20610
Rho 0.84585 ***
Lambda 0.87712 ***
AIC -3901.10 -6028.00 -5948.00
p-value of residuals (LM) 2.92E-05 0.99900
1960 - White isolation
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
186

Table 30. Regression results, Black isolation, 1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
-0.21272 ** -0.25630 ** -0.28949 *** 0.00528 0.10169
Distance to core 0.02869 *** 0.02859 *** 0.03118 *** -0.00045 -0.00813
Percent Subway Proximate 0.02637 ** 0.02790 ** 0.02598 ** -0.00200 -0.00476
Percent HOLC A areas 0.01276
Percent HOLC B areas 0.05791 0.06336 ** 0.07170 ** -0.00287 -0.01173 .
Percent HOLC C areas 0.02135 0.02690 * 0.03391 **
Percent HOLC D areas 0.09935 * 0.10734 *** 0.11095 *** 0.00713 * 0.00950 .
Percent Non-HOLC areas -0.01604 -0.00273 0.00466
Percent Urban Renewal Areas 0.10599 ** 0.10602 ** 0.10832 ** -0.00223 -0.00010
Public housing density per acre 0.01337 *** 0.01302 *** 0.01411 *** 0.00214 * -0.00006
Percent highway proximate -0.07468 *** -0.07024 *** -0.07601 *** 0.00126 0.01569 .
Percent single-family zoning -0.06172 *
Percent two-family zoning -0.17244 *** -0.12262 *** -0.12117 *** 0.00013 0.00637
Percent 'contextual' multi-family zoning -0.25381 *** -0.20370 *** -0.19352 *** -0.01579 . -0.00580
Percent 'non-contextual' multi-family zoning -0.05615 . .
Percent high-density 'tower' districts -0.05551
Percent former M-districts -0.15939 ** -0.11636 *
Percent pure non-residential districts -0.05988 * -0.02783 .
Percent Special Purpose District -0.09059 *** -0.09205 *** -0.09035 *** -0.00273 -0.00531
Percent historic district 0.01139 *** 0.01123 *** 0.01138 ** -0.00077 -0.00310 *
adjusted R-squared 0.14000 0.14020 0.13780
Rho 0.94918 ***
Lambda 0.95331 ***
AIC -3463.00 -8230.40 -8228.30
p-value of residuals (LM) < 2.2e-16 1.35E-06 0.99700
1990 - Black isolation
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
187

Table 31. Regression results, White isolation, 1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.34292 *** 0.33791 *** 0.38540 *** 0.04740 * 0.39531 ***
Distance to core -0.03881 *** -0.03854 *** -0.04069 *** -0.00865 *** -0.03838 ***
Percent Subway Proximate -0.03317 *** -0.03330 *** -0.03394 *** -0.00099 0.00107
Percent HOLC A areas 0.23920 *** 0.23916 *** 0.26218 *** 0.08192 *** 0.05303 .
Percent HOLC B areas 0.18644 *** 0.18621 *** 0.20294 *** 0.07105 *** 0.05850 **
Percent HOLC C areas 0.19553 *** 0.19513 *** 0.21137 *** 0.07575 *** 0.06778 **
Percent HOLC D areas 0.14158 *** 0.14110 *** 0.15107 *** 0.06320 ** 0.05236 *
Percent Non-HOLC areas 0.25533 *** 0.25518 *** 0.26483 *** 0.07962 *** 0.07790 ***
Percent Urban Renewal Areas -0.06221 * -0.06161 *
Public housing density per acre -0.01513 *** -0.01508 *** -0.01507 *** -0.00413 *** -0.00260 *
Percent highway proximate 0.03653 * 0.03665 *
Percent single-family zoning 0.04041 0.04321 .
Percent two-family zoning 0.14667 *** 0.14974 *** 0.10467 *** 0.00024 -0.01660
Percent 'contextual' multi-family zoning 0.49312 *** 0.49648 *** 0.47033 *** 0.09502 *** 0.06754 **
Percent 'non-contextual' multi-family zoning 0.04402 0.04755 *
Percent high-density 'tower' districts 0.04662 0.05053 .
Percent former M-districts -0.01486
Percent pure non-residential districts -0.06725 ** -0.06435 ** -0.10224 *** -0.01809 ** 0.00372
Percent Special Purpose District 0.10729 *** 0.10697 *** 0.10924 *** 0.00758 0.01702
Percent historic district 0.00579 * 0.00574 *
adjusted R-squared 0.32430 0.32450 0.32120
Rho 0.85465 ***
Lambda 0.89463 ***
AIC -4415.60 -7174.30 -7105.00
p-value of residuals (LM) < 2.2e-16 < 2.2e-16 0.99900
Spatial Lag Spatial Error
1990 - White isolation
OLS run 1 OLS run 2 OLS run 3
188

Table 32. Regression results, Black isolation, 2020

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
-0.29407 *** -0.31085 *** -0.31058 *** -0.03033 * -0.03033
Distance to core 0.04538 *** 0.04467 *** 0.04409 *** 0.00356 ** 0.00356 **
Percent Subway Proximate -0.02144 * -0.02110 *
Percent HOLC A areas -0.18886 *** -0.12818 *** -0.11870 *** -0.00908 -0.00908
Percent HOLC B areas -0.12905 ** -0.06753 *** -0.06573 *** -0.01403 *** -0.01403 ***
Percent HOLC C areas -0.15100 ** -0.08826 *** -0.08911 *** -0.00890 * -0.00890 *
Percent HOLC D areas -0.06288
Percent Non-HOLC areas -0.14154 ** -0.08814 *** -0.08348 *** -0.00481 -0.00481
Percent Urban Renewal Areas 0.11317 ** 0.11404 ** 0.10625 ** 0.02953 * 0.02953 *
Public housing density per acre 0.00998 ** 0.01029 ** 0.01006 ** -0.00036 -0.00036
Percent highway proximate -0.09206 *** -0.09578 *** -0.09574 *** 0.00659 0.00659
Percent single-family zoning -0.03525 -0.06062 *** -0.05153 ** -0.00818 -0.00818
Percent two-family zoning -0.03219 -0.05731 *** -0.05075 *** -0.00417 -0.00417
Percent 'contextual' multi-family zoning -0.02133 -0.04959 *** -0.05076 *** 0.00082 0.00082
Percent 'non-contextual' multi-family zoning 0.03130
Percent high-density 'tower' districts 0.03255
Percent former M-districts 0.00649
Percent pure non-residential districts 0.01531 .
Percent Special Purpose District -0.09259 *** -0.09240 *** -0.09680 *** -0.00503 -0.00503
Percent historic district 0.04837 * 0.04596 *
adjusted R-squared 0.14510 0.14560 0.14360
Rho 0.93075 ***
Lambda 0.93075 ***
AIC -3472.80 -8302.60 -8305.20
p-value of residuals (LM) < 2.2e-16 4.59E-14 0.99900
2020 - Black isolation
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
189

Table 33. Regression results, White isolation, 2020

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.35541 *** 0.35216 *** 0.33930 *** -0.03712 * -0.05374 *
Distance to core -0.05230 *** -0.05131 *** -0.05032 *** -0.00548 *** 0.00237
Percent Subway Proximate -0.01823 * -0.02001 **
Percent HOLC A areas 0.46957 *** 0.47550 *** 0.47615 *** 0.11400 *** 0.10214 ***
Percent HOLC B areas 0.34888 *** 0.35355 *** 0.34902 *** 0.08043 *** 0.05726 **
Percent HOLC C areas 0.32420 *** 0.32857 *** 0.32358 *** 0.08198 *** 0.06542 ***
Percent HOLC D areas 0.34416 *** 0.34726 *** 0.34334 *** 0.08298 *** 0.06561 ***
Percent Non-HOLC areas 0.36472 *** 0.36650 *** 0.36728 *** 0.07125 *** 0.04580 ***
Percent Urban Renewal Areas -0.01161
Public housing density per acre -0.01170 *** -0.01191 *** -0.01222 *** -0.00331 *** -0.00132 **
Percent highway proximate -0.01154
Percent single-family zoning 0.03122
Percent two-family zoning 0.05072 * 0.03835 *** 0.04449 *** -0.00376 -0.02007 ***
Percent 'contextual' multi-family zoning 0.07174 *** 0.05932 *** 0.05440 *** 0.00882 * 0.00788
Percent 'non-contextual' multi-family zoning 0.01306
Percent high-density 'tower' districts 0.12443 *** 0.11381 *** 0.10586 *** 0.01466 . 0.01344
Percent former M-districts -0.10148 ** -0.11139 ** -0.11884 *** -0.01849 -0.03346 *
Percent pure non-residential districts -0.05413 ** -0.06736 *** -0.06861 *** -0.00815 . 0.01296 .
Percent Special Purpose District 0.12028 *** 0.12006 *** 0.11744 *** 0.00792 . 0.02123 **
Percent historic district 0.14469 *** 0.14692 *** 0.14199 *** 0.02410 ** 0.0103 **
adjusted R-squared 0.33340 0.33370 0.33200
Rho 0.93150 ***
Lambda 0.95358 ***
AIC -4465.20 -8363.20 -8315.30
p-value of residuals (LM) < 2.2e-16 < 2.2e-16 0.8260
Spatial Lag Spatial Error
2020 - White isolation
OLS run 1 OLS run 2 OLS run 3
190

Appendix D: Detailed Regression Results – HOLC boundaries

Table 34. Regression results, HOLC A boundaries, 1960

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.01419 0.00237 0.00304 . -0.00130 0.00574
Percent Urban Renewal Areas -0.06979
Public housing density per acre -0.00143
Percent highway proximate 0.02302 . 0.02425 .
Percent single-family zoning 0.08453 *** 0.09962 *** 0.10295 *** 0.05274 *** 0.10320 ***
Percent two-family zoning 0.03066 0.04635 *
Percent 'contextual' multi-family zoning N/A
Percent 'non-contextual' multi-family zoning -0.01667
Percent high-density 'tower' districts 0.21153 *** 0.22566 *** 0.22563 *** 0.06459 *** 0.18859 ***
Percent former M-districts N/A
Percent pure non-residential districts -0.03056 -0.01625 .
Percent Special Purpose District N/A
Percent historic district N/A
adjusted R-squared 0.11760 0.11770 0.11370
Rho 0.85522 ***
Lambda 0.86970 ***
AIC -5346.10 -6942.60 -6998.10
p-value of residuals (LM) < 2.2e-16 1.53E-08 0.02000
1960 - HOLC - A
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
191

Table 35. Regression results, HOLC B boundaries, 1960

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.08140 * 0.12252 *** 0.12384 *** 0.02186 *** 0.09742 ***
Percent Urban Renewal Areas -0.25677 -0.25796
Public housing density per acre -0.03242 *** -0.03205 *** -0.03293 *** -0.00857 * -0.00660
Percent highway proximate 0.03848
Percent single-family zoning 0.13889 ** 0.09100 ** 0.09090 ** 0.03664 * 0.10649
Percent two-family zoning 0.15528 * 0.10185 .
Percent 'contextual' multi-family zoning N/A
Percent 'non-contextual' multi-family zoning 0.05703
Percent high-density 'tower' districts 0.04979
Percent former M-districts N/A
Percent pure non-residential districts -0.16514 *** -0.20972 *** -0.21288 *** -0.07277 *** -0.08473
Percent Special Purpose District N/A
Percent historic district N/A
adjusted R-squared 0.04902 0.04944 0.04767
Rho 0.87939 ***
Lambda 0.88549 ***
AIC -576.80 -2666.60 -2653.40
p-value of residuals (LM) < 2.2e-16 0.1139 0.2170
1960 - HOLC - B
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
192

Table 36. Regression results, HOLC C boundaries, 1960

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
-0.03649 0.00480 -0.00259 -0.06602 *** 0.09516 ***
Percent Urban Renewal Areas -0.41154 . -0.41383 .
Public housing density per acre -0.06703 *** -0.06721 *** -0.06636 *** -0.01527 ** -0.01052 .
Percent highway proximate 0.10791 * 0.10606 *
Percent single-family zoning 0.42244 *** 0.36919 *** 0.39137 *** 0.11035 *** 0.14673 **
Percent two-family zoning 0.68755 *** 0.63279 *** 0.64623 *** 0.21215 *** 0.28322 ***
Percent 'contextual' multi-family zoning N/A
Percent 'non-contextual' multi-family zoning 0.52731 *** 0.46956 *** 0.48875 *** 0.19678 *** 0.25286 ***
Percent high-density 'tower' districts -0.09112 -0.14431 *
Percent former M-districts N/A
Percent pure non-residential districts 0.06535
Percent Special Purpose District N/A
Percent historic district N/A
adjusted R-squared 0.13260 0.13260 0.12800
Rho 0.85280 ***
Lambda 0.87451 ***
AIC 605.15 -1327.40 -1301.60
p-value of residuals (LM) < 2.2e-16 0.3994 0.9460
OLS run 2
1960 - HOLC - C
OLS run 1 OLS run 3 Spatial Lag Spatial Error
193

Table 37. Regression results, HOLC D boundaries, 1960

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.00214 -0.03966 . -0.03966 . -0.04709 *** 0.03570
Percent Urban Renewal Areas 0.79978 *** 0.80410 *** 0.80410 *** 0.38440 *** 0.29692 **
Public housing density per acre 0.07697 *** 0.07741 *** 0.07741 *** 0.00992 * -0.00202
Percent highway proximate -0.24698 *** -0.24274 *** -0.24274 *** -0.07778 ** -0.11367 ***
Percent single-family zoning -0.08142
Percent two-family zoning -0.04664
Percent 'contextual' multi-family zoning N/A
Percent 'non-contextual' multi-family zoning 0.28128 *** 0.33980 *** 0.33980 *** 0.12412 *** 0.17939 ***
Percent high-density 'tower' districts 0.30470 *** 0.35961 *** 0.35961 *** 0.07282 * 0.02812
Percent former M-districts N/A
Percent pure non-residential districts 0.36483 *** 0.41764 *** 0.41764 *** 0.07118 *** 0.03876
Percent Special Purpose District N/A
Percent historic district N/A
adjusted R-squared 0.10880 0.10900 0.10900
Rho 0.87789 ***
Lambda 0.89845 ***
AIC 263.54 -1947.20 -1953.40
p-value of residuals (LM) < 2.2e-16 0.7538 0.8610
1960 - HOLC - D
Spatial Lag Spatial Error OLS run 1 OLS run 2 OLS run 3
194

Table 38. Regression results, HOLC A boundaries, 1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.01423 0.01602 . -0.00463 ** -0.00434 *** -0.00179
Percent Urban Renewal Areas -0.03270 -0.03318
Public housing density per acre 0.00029
Percent highway proximate 0.01075
Percent single-family zoning 0.04869 ** 0.04748 ** 0.07501 *** 0.04674 *** 0.09243 ***
Percent two-family zoning 0.00237
Percent 'contextual' multi-family zoning 0.15030 *** 0.14866 *** 0.17614 *** 0.02621 . 0.01065
Percent 'non-contextual' multi-family zoning -0.02732 -0.02886 *
Percent high-density 'tower' districts 0.12427 *** 0.12276 *** 0.14650 *** 0.03332 ** 0.13326 ***
Percent former M-districts -0.07102 * -0.07230 *
Percent pure non-residential districts -0.02969 . -0.02972 *
Percent Special Purpose District 0.10087 *** 0.10104 *** 0.10197 *** 0.04211 *** 0.08344 ***
Percent historic district 0.00835 *** 0.00825 *** 0.00762 *** 0.00530 *** 0.00514 .
adjusted R-squared 0.19890 0.19970 0.19700
Rho 0.83123 ***
Lambda 0.85466 ***
AIC -5553.70 -6876.50 -6923.60
p-value of residuals (LM) < 2.2e-16 3.84E-13 0.00300
1990 - HOLC - A
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
195

Table 39. Regression results, HOLC B boundaries, 1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.07265 . 0.12217 *** 0.12922 *** 0.02343 *** 0.11822 ***
Percent Urban Renewal Areas -0.22221 ** -0.22255 ** -0.23607 *** -0.08465 * -0.07950 .
Public housing density per acre -0.03079 *** -0.03103 *** -0.03288 *** -0.00927 ** -0.00745 *
Percent highway proximate -0.02268
Percent single-family zoning 0.15465 ** 0.09163 **
Percent two-family zoning 0.11885 .
Percent 'contextual' multi-family zoning 0.15473 * 0.09443
Percent 'non-contextual' multi-family zoning 0.07203
Percent high-density 'tower' districts -0.18574 ** -0.25431 *** -0.24938 *** -0.05085 . -0.05719
Percent former M-districts -0.24774 * -0.31058 ** -0.32663 *** -0.08922 . -0.10260
Percent pure non-residential districts -0.13147 * -0.19742 *** -0.21073 *** -0.07128 *** -0.07201 ***
Percent Special Purpose District 0.12848 *** 0.13169 *** 0.13190 *** 0.02075 0.02490
Percent historic district 0.02862 *** 0.02864 *** 0.03027 *** 0.01148 *** 0.01324 **
adjusted R-squared 0.07727 0.07700 0.07372
Rho 0.88667 ***
Lambda 0.89483 ***
AIC -533.48 -2636.10 -2605.10
p-value of residuals (LM) < 2.2e-16 0.1264 0.1660
1990 - HOLC - B
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
196

Table 40. Regression results, HOLC C boundaries, 1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.00089 0.05129 * 0.07239 *** -0.04815 *** 0.12613 ***
Percent Urban Renewal Areas -0.59281 *** -0.58393 *** -0.60111 *** -0.15627 ** -0.11743 .
Public housing density per acre -0.06540 *** -0.06529 *** -0.06514 *** -0.01409 ** -0.01246 *
Percent highway proximate 0.07933 0.07883
Percent single-family zoning 0.38535 *** 0.32085 *** 0.30220 *** 0.07669 ** 0.12898 **
Percent two-family zoning 0.60149 *** 0.53574 *** 0.51079 *** 0.22567 *** 0.35965 ***
Percent 'contextual' multi-family zoning 0.17558 . 0.12300
Percent 'non-contextual' multi-family zoning 0.51532 *** 0.44403 *** 0.42021 *** 0.17430 *** 0.24658 ***
Percent high-density 'tower' districts 0.09045
Percent former M-districts 0.01550
Percent pure non-residential districts 0.07792
Percent Special Purpose District -0.17853 *** -0.17329 *** -0.17486 *** -0.02910 -0.05339
Percent historic district -0.02876 *** -0.02886 *** -0.02747
***
0.00396 0.00825
adjusted R-squared 0.16070 0.16130 0.16050
Rho 0.85501 ***
Lambda 0.88008 ***
AIC 535.13 -1309.80 -1299.80
p-value of residuals (LM) < 2.2e-16 0.0237 0.3810
OLS run 2
1990 - HOLC - C
OLS run 1 OLS run 3 Spatial Lag Spatial Error
197

Table 41. Regression results, HOLC D boundaries, 1990

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
-0.00667 -0.03040 -0.03613 . -0.04288 *** 0.08299
Percent Urban Renewal Areas 0.78252 *** 0.78095 *** 0.79066 *** 0.22495 *** 0.18192 ***
Public housing density per acre 0.08816 *** 0.08831 *** 0.08942 *** 0.01884 *** 0.01015 *
Percent highway proximate -0.17691 *** -0.17444 *** -0.17760 *** -0.04951 . -0.06325 .
Percent single-family zoning -0.05114
Percent two-family zoning -0.00848
Percent 'contextual' multi-family zoning 0.49807 *** 0.51258 *** 0.52601 *** 0.18540 *** 0.33549 ***
Percent 'non-contextual' multi-family zoning 0.26010 *** 0.29199 *** 0.29468 *** 0.10354 *** 0.14541 ***
Percent high-density 'tower' districts 0.25847 ** 0.28189 *** 0.24086 *** 0.02217 -0.12386 .
Percent former M-districts 0.69990 *** 0.71512 *** 0.70787 *** 0.24477 *** 0.26911 ***
Percent pure non-residential districts 0.31385 *** 0.34371 *** 0.34783 *** 0.06768 ** 0.04456
Percent Special Purpose District -0.07738 * -0.07727 *
Percent historic district -0.00586
adjusted R-squared 0.19020 0.19080 0.18910
Rho 0.87549 ***
Lambda 0.90550 ***
AIC 122.10 -2015.10 -1986.30
p-value of residuals (LM) < 2.2e-16 0.7662 0.7890
1990 - HOLC - D
Spatial Lag Spatial Error OLS run 1 OLS run 2 OLS run 3
198

Table 42. Regression results, HOLC A boundaries, 2020

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.01151 0.00000 -0.00254 -0.00370 ** -0.00056
Percent Urban Renewal Areas -0.05182 * -0.05514 **
Public housing density per acre -0.00125
Percent highway proximate 0.00941
Percent single-family zoning 0.03619 * 0.05128 *** 0.05649 *** 0.03558 *** 0.06832 ***
Percent two-family zoning -0.01764
Percent 'contextual' multi-family zoning -0.01336
Percent 'non-contextual' multi-family zoning -0.01721
Percent high-density 'tower' districts 0.16758 *** 0.18340 *** 0.18361 *** 0.04629 *** 0.10444 ***
Percent former M-districts -0.07171 ** -0.05576 *
Percent pure non-residential districts -0.03044 * -0.01527
Percent Special Purpose District 0.04079 *** 0.04147 *** 0.03902 *** 0.01893 *** 0.04226 ***
Percent historic district 0.09105 *** 0.09251 *** 0.09154 *** 0.05082 *** 0.06518 ***
adjusted R-squared 0.14080 0.14140
Rho 0.80788 ***
Lambda 0.82425 ***
AIC -5932.70 -7276.70 -7303.60
p-value of residuals (LM) < 2.2e-16 8.23E-03 0.28100
2020 - HOLC - A
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
199

Table 43. Regression results, HOLC B boundaries, 2020

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.08642 ** 0.10518 *** 0.13135 *** 0.01444 *** 0.11050 ***
Percent Urban Renewal Areas -0.25919 *** -0.25794 *** -0.25333 *** -0.07201 . -0.06629
Public housing density per acre -0.03328 *** -0.03320 *** -0.03151 *** -0.00637 * -0.00512
Percent highway proximate -0.00434
Percent single-family zoning 0.15012 ** 0.12561 *** 0.09501 ** 0.07967 *** 0.20882 ***
Percent two-family zoning -0.05667 -0.07968 ** -0.11439 *** -0.00009 0.07027 ***
Percent 'contextual' multi-family zoning 0.03111
Percent 'non-contextual' multi-family zoning 0.07856 . 0.05206 *
Percent high-density 'tower' districts -0.17315 ** -0.19682 *** -0.23723 *** -0.04310 . -0.06041
Percent former M-districts -0.31099 *** -0.33371 *** -0.37147 *** -0.07352 . -0.06289
Percent pure non-residential districts -0.15923 *** -0.18236 *** -0.21124 *** -0.05238 ** -0.03892 .
Percent Special Purpose District 0.08855 *** 0.08833 *** 0.08783 *** 0.00505 -0.01609
Percent historic district 0.23715 *** 0.24012 *** 0.22188 *** 0.13263 *** 0.17092 ***
adjusted R-squared 0.08360 0.08416 0.08223
Rho 0.89181 ***
Lambda 0.90313 ***
AIC -691.05 -2876.40 -2894.80
p-value of residuals (LM) < 2.2e-16 0.0207 0.2170
Spatial Lag Spatial Error
2020 - HOLC - B
OLS run 1 OLS run 2 OLS run 3
200

Table 44. Regression results, HOLC C boundaries, 2020

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
-0.02521 -0.00327 0.04498 * -0.05831 *** 0.12776 ***
Percent Urban Renewal Areas -0.51029 *** -0.50575 *** -0.50658 *** -0.07932 -0.05961
Public housing density per acre -0.05121 *** -0.05101 *** -0.05092 *** -0.00975 * -0.00859 .
Percent highway proximate 0.08572 . 0.08510 .
Percent single-family zoning 0.41066 *** 0.38150 *** 0.32997 *** 0.09336 *** 0.15614 ***
Percent two-family zoning 0.68998 *** 0.66118 *** 0.60706 *** 0.25822 *** 0.40306 ***
Percent 'contextual' multi-family zoning 0.41323 *** 0.38503 *** 0.32882 *** 0.16741 *** 0.22755 ***
Percent 'non-contextual' multi-family zoning 0.48301 *** 0.45150 *** 0.39757 *** 0.15226 *** 0.17436 ***
Percent high-density 'tower' districts 0.06155
Percent former M-districts 0.06040
Percent pure non-residential districts 0.12949 * 0.10399 *
Percent Special Purpose District -0.14858 *** -0.14053 *** -0.14691 *** -0.01792 -0.02958
Percent historic district -0.14775 ** -0.14364 ** -0.16338 ** -0.02754 -0.04765
adjusted R-squared 0.19270 0.19310 0.19100
Rho 0.85089 ***
Lambda 0.88457 ***
AIC 450.78 -1457.00 -1476.40
p-value of residuals (LM) < 2.2e-16 0.0127 0.4220
2020 - HOLC - C
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error
201

Table 45. Regression results, HOLC D boundaries, 2020

Variables Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
y-intercept
0.02491 0.05185 *** 0.02524 * -0.01069 0.13217 ***
Percent Urban Renewal Areas 0.76661 *** 0.76426 *** 0.76884 *** 0.23469 *** 0.18159 ***
Public housing density per acre 0.08450 *** 0.08437 *** 0.08530 *** 0.02087 *** 0.01507 ***
Percent highway proximate -0.19318 *** -0.19655 *** -0.20820 *** -0.07347 ** -0.08204 *
Percent single-family zoning -0.06879 -0.10093 *
Percent two-family zoning 0.04327
Percent 'contextual' multi-family zoning 0.43648 *** 0.40264 *** 0.43565 *** 0.13087 *** 0.20559 ***
Percent 'non-contextual' multi-family zoning 0.18174 *** 0.14513 *** 0.17762 *** 0.05604 *** 0.09610 ***
Percent high-density 'tower' districts 0.29748 *** 0.26248 *** 0.24696 *** 0.00297 -0.13269 *
Percent former M-districts 0.72183 *** 0.68877 *** 0.68145 *** 0.25953 *** 0.28005 ***
Percent pure non-residential districts 0.14704 ** 0.11566 ** 0.14497 *** -0.03284 -0.04988 .
Percent Special Purpose District -0.07062 * -0.07016 *
Percent historic district -0.16538 *** -0.16889 *** -0.17520 *** -0.13730 *** -0.14388 ***
adjusted R-squared 0.23760 0.23770 0.23420
Rho 0.86716 ***
Lambda 0.90166 ***
AIC -59.67 -2168.20 -2139.20
p-value of residuals (LM) < 2.2e-16 0.8811 0.8460
2020 - HOLC - D
OLS run 1 OLS run 2 OLS run 3 Spatial Lag Spatial Error

Abstract (if available)

Abstract In the summer of 2020, sustained violence against Black Americans by law enforcement erupted into nationwide protests following the callous murder of George Floyd. The cultural zeitgeist prompted a call to action, not only to rethink our policing, but also to examine larger systemic and institutionalized racism in our society. In urban planning circles, this discussion often begins with an examination of the role “redlining” maps created in the 1930s by the federal government, which controversially appraised lending risk with a racial lens, stigmatizing areas with Black residents, outlined in red, as risky for investment, and contributing to ensuing segregation.

Through examination of the nation’s largest metropolis, New York, this thesis evaluates whether redlining was only one factor of government policy – federal or municipal – entrenching segregation in the landscape. Global and local spatial clustering and segregation measures were conducted in 10-year intervals from 1910 to 2020 to evaluate underlying shifts in the spatial patterns of Black and White population segments over time. Linear regression, spatial error, and spatial lag models were then constructed to evaluate the degree to which redlining, urban renewal designations, public housing concentrations, zoning designations and historic districting contributed to the spatial segregation of Black and White populations in three distinct years: 1960, 1990 and 2020. The findings showed each era of new urban planning policy contributed to persisting segregation. The findings also showed that oftentimes a new generation of policy would spatially reference a prior era, to the benefit or detriment of a particular population: Urban renewal designations mimicked redlined areas and disproportionately concentrated public housing into increasingly Black enclaves, while exclusionary zoning tools like single-family zoning, often mimicked the safest investment designations in redlining maps, prolonging the privilege of predominantly White communities.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Spatiotemporal analysis of the SLOSH and ADCIRC storm surge models: a case study of hurricane Ida

PDF

The impact of severe coastal flooding on economic recovery disparities: a study of New Jersey communities following Hurricane Sandy

PDF

Modeling the spatio-temporal variability of solar radiation on buildings: a case study of Lewis Hall

PDF

The role of precision in spatial narratives: using a modified discourse quality index to measure the quality of deliberative spatial data

PDF

The impact of definition criteria on mapped wildland-urban interface: a case study for ten counties along the Oregon-California border

PDF

Geographic information systems and marketing: a transdisciplinary approach to curriculum development

PDF

Development of a historical urban land use web application for the city of Hong Kong

PDF

Investigating bus route walkability: comparative case study in Orange County, California

PDF

Assessing modern conflict to monitor human rights with remote sensing: Russia's war in Ukraine

PDF

Spatiotemporal visualization and analysis as a policy support tool: a case study of the economic geography of tobacco farming in the Philippines

PDF

An exploratory spatial analysis of fire service and EMS accessibility in northeastern Illinois communities

PDF

Developing a replicable approach for the creation of urban climatic maps for urban heat island analysis: a case study for the city of Los Angeles, California

PDF

Advancing Redwood City's bicycle infrastructure through a geodesign workflow

PDF

Providing a new low-cost primary care facility for under-served communities: a site suitability analysis for Service Planning Area 6 in Los Angeles County, California

PDF

Implementing spatial thinking with Web GIS in the non-profit sector: a case study of ArcGIS Online in the Pacific Symphony

PDF

Generating bicyclist counts using volunteered and professional geographic information through a mobile application

PDF

Philly Bike Report: a mobile app for mapping and sharing real-time reports of illegally blocked bike lanes in Philadelphia

PDF

A spatial analysis of veteran healthcare accessibility

PDF

Urban green space accessibility and environmental justice: a GIS-based analysis in the city of Phoenix, Arizona

PDF

Exploring global natural disaster and climate migration data: a Web GIS application

Asset Metadata

Creator Hayner, Christopher Steven (author)

Core Title Exploring the pernicious effects of redlining and discriminatory policies on an American city: a spatio-temporal case study of New York City

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Degree Conferral Date 2022-12

Publication Date 01/07/2023

Defense Date 12/14/2023

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

Tag OAI-PMH Harvest,redlining,segregation,Urban planning,zoning

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Sedano, Elisabeth (committee chair), Ruddell, Darren (committee member), Vos, Robert (committee member)

Creator Email chayner@usc.edu,christopher.hayner@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC112710936

Unique identifier UC112710936

Identifier etd-HaynerChri-11404.pdf (filename)

Legacy Identifier etd-HaynerChri-11404

Document Type Thesis

Format theses (aat)

Rights Hayner, Christopher Steven

Internet Media Type application/pdf

Type texts

Source 20230111-usctheses-batch-1000 (batch), 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 cisadmin@lib.usc.edu