University of Southern California Dissertations and Theses

Urban consumer amenities and their accessibility

(USC Thesis Other)

Urban consumer amenities and their accessibility

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Urban Consumer Amenities and their Accessibility
by
Clemens Andre Pilgram
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(URBAN PLANNING AND DEVELOPMENT)
August 2024
Copyright 2024 Clemens Andre Pilgram

To my grandmother Beryl, who departed from us before having a chance to see this
dissertation. May she rest in peace.
ii

Acknowledgements
As is normal for doctoral studies thesedays, I have too many people to thank to fit on one page.
First, I would like to thank my committee chair Marlon Boarnet and my committee members
Geoff Boeing, and Christian Redfearn for the guidance and mentorship provided over the past
years. I would also like to thank Matt Kahn, Deborah Salon, Nicholas Duquette, Jorge de la Roca,
Annette Kim, Dowell Myers, Richard Green, Moussa Diop, Genevieve Giuliano, and David King
for sharpening my understanding of the fields in which I operate, and what it means to be an
academic.
Further, I would like to thank Sarah West, who has acted in roles as diverse as coauthor, mentor,
academic life coach, and friend going all the way back to my days as an undergraduate student at
Macalester College.
To the fellow members of my cohort - Robert Binder, Alanna Coombs, Alina Ha, Eli Joun,
David Flores Moctezuma, Sam Ross-Brown, Renzhi Shen, Rebecca Smith, Bonnie Wang, John
Zhao, and Hai Zhou - as well as program-mates Jaime Lopez, Lizhong Liu, James Gross, and
Qifan Shao - I thank you for the mutual support during the program, and for making our shared
time at USC as engaging and rewarding as it was.
To George Papadimitriou, Gene Burinsky, Guillermo Castro, Liana Engie, Ben Vatterott, Mingyu
Ding, Erfan Zeraat, Kylie Trettner, Su Lei, Cyril Hui, Aaron Wirthwein, Ryan Kynor, Theo Van
Straten, Perle Knauss, Espen Aas, Liz Fu, Leo Lerner, Jay Park, Julian Kaiser, Lynne Cherchia,
Hirona Arai, Steph Britt, Ben Perez, Nico Perez, Kun Du, and all the various other members of the
USC Cycling Club - I would like to thank you for helping me stay sane and for helping me explore
Greater Los Angeles - especially during the long and otherwise lonely early days of the Covid-19
pandemic in which we had little but our bikes and each other.
iii

To Angel Yu, Joseph Natividad, and Jack McCarthy - thank you for supporting my academic
career by allowing me to stay at your places for conferences during my PhD.
To Andy Timm, Brendan Rome, Adam and Lizzy Perruzzi, H Trostle, Andreas Scholten,
Robert Haag, Timofej Gießler, Ayano Alexis Terai, Tori Lewis, Miranda Adams, Jared Willard,
Patrick Riechert, and Perry Campbell - thank you for keeping me entertained and having my back.
To Dolos - thank you for keeping the physical areas surrounding my code bug- and rodent-free.
To Rory Huang - thank you for bearing with me as we navigate the wild ride of being a dualacademic couple.
Finally, I would like to thank my parents Cathy and Paul as well as my brother Mark for never
hesitating to support my dreams, even if it means living an ocean away.
iv

Table of Contents
Dedication . . ii
Acknowledgments iii
List of Tables viii
List of Figures ix
Abstract . . . xii
Chapter 1: Introduction 1
Chapter 2: Transit Station Area Walkability: Identifying Impediments to Walking Using
Scalable, Recomputable Land-Use Measures 6
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Study Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Methods and Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.1 Data Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5.1 Differences in Station-to-Block Measures across Station Areas . . . . . . . 23
2.5.2 Station-to-block pathway measures versus circular buffers . . . . . . . . . 26
2.5.3 Identification of detrimental parcels in station areas . . . . . . . . . . . . . 29
2.6 OpenStreetMap as a data source for land uses . . . . . . . . . . . . . . . . . . . . 32
2.7 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Chapter 3: Pathway buffer measures of pedestrian quality for the entire United States 39
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2.1 Assessing Walkability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2.2 Equity of Public Transit Systems . . . . . . . . . . . . . . . . . . . . . . . 45
3.2.3 The First/Last Mile Problem . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.4 Walking to School . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.5 Contribution of this paper . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3 Data Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
v
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . .

3.3.1 US Census . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3.2 Public Transit Locations: Transitland . . . . . . . . . . . . . . . . . . . . 48
3.3.3 Schools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.4 OpenStreetMap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3.4.1 Street Networks . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3.4.2 Land Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3.5 Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.4.1 Experienced Disamenity Shares . . . . . . . . . . . . . . . . . . . . . . . 53
3.4.2 Regression Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.5 Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.5.1 Nationwide Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.5.2 MSA-Level Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.5.3 City-Level Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.5.4 Regression Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.6 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Chapter 4: Developer’s Choice: The Evolving Provision of New Housing Supply & Declining Elasticity . . 68
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.2 Framing Our “Channels” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3 Channel Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.4 The Dynamics Across Four Supply Channels . . . . . . . . . . . . . . . . . . . . 79
4.5 Changing Channels: Why? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.6 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Chapter 5: Online Housing Listings and Preferences 105
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.3 The Online Housing Listing landscape . . . . . . . . . . . . . . . . . . . . . . . . 111
5.4 Study Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.4.1 Study Area: Los Angeles and Orange Counties . . . . . . . . . . . . . . . 112
5.4.2 New Legal Possibilities: ADUs and Upzoning . . . . . . . . . . . . . . . . 113
5.4.3 New Amenities: Public Transit . . . . . . . . . . . . . . . . . . . . . . . . 114
5.5 Data Collection and Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.6 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.6.1 Labeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.6.2 Generating Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . 125
5.7 Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.7.1 Tracking Changes to References Over Time . . . . . . . . . . . . . . . . . 127
5.7.1.1 Covid-Era Preferences . . . . . . . . . . . . . . . . . . . . . . . 127
5.7.1.2 Housing Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.7.1.3 Public Transit . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.7.2 Differences across Submarkets and Space . . . . . . . . . . . . . . . . . . 134
5.7.2.1 Housing Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
vi
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .

5.7.2.2 Public Transit . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.8 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
Chapter 6: Conclusion 148
References . 151
vii
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

List of Tables
2.1 Pathway Buffers versus 800 Meter Circular Buffer Land Use Shares (%), year 2020 27
2.2 GLUS versus OSM Land Use Measures for Blue Line Station Pathway Buffers,
800 Meters Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1 Population shares that can walk to amenities across all Metropolitan Areas Studied 54
3.2 Experienced Disamenity Land Use Shares Walking to Bus Stops in Select Metropolitan Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3 Experienced Disamenity Land Use Shares Walking to Rail Stations in Select Metropolitan Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4 Experienced Disamenity Land Use Shares Walking to Elementary Schools in Select Metropolitan Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.5 Regression Analyses for Walking Routes to Bus Stops and Elementary Schools . . 62
4.1 Channel Shares, All MSAs combined, 2000 and 2010 . . . . . . . . . . . . . . . . 79
4.2 Share of Net New Units by Supply Channel, 2000-10 and 2010-20 . . . . . . . . . 82
4.3 Changes in Housing Stock by Supply Channel, 2000-10 and 2010-20 . . . . . . . . 84
4.4 Growth Rates by Supply Channel, 2000-10 and 2010-20 . . . . . . . . . . . . . . 86
5.1 Observation Counts by Property Type . . . . . . . . . . . . . . . . . . . . . . . . 116
5.2 Sample Sizes by Date and Property Type . . . . . . . . . . . . . . . . . . . . . . . 117
5.3 References to common attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
viii

List of Figures
2.1 Map of the Study area in 2020, Northern Portion . . . . . . . . . . . . . . . . . . 18
2.2 Map of the Study area in 2020, Southern Portion . . . . . . . . . . . . . . . . . . 19
2.3 Illustration of Routing and Pathway Buffer Generation . . . . . . . . . . . . . . . 21
2.4 A grain silo viewed from three different distances. . . . . . . . . . . . . . . . . . . 22
2.5 Northern Stations Population-weighted Pathway Buffers . . . . . . . . . . . . . . 24
2.6 Southern Stations Population-weighted Pathway Buffers . . . . . . . . . . . . . . 25
2.7 Northern Station Population-weighted Pathway-buffered Detrimental Land Uses . . 30
2.8 Southern Station Population-weighted Pathway-buffered Detrimental Land Uses . . 31
3.1 Share of Disamenity Land Uses Encountered on Walks to Bus Stops in MinneapolisSaint Paul, Minnesota . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.2 Share of Disamenity Land Uses Encountered on Walks to Rail Stations in MinneapolisSaint Paul, Minnesota . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.1 Inflation-adjusted housing costs and Per Capita Housing Starts . . . . . . . . . . . 73
4.2 Housing Supply Over Two 10-Year Periods . . . . . . . . . . . . . . . . . . . . . 80
4.3 NNU Shares vs Stock by Channel: Exurban . . . . . . . . . . . . . . . . . . . . . 88
4.4 NNU Shares vs Stock by Channel: Suburban . . . . . . . . . . . . . . . . . . . . 88
4.5 NNU Shares vs Stock by Channel: Urban . . . . . . . . . . . . . . . . . . . . . . 89
4.6 NNU Shares vs Stock by Channel: Post-Industrial . . . . . . . . . . . . . . . . . . 90
4.7 Exurban Channel Shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.8 Suburban Channel Shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.9 Urban Channel Shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.10 Post-Industrial Channel Shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
ix

4.11 Map of Fast Food Restaurants in the Los Angeles MSA . . . . . . . . . . . . . . . 96
4.12 Map of Non-Chain Restaurants in the Los Angeles MSA . . . . . . . . . . . . . . 97
4.13 Map of Access & New Net Supply – Los Angeles MSA . . . . . . . . . . . . . . . 98
4.14 Map of Access & New Net Supply – Phoenix MSA . . . . . . . . . . . . . . . . . 99
4.15 Map of Access & New Net Supply – Austin MSA . . . . . . . . . . . . . . . . . . 100
4.16 Map of Access & New Net Supply – Chicago MSA . . . . . . . . . . . . . . . . . 101
4.17 Map of Access & New Net Supply – Detroit MSA . . . . . . . . . . . . . . . . . . 102
5.1 Spatial distribution of listings for Single-Family Properties . . . . . . . . . . . . . 118
5.2 Spatial distribution of listings for Condominiums and Townhomes . . . . . . . . . 119
5.3 Spatial distribution of listings for Multi-Family Properties . . . . . . . . . . . . . . 120
5.4 Spatial distribution of listings for Manufactured Homes . . . . . . . . . . . . . . . 121
5.5 Spatial distribution of listings for Vacant Land . . . . . . . . . . . . . . . . . . . . 122
5.6 References to Covid-Era Phenomena: Telework (dashed), Virtual Tours (dotted),
Commuting (solid) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.7 References to Accessory Dwelling Units by Property Type . . . . . . . . . . . . . 129
5.8 References to Senate Bills 9 and 10 by Property Type . . . . . . . . . . . . . . . . 130
5.9 Share of ADU references that discuss hypothetical vs. actual ADUs . . . . . . . . 131
5.10 References to TOC and ED1 over time . . . . . . . . . . . . . . . . . . . . . . . . 132
5.11 References to Public Transit by Property Type . . . . . . . . . . . . . . . . . . . . 133
5.12 References to Accessory Dwelling Units by Lot Size . . . . . . . . . . . . . . . . 134
5.13 References to Accessory Dwelling Units by ZCTA Income . . . . . . . . . . . . . 135
5.14 References to Accessory Dwelling Units by ZCTA Latino Population Share . . . . 136
5.15 References to Accessory Dwelling Units by ZCTA Non-Hispanic White Population Share . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
5.16 Listings in 2020 (Orange: References ADUs, Black: Does not reference ADUs) . . 138
5.17 Listings in 2024 (Orange: References ADUs, Black: Does not reference ADUs) . . 139
5.18 Percentage Point change in references to ADUs, 2020 to 2023 . . . . . . . . . . . 140
x

5.19 References to Public Transit by Property Type and Distance to nearest LA Metro
Rail station . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.20 Share of Sample by distance to nearest LA Metro Rail station . . . . . . . . . . . . 143
5.21 Spatial distribution of references to public transit . . . . . . . . . . . . . . . . . . 145
xi

Abstract
This dissertation discusses issues related to the urban experience, as it relates to transportation,
housing searches, and where new housing units are being developed. All four papers are at the intersection of data analytics and urban policy analysis, broadly falling under the heading of Regional
Science. They explore how novel data sources – including volunteer-generated geographical information such as OpenStreetMap and scraped online housing listings – can inform our understanding
of local-scale urban experiences. Specifically, I explore how to measure the quality of pedestrian
experiences, and what role consumption amenities and their accessibility play in housing development and the marketing of residential units, exploring such issues at broader geographic scales
than were common previously.
xii

Chapter 1
Introduction
Everyone wants to live somewhere full of nice things: Beyond just being places with agglomeration effects, much of the urban experience is about consumption (Couture and Handbury, 2020).
Indeed, much of urban studies discourse in the past decade has been about the return of the city;
about how the Millenial generation demands high-amenity urban lifestyles and acts as a driving
force behind gentrification (H. Lee, 2020; Moos, 2016; Myers, 2016). Indeed, in the fourth paper
of this dissertation, I observe through online housing listings - now the most common way of marketing residential properties - that sellers and their agents routinely call attention to the presence
of nearby shopping and dining.
But - how do people access such amenities? Per the 2017 National Household Travel Survey -
the perhaps most representative source on how Americans meet their transportation needs, 92.8%
of all shopping trips and 88.6% of all restaurant trips are conducted by personal automobile. Even
for trips under a mile, cars remain the most common mode of transportation recorded in the NHTS;
only in dense environments do other modes of transportation account for greater mode shares (Federal Highway Administration, 2017). This raises the policy question of how to promote the kinds
of urban environments in which urbanites (and suburbanites) can access employment as well as
urban consumer amenities by modes other than by car, and how to make such trips pleasant and
inviting. Several bodies of research relate to this, with streams of research on urban design (e.g.
Calthorpe, 1989), or on what land uses are associated with active modes of travel (e.g. Boarnet,
Forsyth, Day, and Oakes, 2011). Another approach to promoting the creation of highly amenitized
1

urban environments is via infill development, as infill development necessary to achieve population densities densities at which active travel is possible (Cerin, Sallis, Salvo, Hinckson, Conway,
Owen, et al., 2022), and at which public transit infrastructure becomes financially viable (Cervero
and Guerra, 2011).
This is by no means the first dissertation on any of these topics. However, in the past, many
inquiries into such topics were constrained in scale by data availability or by available computing
resources. As a result, much research in the past has taken the form of local case studies, limiting itself to whatever was feasible within the given constraints. Recent advances in computing
power and data availability have allowed building on such case studies, but at far broader a scale:
In what could be referred to as ”high-N descriptive work” or ”social science without samples”,
researchers attempt to study phenomena by looking at the entire relevant population rather than
basing studies on more accessible subsets or samples. An example for this is Boeing, Pilgram, and
Lu, 2024’s attempt to explore the relationships between street network design and transport sector
CO2 emissions; which reviews an existing local case-study based literature on the same topic and
then tests whether the same associations exist on a global scale. In this spirit, I seek to take existing
questions - about walkability, about real estate development, and about related urban issues - and
build on existing work by scaling up and zooming out. In doing so, I hope to nudge the literature
toward questions about generalizability and to ask whether prior research discusses general trends,
or whether it suffers from bias due to choices of where studies were conducted.
This dissertation consists of four papers: Chapters 2 and 3 introduce and apply a pathwaybuffer based measure of approximating walkability: In Chapter 2, ”Transit Station Area Walkability: Identifying Impediments to Walking Using Scalable, Recomputable Land-Use Measures”,
Sarah West and I review the literature on the quality of the pedestrian experience and on how to determine whether any given walk is pleasant, propose a novel, pathway buffer-based and populationweighted way of measuring what land uses pedestrians are likely to encounter on a particular kind
of trip. This method is based on the premise of that there are associations between land use and
pleasure or displeasure among pedestrians, as is documented by the environmental psychology
2

literature (Guo and Ferreira, 2008; Hartig, Evans, Jamner, Davis, and G arling, 2003; LaJeunesse, ¨
Ryus, Kumfer, Kothuri, and Nordback, 2021). We then apply said method to first-/last-mile trips to
light rail stations in South Minneapolis. We demonstrate that this approach yields markedly different estimates of land use composition within station areas than simple circular buffers - which remain common in mapping of station areas. This paper responds to several literatures that attempt to
measure walkability via environmental audits (Boarnet, Day, Alfonzo, Forsyth, and Oakes, 2006;
Boarnet, Forsyth, Day, and Oakes, 2011; Brown, Werner, Amburgey, and Szalay, 2007; S. Lee and
Talen, 2014) or computer vision methods (Yin and Wang, 2016; F. Zhang, Zhou, Liu, Liu, Fung,
Lin, and Ratti, 2018), building on it by offering a scalable method that allows for comparative
statements of walkability across places using open source geographic data and requiring relatively
little computational resources. It thus seeks to complement more locally grounded, qualitative
methods with scale.
Chapter 3, ”Pathway buffer measures of pedestrian quality for the entire United States”, takes
this same analytical approach and applies it to walks from each census block to the nearest bus
stop, rail transit stop, and elementary school in almost 200 metropolitan areas in the United States
to create population-wide estimates of pedestrian quality. I demonstrate that significant inequities
along racial lines exist not only in whether accomplishing certain errands on foot is possible within
a set distance (half a mile, a full mile), but also in what land uses exist along the way for such walks:
While Non-Hispanic White populations tend to live in lower-density, lower-access neighborhoods
compared to other ethnic groups, they face fewer ”disamenity” land uses such as industrial corridors, railways, highways, or major roadways on such walks conditional on being within half a
mile (or a full mile) walking distance of such destinations. To my knowledge, it is the first effort
to score walkability on this broad a geographic scale; with it, I seek to frame disparities in the
pedestrian experience as an environmental justice issue.
Chapter 4, ”Developer’s Choice: The Evolving Provision of New Housing Supply & Declining
Elasticity”, explores how residential real estate development has shifted from greenfield to infill
across 52 metropolitan statistical areas in the United States. This paper contributes to a literature
3

documenting a decline in elasticity of housing supply over the past decades (Baum-Snow, 2023a;
Orlando and Redfearn, 2024b), as well as more broadly the literature on infill housing development and housing supply (Dong, 2023; Freemark, 2023; Schuetz, 2020). In this paper, Christian
Redfearn and I classify census blockgroups in those MSAs into four types (”channels”) based on
prior housing unit density, density of employment, and centrality, along lines that coincide with the
structure of the construction industry. We demonstrate that the composition of new units in terms
of what kinds of places they are built in has shifted toward more infill when comparing the 2010s
to the 2000s. We find that the share of net new units built in exurban greenfield settings declined
from around 60 percent of all new housing units in the MSAs studied in the 2000s to 44 percent
in the 2010s, while the share of all new units in already-dense urban environments increased from
14 percent in the 2000s to 24 percent in the 2010s. Further, we observe that the share of new units
built in already-dense portions of the MSAs is greater in MSAs with high housing costs. Given the
higher cost of construction in infill settings and the notion that housing costs might be bounded by
the cost of the marginal housing unit, this shift may be part of the explanation behind the declining
elasticity of housing supply over that same period.
Finally, Chapter 5, ”Online Housing Listings and Preferences,” explores how residential properties are marketed, tracking references to novelties in housing - such as Accessible Dwelling Units
- and to transit infrastructure in Southern California housing listings collected from 2020 to 2024.
I demonstrate that the text sellers and their agents use to promote their properties reflects local
amenities, and that it is responsive to external changes: Properties within close proximity to rail
transit tend to advertise this fact. While condominiums and townhomes make such reference more
often than other property types, this is largely a composition effect due to a greater share of such
properties being located close to public transit rather than suggesting that prospective buyers for
such properties are looking for anything fundamentally different than are buyers for, say, singlefamily homes. Further, listings respond to changes in zoning laws that permit more infill development: Over the course of my period of observation, the share of single-family housing listings that
call attention to accessory dwelling units increased from approximately four percent to over twelve
4

percent. It joins a nascent literature that uses text in online housing listings to track neighborhood
changes (Nilsson and Delmelle, 2022), changes in preferences (J. Lee and Lee, 2023), or adoption
of policies (Blanco and Song, 2024), as well as contributing to a literature exploring the potential
of accessory dwelling units as a form of promoting infill development (Brueckner and Thomaz,
2024; Marantz, Elmendorf, and Kim, 2023a).
The papers in this dissertation use data sources that are openly available to anybody interested:
OpenStreetMap is an open source data source, the Census Bureau’s data products are publicly
available, and online housing listings are - while not strictly open source data - visible to the
public. Wherever possible, I ask questions at a national scale, but based on bottom-up, highly local
analyses, that I then scale up to the national scale.
5

Chapter 2
Transit Station Area Walkability: Identifying Impediments to
Walking Using Scalable, Recomputable Land-Use Measures
Clemens A. Pilgram and Sarah E. West
Abstract
Transit station area land use characteristics can increase or decrease the perceived costs of riding
rail relative to driving or taking other modes. This paper focuses on those characteristics that create
discomfort to riders who are walking between stations and destinations, with the aim of providing
researchers and planners with a tool that can be used to identify pain points in any existing or
potential station areas. We propose and demonstrate a scalable, recomputable method of measuring
pedestrian quality for trips that relies solely on datasets readily available for almost any location
in the United States, and we compare results using data from a global source, OpenStreetMap.
We illustrate our tool in neighborhoods surrounding the Blue Line light rail line in Minneapolis,
Minnesota, calculating the population-weighted distribution of land uses within pathway buffers
of walks from stations to nearby destinations. We focus on land uses that pose a disutility to
pedestrians such as major highways or industrial tracts, and compare disamenity levels across
6

station areas. Despite their simplicity, our measures capture important differences in land-userelated pedestrian experiences and reveal the inadequacy of using circular buffers to designate and
characterize station catchment areas.
7

2.1 Introduction
Researchers and planners have long known that the land use characteristics of transit station areas
affect pedestrian experiences. An established literature finds that the nature and degree of land
use mix affects mode choice, walking, and transit ridership. 1
In addition, research that uses environmental audit methods reveals the importance to potential pedestrians of specific impediments
to walking (e.g. Day, Boarnet, Alfonzo, and Forsyth, 2006). Such methods, however, are laborintensive and difficult to scale, and cannot be easily employed to conduct cross-city comparisons.
In this paper, we propose and demonstrate a scalable, recomputable method for generating descriptive statistics of land uses surrounding stations – or any other destinations where substantial foot
traffic is expected or desirable– that complement more complex measures.
We focus on land uses that create discomfort to riders who are walking between stations and
destinations, with the aim of providing researchers and planners with a tool that can be used to
identify pain points in any existing or potential station areas. Our method relies solely on datasets
readily available for almost any location in the United States, and we compare results using data
from an open-access global source, OpenStreetMap. These data are commonly used by planners,
and our easy-to-scale techniques enable quick identification of specific problem areas that impede
station access and egress, which promotes large-scale comparisons within and across geographic
areas. We illustrate our tool in neighborhoods surrounding the Blue Line light rail line in Minneapolis, Minnesota, calculating the distribution of land uses within pathway buffers of walks
from stations to nearby destinations.
To identify areas associated with disamenities such as industrial areas, major roadways, rail
tracks, or vacant lots, we leverage understanding of relationships between land use and pedestrian experiences known through prior research (Basu, Haque, King, Kamruzzaman, and OviedoTrespalacios, 2022; Guo and Ferreira, 2008; Hartig, Evans, Jamner, Davis, and Garling, ¨ 2003;
LaJeunesse, Ryus, Kumfer, Kothuri, and Nordback, 2021; S. Park, Deakin, and Lee, 2014; Tribby,
1See for example Cervero, 2002; Cervero and Kockelman, 1997; Ding, Cao, and Liu, 2019; Ewing and Cervero,
2001, 2010; Gutierrez, Cardozo, and Garc ´ ´ıa-Palomares, 2011; Jun, Choi, Jeong, Kwon, and Kim, 2015; J. Liu, Xiao,
and Zhou, 2021; Q. Zhang, Moeckel, and Clifton, 2022.
8

Miller, Brown, Werner, and Smith, 2017). We identify geographic locations within 800 meters of
stations (about one half-mile) at which light rail users are more likely to come into conflict with
cars or trains or experience noise, difficult-to-traverse walkways, and other disamenities associated with industrial land, roadways, and rail tracks. We overlay grid cells on station area land use
maps, designate each grid cell according to its land use, and flag those associated with disamenities. Then, we weight each cell by the population of potential transit riders that pass through the
area on the way from their census block to the station. These steps generate population-weighted
pathways that we map along the Minneapolis Blue Line.
Our measures contrast with most traditional land use indices, which do not accurately capture
the most likely routes between station and surrounding blocks, or the potential number of people
taking them. For example, a large industrial site may be near a station, but if few parcels lie beyond
it, or few people live beyond it, then simply calculating its ratio to total land area within a radius of
the station would over-weight the importance of the industrial site for pedestrian experiences. In
addition, our tool uses only land use and street network data, generating measures that complement
more complex assessments that, for example, characterize streetscapes. By doing so, we do not
mean to suggest that more complex measures are less important or less useful, just as we do not
mean to imply that local knowledge should not play a primary role in station area development
planning, or that proportion of land use should be the only measure used by planners to determine
walkability of a station area. On the contrary, we promote our approach as one that can draw
attention to potential problem areas that require further in-depth and local investigation.
To demonstrate the feasibility of our tool with a sample application, we rely primarily on
the Generalized Land Use Survey (”GLUS”) from Minneapolis. Our use of the GLUS enables
us to present highly detailed and complete analysis of station areas in our chosen city. To further demonstrate our approach’s potential, we also characterize station areas using only data from
OpenStreetMap (”OSM”), an open-access, crowd-sourced mapping project that provides data on
9

hundreds of cities worldwide. 2 The measures we derive using OSM are broadly consistent with
those using local land use data, suggesting that researchers can use OSM to conduct broad, global,
cross-city land-use based comparisons.
Because our measures are based on land uses within buffered population-weighted pedestrian
pathways between stations and destinations, our station area land use characterizations differ from
those generated using circular buffers. In nearly all of the station areas in our study area, our pathway land use measures detect more exposure to land uses associated with disamenities, sometimes
substantially so, than suggested by traditional circular buffer measures. Since we identify these
problematic areas at the grid-cell level, we identify hyper-local areas for improvement and funding. And because land use data like ours are widely available, our method - for which we will
make data and computer programs publicly available - can be used to map such areas in nearly
any existing or potential station area and enable planners and cities to identify, avoid, mitigate, or
ameliorate station area characteristics that deter transit use. As such, our method and estimates
can be used to address the challenges posed by how transit users are to cover the distance between
stations and their actual trip origins or destinations (Givoni and Rietveld, 2007; Y. Liu, Yang, Timmermans, and de Vries, 2020; Zellner, Massey, Shiftan, Levine, and Arquero, 2016). We also
expect researchers to find our station area land designations useful when estimating the effects of
new or improved transit on outcomes such as ridership, land use change, and property values.
We proceed as follows. In Section 3.2, we review existing research on the relationship between
the built environment and mode choice, transit use, and walking, and clarify our contribution to this
literature. We describe the study area in Section 2.3, discuss our methods and data in Section 2.4,
and present our findings using local land use data in Section 3.5. In Section 2.6, we recharacterize
station areas using OpenStreetMap data. Finally, in Section 2.7 we discuss the implications and
applications of our findings and conclude.
2Boeing, Higgs, Liu, Giles-Corti, Sallis, Cerin, et al., 2022 demonstrate the general usefulness of OpenStreetMap
for measuring neighborhood-level spatial indicators of urban design and transport features, using the open-access
project to generate indicators for 25 cities in 19 countries.
10

2.2 Literature Review
In this section, we review the literature that relates land use to mode choice, transit use, walkability, and path choice, and establish the ways in which our paper builds on and complements
this research. We begin by examining a well-established literature that focuses on the relationship
between the complexity, composition, or configuration of land use and transit use. We then review
papers that use environmental audit methods to generate detailed local walkability assessments.
Finally, we discuss articles from urban planning and environmental psychology that provide the
basis for using land use data to identify impediments to walking.
A rich literature describes the relationship between land use composition, configuration, and
complexity and mode choice in general, on transit use in specific, on walking propensities, and
on physical health. As explained by Gehrke and Clifton, 2019, composition is ”the number of
land use patches or proportion of each type,” and configuration ”reflects the spatial arrangements,
shape, and dissimilarity of landscape patches” (p. 13). Both composition and configuration can
be more or less ”complex.” Like us, many of these papers use land use data to characterize station
areas (e.g. Cervero, 2002; Cervero and Kockelman, 1997, also see Ewing and Cervero, 2001 and
Ewing and Cervero, 2010 for reviews). This literature hypothesizes that greater degrees of land use
mix around stations induces greater use of transit using entropy or dissimilarity indexes to measure
mixedness.3
These papers typically use data from the census or geographic information systems and find
that land use diversity affects mode choice, including transit ridership. For example, Ding, Cao,
and Liu, 2019 use an entropy index and estimate that built environment characteristics explain
about a third of station boarding variation in the heavy-rail network in Washington D.C.. Similarly,
Gutierrez, Cardozo, and Garc ´ ´ıa-Palomares, 2011 find that greater levels of land use diversity are
associated with greater boarding numbers in Madrid. Jun, Choi, Jeong, Kwon, and Kim, 2015 use
3While entropy indices measure the degree of balance of different land use types in a given area, dissimilarity
indexes measure how well the land uses are mixed up. For example, one could have an entropy index equal to 1 (the
highest possible value) if all land uses are represented equally in an area but have a low dissimilarity index for the
same area because parcels of common land use type are clumped together. The Minneapolis Blue Line station areas
are likely to have moderate levels of entropy, but low levels of dissimilarity (because of clumping).
11

Bhat and Guo, 2007’s balance index measure of land use diversity, where land use is classified
into residential, manufacturing, and office categories, and find that greater levels of mix (balance)
correlate with boardings in Seoul’s subway system in catchment areas most immediate to stations.
Q. Zhang, Moeckel, and Clifton, 2022’s simulation model of pedestrian demand suggests that
denser, more diverse land-use plans in Portland can improve promotion of walking trips. And J.
Liu, Xiao, and Zhou, 2021 find that areas with greater balance in land use increase walking for all
kinds of trips in Xiamen, China, but particularly walking when commuting.4
Areas with greater entropy (balance of land uses) or dissimilarity (spatial mix of uses), however,
may not be of greater value if the balance and mix are being generated by unpleasant land uses.
For example, industrial uses and rights of way do not add amenities that, when mixed well with
residential and commercial uses, make for enjoyable walking commutes to stations. Instead, they
may be replete with obstacles that deter passage. In addition, traditional land use indices generally
do not weight land use by population that passes through the land use on specific pathways to the
station, and therefore do not accurately capture the number of potential routes between station and
surrounding parcels. As Guo and Ferreira, 2008 put it, “Path-based approaches better describe
the actual travel decision of a pedestrian than zone-based methods (p. 462).” A disamenity passed
by many potential transit riders should be weighted more heavily than one passed by few; by
weighting land uses according to the populations that might experience them, along the pathways
that they take, one can generate more accurate measures of pedestrian experiences in station areas.
In contrast to the more traditional land use and mode choice research, another body of literature
uses labor-intensive environmental audit methods to collect data on street and neighborhood characteristics to identify specific neighborhood obstacles and walking disamenities. Boarnet, Forsyth,
Day, and Oakes, 2011 use the Irvine Minnesota Inventory (”IMI”) to predict the effect of the built
4Note that we have taken care here to describe the estimated relationships in this paragraph and elsewhere in this
section as ”correlations” or ”associations,” rather than causal relationships. The studies we summarize here generally
do not focus on establishing causation (though they use varying degrees of caution when describing their findings).
Our paper also leaves causal inference for other research, focusing instead on new measures of the built environment
that can be used in exercises that probe either correlation or causation.
12

environment on physical activity and walking. 5 The inventory took two years to develop, was
tested in 27 different field settings, and includes information on 162 characteristics of the built
environment ranging from streetscape features like the presence of porches or sidewalk cracks,
to intersection characteristics, and parcel land uses (e.g. indicating the presence of a restaurant)
(Boarnet, Day, Alfonzo, Forsyth, and Oakes, 2006). Together this information offers exceptionally detailed data on the Twin Cities, Minnesota, and Boarnet, Forsyth, Day, and Oakes, 2011
demonstrate that it is possible to use such data to predict behavioral outcomes among residents of
those cities. Adkins, Dill, Luhr, and Neal, 2012 find a statistical link between built environment
characteristics, also recorded via an environmental audit, and survey respondents’ perceptions of
walkability in Portland, Oregon. Similarly, S. Park, Deakin, and Lee, 2014 construct a walkability
index using detailed pedestrian survey information from one station area in suburban San Francisco, and find it predicts the likelihood of walking versus taking a car. Finally, Werner, Brown,
and Gallimore, 2010, using data from an audit developed and described by Brown, Werner, Amburgey, and Szalay, 2007, find that light rail use is more likely on more “walkable” blocks, this
time in Salt Lake City, Utah. These studies offer useful levels of descriptive detail for specific
cities but are difficult and costly to scale and apply to new areas.6
From a planning perspective, ensuring quality pedestrian experiences in station areas is of particular importance. Such areas are intended to be ”pedestrian pockets,” a concept central to transit
oriented development (Calthorpe, 1989; Renne and Appleyard, 2019). A related body of research
focuses on observed behaviors on the “last mile” between a transit station and a destination, such as
a home. This literature also finds that passengers’ perception and likelihood of traveling via public
transit is affected by specific characteristics of the local neighborhood’s built environment (Givoni
and Rietveld, 2007; Y. Liu, Yang, Timmermans, and de Vries, 2020; Zellner, Massey, Shiftan,
5Development of the IMI is described in Day, Boarnet, Alfonzo, and Forsyth, 2006 and tested in Boarnet, Day,
Alfonzo, Forsyth, and Oakes, 2006.
6A more computationally technical literature uses street-view images and/or machine learning to describe pedestrian experiences or predict housing market outcomes (Cetintahra and Cubukcu, 2015; Naik, Raskar, and Hidalgo,
2016; Yin, Cheng, Wang, and Shao, 2015; Yin and Wang, 2016; F. Zhang, Zhou, Liu, Liu, Fung, Lin, and Ratti,
2018). While these methods are easier to replicate in new areas, they require techniques that are not typically used by
practitioners.
13

Levine, and Arquero, 2016). Perceptions of safety, which are affected by surrounding land use,
also affect the choice to walk and levels of satisfaction with public transit (Loukaitou-Sideris, 2006;
K. Park, Farb, and Chen, 2021; Venter, 2020).
Findings from both the environmental psychology and urban planning literatures underscore
the importance of station area characteristics for choosing transit and provide the foundation for
our use of land use to identify areas that are likely to impede or make walking unpleasant. In a
landmark study on a group of randomly selected pedestrians, Hartig, Evans, Jamner, Davis, and
Garling, ¨ 2003 find that in general, concrete-abundant urban environments increase stress, while
natural and tree-filled environments reduce it. LaJeunesse, Ryus, Kumfer, Kothuri, and Nordback,
2021 disaggregate urban land uses to understand how city surroundings affect pedestrian stress levels. They find that walking in proximity to major streets and in areas with industrial and mixed use
are associated with higher stress, while traversing residential areas, forests, parks, and university
campuses produces lower stress levels.
Similarly, literature on pedestrian route choice further supports the finding that land use affects
walkers’ choice of path. Guo and Ferreira, 2008 define “pedestrian-friendly” parcels as those with
retail, commerce, and mixed development, and find that paths that pass though areas with such
uses are more likely to be chosen by walkers in Boston. Summarizing the literature on pedestrian
route choice in general, Basu, Haque, King, Kamruzzaman, and Oviedo-Trespalacios, 2022 report
that pedestrians are more likely to choose routes that pass residential and commercial buildings,
and less likely to take paths through areas with industrial uses, vacant land, or traffic. Tribby,
Miller, Brown, Werner, and Smith, 2017, for example, which uses random forest techniques to
find correlates with route choice in Salt Lake City, find that noise and industrial land use deter
pedestrians. They also find that on-street parking increases the chance that a route will be chosen,
attributing the effect to the buffer such parking provides to automobile traffic (Marshall, Garrick,
and Hansen, 2008).
We develop land-used-based measures that are informed by this literature and complement
more complex and labor-intensive methods of characterizing the walkability of transit station areas.
14

We use population-weighted pathways, which pinpoint the routes that pedestrians are more likely
to use, rather than examining land use in circular buffers. Because our measures are easy to scale,
they enable quick identification of specific impediments to transit station access, and facilitate
large-scale comparisons across and within cities.
2.3 Study Area
Our method uses publicly available land use data to develop proxy measures of the obstacles,
disamenities, and problem areas that may impede pedestrian access to stations.7 Using OSM data
to create routes between stations and census blocks destinations within 800 meters (about one halfmile) of a station, we account for the fact that land use measures calculated within a circular buffer
may not accurately capture the experiences of people traveling between stations and neighborhood
places.
We use this method to evaluate areas within surrounding stations of Minneapolis’ Metro Blue
Line Light Rail, which connects Downtown Minneapolis to the Minneapolis-Saint Paul International Airport and the Mall of America via a highway and freight rail corridor along Highway
55 - Hiawatha Avenue. We chose this study area because it is an instance of new rail investment
retrofitted into an existing built-up urban area, and it contains a diverse set of land uses across
the different stations within the corridor. We therefore follow several other papers studying travel
behavior (Cao and Schoner, 2014), labor market accessibility (Fan, Guthrie, and Levinson, 2012),
residential preferences (Cao, 2015), housing price appreciation (C. A. Pilgram and West, 2018)
or propensity of land redevelopment (Agustini and West, 2022) in the same study area. We use
pathways 800m long to be consistent with this and other prior literature that suggests that transit
catchment areas are generally limited to about one half-mile (804.5 meters) from stations (Federal
Transit Administration, 2011). While our findings are specific to the area we study, our method
7Detrimental land uses may contain specific obstacles that directly impede walking such as uneven railroad tracks,
disamenities like loud grain elevators, or more general problem areas such as roadside walkways that are exposed to
strong winds. While the specific meanings of these terms as we imagine them vary to some degree, all are examples
of the kinds of impediments we aim to capture in our land use measures.
15

should be near-universally applicable to any location or pathway distance for which the required
types of input data can be obtained.
As is described in Hurst and West, 2014, the Hiawatha Avenue corridor was initially considered
as an alignment for what is now Interstate 35W, only to remain in limbo for several decades before
the Minnesota State Legislature approved funds that in conjunction with federal funding made the
corridor the location of the region’s first light rail line. Construction on the line began in 2001, and
the line went into service in 2004. We focus on the six station areas in Minneapolis that are outside
of downtown along the Hiawatha corridor: Franklin Avenue, Lake Street, 38th Street, 46th Street,
50th Street - Minnehaha Park, and Veteran’s Administration (VA) Medical Center.
Figures 2.1 and 2.2 show the land uses in 800-meter buffers surrounding these stations. Highway 55 - Hiawatha Avenue itself is a six-lane arterial road. It and other highways are indicated
in black. The light rail line is indicated in brown, as are freight rail lines. Industrial land (in red)
is concentrated just north of the Lake Street station. Single-family residential (bright yellow) and
multi-family (tan) land dominate station areas surrounding 38th and 46th Street stations. Between
the Lake Street and 46th Street stations, the light rail line runs along the western side of Highway
55, running in between rather than serving central portions of neighborhoods. Its stations, located
within a corridor of industrial and highway land uses anchored by Hiawatha Avenue, are therefore located within what Jane Jacobs, 1961 would likely consider a ”border vacuum”. 8 This is
exacerbated by the fact that rail passengers traveling to ultimate destinations east of Hiawatha Avenue must cross the six-lane road at grade level. Such walkers then pass through an industrial and
freight rail corridor approximately 200 meters (one eighth of a mile) wide. South of 46th Street,
the line splits from Hiawatha Avenue, instead running along a smaller parallel road. Recreational
land, in green, is scarce along the northern portion of the line but abundant in the circular buffers
around the southern stations in our study area. The southernmost station area is dominated by the
8So-called ”border vacuums” exist in areas proximate to features acting as borders to pedestrian activities - such as
train tracks or major roads - that are of low urban vitality due to proximity to those borders. Jane Jacobs writes, ”The
root trouble with borders, as city neighbors, is that they are apt to form dead ends for most users of city streets. They
represent, for most people, most of the time, barriers. Consequently, the street that adjoins a border is a terminus of
generalized use.”(Jane Jacobs, 1961, p. 259).
16

institution for which it is named, the VA Medical Center, which is visible in blue, the color for
”institutional” land use.
2.4 Methods and Data
2.4.1 Data Sources
In this section, we describe the data that we use to characterize pathway buffer areas in Minneapolis
light rail station areas. While we use local land use data for this main exercise, later in the paper
we present results based on open-source mapping data for comparison. We rely on four types of
data for our primary analysis:
• Street Networks: We obtain street network information from OpenStreetMap (”OSM”)
via the OSMnx Python package (Boeing, 2017), filtering the network to contain all elements of the network available to pedestrians. 9 OpenStreetMap is an online mapping service generated from volunteer efforts. It is fairly complete, and withstands groundtruthing
audits (Barrington-Leigh and Millard-Ball, 2017; Bright, De Sabbata, Lee, Ganesh, and
Humphreys, 2018). Its near-universal coverage makes it a useful data source for replicable,
re-computable studies where the same analysis is performed repeatedly across a wide range
of different places (Boeing, 2020a, 2021).
• Land Use Data: Land use data are taken from the Generalized Land Use Survey (”GLUS”)
and consist of a Shapefile containing Polygons with information on how each piece of land in
Minneapolis was used in 1990, 1997, 2000, 2005, 2010, 2015, and 2020. We aggregate land
uses to nine categories to obtain consistency across years: Commercial, Institutional, MultiFamily Residential, Recreational, Residential, Highway, Industrial and Railway, Roadway,
9Specifically, we filter using the custom filter ’[”highway”] [”area”!∼”yes”]
[”highway”!∼”abandoned—construction—planned—platform—proposed—raceway—motorway—
motorway link—trunk”] [”service”!∼”private”]’. This filter is more expansive than the OSMnx default for
pedestrian infrastructure, yet excludes several types of roads that are only available to motorists.
17

Figure 2.1: Map of the Study area in 2020, Northern Portion
18

Figure 2.2: Map of the Study area in 2020, Southern Portion
19

and Vacant.10 Consistent with literature reviewed above that relates walk route choice and
pedestrian displeasure to land use, we identify and focus our analysis on four land uses that
likely make walking unpleasant: Highway, Industrial and Railway, Roadway, and Vacant.
While we report shares of the other land uses in our analysis, we remain agnostic about their
effect on walking experiences.
• Census Data: Specifically, we use census block locations and populations. For the neighborhood end of neighborhood-station trips, we rely on census block centroids from the 2010
Census; to weight each block’s neighborhood-station trips, we rely on populations of each
census block as recorded in the 2010 Census.
• Station Locations: For the other end of trips, station locations are taken from a shapefile
provided by the Twin Cities’ regional planning authority (Minnesota Geospatial Commons,
2022).
2.4.2 Methods
We illustrate our method in Figure 2.3 using an example route. We start by using OSMnx to calculate street-network based routes from each light rail station in south Minneapolis to the network
node closest to each census block our study area (first panel of Figure 2.3). Routing is performed in
Python using the ”OSMnx” software package, while all subsequent calculations and spatial transformations are performed in R using the ”sf” software package. Since - particularly in our flat
10We group ”Agricultural” (used almost exclusively for community gardens in our study area), ”Golf Course,”
”Open Water,” ”Open Water Bodies,” ”Park, Recreational, or Preserve,” ”Parks and Recreation Areas,” and ”Water”
into ”Recreational;” ”Airport,” ”Airport or Airstrip,” and ”Airports” into ”Airport;” ”Commercial,” ”Mixed Use Commercial,” ”Mixed Use Commercial and Other,” ”Office,” and ”Retail and Other Commercial” into ”Commercial;” ”Industrial,” ”Industrial Parks not Developed,” ”Industrial and Utility,” ”Industrial or Utility,” ”Mixed Use Industrial,” and
”Public Industrial (1997 only)” into ”Industrial;” ”Institutional” and ”Public Semi-Public” into ”Institutional;” ”Major Four Lane Highways” and ”Major Highway” into ”Highway;” ”Major Railway” and ”Railway” into ”Railway;”
”Mixed Use Residential,” ”Single Family Attached,” ”Single Family Detached,” and ”Single Family Residential” into
”Residential;” ”Multi-Family Residential” and ”Multifamily” into ”Multi-Family Residential;” and ”Public & SemiPublic Vacant,” ”Undeveloped,” and ”Vacant/Agricultural” into ”Vacant.” In addition, we mark all areas that are within
5 meters of the centerline of a road labelled “primary,” “secondary,” “tertiary,” or “trunk” as “Roadway.” Since – with
the exception of the Blue Line light rail – all railways in our study area are part of industrial parcels of land, we
combine Industrial and Railway land uses into one category for purposes of illustrating our method.
20

Figure 2.3: Illustration of Routing and Pathway Buffer Generation
study area - pedestrians typically do not walk faster on one type of road than on another, we determine routing by having OSMnx choose the routes that minimize distance, using only paths that
are available for walking. We then buffer each route by 75 meters to generate ”pathway buffers”
of each walk from a station (second panel of Figure 2.3).
We chose 75 meters as a distance for buffering pathways to capture the furthest distance at
which land uses may influence perceptions of how pleasant or unpleasant that walk might be.
Consider the grain silos near the 38 th Street Station, presented in Figure 2.4: From a distance of
150 meters, the grain silos act as a mere backdrop; at 75 meters, they are far more likely to impact
pedestrians’ experiences.11
Next, we subset to routes of no more than 800 meters network distance, take their ”pathway
buffers,” intersect them with land use information, and calculate the share of the pathway buffers
occupied by each land use (third panel of Figure 2.3).
Finally, we sum areas for each land use across all paths leading to any given station and divide
each land use’s sum by the total amount of land in all pathway buffers for that station; weighting
11We also experimented with using a distance of 25 meters to capture only land uses located immediately along
paths. Results remain essentially the same.
21

(a) Grain Silos on 38th Street and Hiawatha viewed from 150 meters distance.
(b) The same grain silos, viewed from 75 meters distance.
(c) The same grain silos, viewed from up close.
Figure 2.4: A grain silo viewed from three different distances.
22

by the population of that census block per the 2010 US Census. 12 Because in our study area land
use mixes differ substantially depending on whether they lie to the east or to the west of the station,
we calculate shares separately for each side of the corridor.
Figures 2.5 and 2.6 map these population-weighted pathways, along with dotted lines indicating 800-meter circular buffers. s13 The largest number of transit users pass through warmer colored
(red, orange, and yellow) areas when travelling to and from stations. Cooler colors such as blue
and green indicate that a non-zero but smaller number of people experience those areas. Note that
many fewer destinations fall within actual 800-meter network walking distance than within the
circular buffers of the same size, and that these destinations are not located in neat radials.
Finally, to map detrimental land uses we overlay all pathway buffers generated as described
above. By summing the populations of destination census blocks of those pathway buffers, we
obtain a measure of how many people may experience the land use on their way to the station.
2.5 Findings
2.5.1 Differences in Station-to-Block Measures across Station Areas
We begin our analysis by quantifying the spatial configuration and intensity of land uses we associate with obstacles and disamenities (Highway, Industrial and Railway, Roadway, and Vacant)
versus other uses encountered while traversing a route to or from a station and a neighborhood
destination. Panel A of Table 2.1 presents land use shares in the pathway buffers along 800 meterslong station-to-census-block pathways, while Panel B shows the land use shares within a 800 meter
radius around stations. Both panels enumerate the differences in land use mix between and within
station areas that are visible in Figures 2.1 and 2.2. Moving from north to south along the line
12The 2010 US Census records nighttime populations of each census block in the United States. For a measure of
daytime population, one could use blocks-level employment counts from the Census Bureau’s LODES WAC dataset,
as weighting using block populations biases against block-station routes used primarily for work or leisure purposes.
13Note that some areas outside the 800 meter circular station area buffers fall within the 75 meter buffers ”visible”
from pathways. This phenomenon occurs when pathways begin more than 725 meters way from stations, and is
particularly apparent in our maps in Figure 2.5 west of the Franklin Avenue station.
23

Figure 2.5: Northern Stations Population-weighted Pathway Buffers
24

Figure 2.6: Southern Stations Population-weighted Pathway Buffers
25

along the Blue Line takes passengers from neighborhoods that have greater levels of multifamily
housing, commercial areas, and highway lanes, to areas that have more single-family residential
neighborhoods and recreational parkland, with substantial industrial tracts east of the line separating stations from residential areas. The 50th Street - Minnehaha Park station’s proximity to
Minnehaha Park, a large city park with a famous waterfall (Minnehaha Falls), explains its high
share of recreational use. The station at VA Medical Center is unique, as it abuts the federal territory at the southern border of the city of Minneapolis, which is dominated by the “Institutional”
land of the Veterans’ Administration Medical Center and Veterans’ Home.
2.5.2 Station-to-block pathway measures versus circular buffers
Comparing the numbers in Table 2.1, Panels A and B shows the importance of accounting for the
location of trip origin or destination parcels when calculating land use shares and demonstrates the
increased accuracy of using OSM to plot routes rather than relying only on GLUS data to measure
land use areas surrounding a station.14
For example, while land use shares in circular buffers indicate that the largest share of land use
(31.3%) within 800 meters and east of the Lake Street station is industrial and railway, the location
of this land is such that only 5.3% of population-weighted pathway pathway buffers east of the
station are comprised of it. The map in Figure 2.5 makes this clear: The upper right-hand quadrant
of the Lake Street station area has shaded pathways only very close to the station. Table 2.1 also
shows that at the Lake Street station, pathways to census blocks east of the station pass through far
more commercial land (42.1%) than comprises the eastern 800 meter circular buffer (22.8%). On
the other hand, eastbound pathways also pass through more highway (15.4%) than comprises the
eastern circular buffer overall (8.0%).
14Land use shares for the area west of VA Medical Center station do not sum to 1 because we omit ”airport”
land from these tables and the Minneapolis-Saint Paul International Airport falls within this station’s surroundings,
comprising 1.6% of the total land within 800 meters and west of the station. None of the other stations have airport
land near them.
26

Table 2.1: Pathway Buffers versus 800 Meter Circular Buffer Land Use Shares (%), year 2020
27

The VA Medical Center Station area land use shares calculated using the two measures differ
even more markedly. As shown in Figure 2.2, a good deal of recreational land lies east of the station
alongside the Mississippi River; we find that in 2020 such land makes up 63.5% of the 800 meter
buffer east of the station. But, as shown in Figure 2.6, there are no inhabited parcels that lie past
that parkland, so the eastern circular buffer land use share wildly overstates the degree of exposure
to such land that potential light rail users can experience while walking to or from the station—the
eastern pathway buffer share of recreational land is nearly zero (0.2%) using our measure. Instead,
walking pathways pass through the institutional land of the Veteran’s Administration Hospital and
Home, and residential parcels.
One type of land use in our study area disappears entirely when applying pathway buffer rather
than the zone-based buffer approach: The Minneapolis-Saint Paul International Airport lies south
of the VA Medical Center station area, and “airport” land comprises 1.6% of the total within a 800
meters of the station. This is the only station with airport land in is vicinity; for this reason we omit
“airport” land use from our tables. Notably, because there are no walkable origins or destinations
within 800 meters of the VA Medical Center station and in or past the airport parcels, airport is not
among the land use types detected using our pathway buffer measure, which again demonstrates
the advantage of our measure over traditional circular buffers.
In general, comparison of Panel A and B in Table 2.1 shows that in most station areas, pathway
buffer land use shares along the Blue Line in Minneapolis measure substantially lower degrees of
pedestrian exposure to parcels with single-family homes (”Residential”), on both east and west
sides of the light-rail line, compared to measures of land use using 800 meter circular buffers.
The drop in residential shares is particularly large at the 38 th Street and 46th Street stations, and
dramatically affects the overall share of disamenity areas at those stations. Station-to-census block
pathway measures generally detect more exposure to commercial parcels, to Highway 55 - Hiawatha Avenue, and to industrial and railways. Exposure to industrial parcels, recreational land,
multifamily housing, roadways, and vacant land is generally similar across the two measures.
28

For ten of the twelve sides (east and west) of our six station areas, pathway land use measures
detect more exposure, sometimes substantially, to land uses associated with unpleasurable walking experiences: Transit users walking to Blue Line stations experience more disamenities than
suggested by approaches using circular buffers.
2.5.3 Identification of detrimental parcels in station areas
Our method enables researchers and planners to scan maps across and within station areas to identify potential parcels for remediation and improvement, and to prioritize based on their visibility.
The maps in Figures 2.7 and 2.8 isolate and aggregate the four land uses associated with disamenities and show the number of potential transit users who would pass the detrimental areas on their
way to stations from the centroids of their census blocks. Looking across station areas, the maps
indicate that the largest populations of potential transit users experience land uses associated with
disamenities at the northern stations at Franklin, Lake, and 38th Streets. While all station areas have
areas with disamenities, the southern-most stations have relatively small populations that may experience them. Our results therefore would direct planners to focus more resources on the more
visible areas at northern stations. It is also interesting to note that the southern stations’ detrimental areas are relatively linear, concentrated along highways, railways, and roadways, while those
at northern stations are spread out more broadly in surrounding neighborhoods, because of both
population and land use geographies. Neighborhoods at the northern stations may therefore offer
more opportunities for improvement if rail and road are less transmutable than other land uses.
Researchers and planners interested in identifying the most visible detrimental land uses within
a station area can also use the population weighted pathway buffer maps in Figures 2.7 and 2.8, in
ways that are more informative than the circular buffer land use maps in Figure 2.5. For example,
Figure 2.5 shows that at the Franklin station, industrial and rail land uses concentrate along the
diagonal corridor in which the light rail locates. But pathway analysis in Figure 2.7 redirects us
to focus on potential improvements along Franklin Avenue and Cedar Avenues, which intersect
at a perpendicular just west of the station. Rather than suggesting we direct attention to the light
29

Figure 2.7: Northern Station Population-weighted Pathway-buffered Detrimental Land Uses
30

Figure 2.8: Southern Station Population-weighted Pathway-buffered Detrimental Land Uses
31

rail corridor itself, the map of Franklin station pulls us east along Franklin Avenue, and northeast
into the Cedar-Riverside areas, both vibrant commercial areas whose links to the Franklin Station
are tenuous and punctuated by disamenities. The Lake Street station’s most visible disamenities
are more concentrated in the immediate area surrounding the boarding platform, but also extend in
both directions down Lake Street, a corridor whose commercial districts are interrupted by large
roadways and vacant land.
The 38th Street station map in Figure 2.7 shows even greater concentration of disamenities in
the immediate station area, but points planners to the immediate east and southeast of the platform,
where a car-oriented intersection of highway and roadway make passage particularly difficult, and
where the railroad and grain elevators shown in pictures in section 2.4.2 create unpleasant walking
for pedestrians coming from residential areas to the east. Similar kinds of analysis can be applied
to each of the station areas in the panel. By contrast, disamenities surrounding the 50 th Street -
Minnehaha Park, and VA Medical Center stations displayed in Figure 2.8 appear to be limited to
highways and roadways.
In these ways, our scalable and easy to reproduce methods provide a first pass at identifying
station area parcels, which, by virtue of their disamenities, may create obstacles to station area
access to residents in surrounding neighborhoods.
2.6 OpenStreetMap as a data source for land uses
So far, this paper has presented land use shares generated from a local data source - the GLUS,
a dataset specific to the Minneapolis-Saint Paul Metropolitan Area. However, the exact same
pathway buffer-based approach to evaluating land uses within pathways can also rely on other
sources, such as OpenStreetMap (”OSM”) - the open source global volunteer mapping project. An
advantage of OSM over other sources its global availability and relatively consistent recording:
While sources such as GLUS - recorded by local authorities - are likely more precise and more
complete than OSM, their level of detail and types of land uses recorded vary considerably from
32

city to city. By contrast, OSM attempts to record land uses consistently across locations and does
so well enough to serve as a basemap for most uses (Barrington-Leigh and Millard-Ball, 2017;
Bright, De Sabbata, Lee, Ganesh, and Humphreys, 2018; Brovelli and Zamboni, 2018; Zhou,
Wang, and Liu, 2022). As such, OSM may be more suitable for city-to-city comparisons than
reliance on several local data sources.
To generate a land use dataset comparable with GLUS - in which any given place is assigned
one land use (that is, there are no overlapping land uses), we download all OSM features with
tags we consider synonymous with various land uses.15 Further, we buffer linear road and railway
features to approximate areas perceived as railways or roadways.16 To resolve overlaps, we assume
a hierarchy of land uses, from least to most dominant in perception: Recreational land is the least
dominant, followed by Multifamily Housing, Institutional, Commercial/Office, Industrial, Airport,
Roadway, Highway and Railway. As all railways in our study area besides the Blue Line are
industrial, we group Industrial and Railway when reporting area shares.
Table 2.2 shows how OSM land-use-based pathway buffer measures compare to those based on
land uses from the Minneapolis-Saint Paul area’s GLUS dataset. Because not all areas are tagged
as any particular use in OSM, substantial areas around stations - mostly single family residential -
have no recorded land use. This is reflected in the generally lower shares of all land uses in Panel
B. Overall, however, we find that OSM is a decent substitute for GLUS. For example, relative
shares of disamenity-associated land uses across and within station areas measured using OSM are
15We define different land uses as follows: Any area with a value for the variable aeroway is considered part of
an Airport; any area with a value of ”commercial” or ”retail” for the variable landuse or a value of ”commercial”
for the variable building or any value for the variable office is considered ”Commercial/Office”; any area with a
value of ”industrial” for the variables landuse or building, ”substation” for the variable power, or ”wastewater plant”
for the variable man made is considered ”Industrial”; any area with the values ”college”, ”hospital”, ”library”,
”place of worship”, ”school”, ”social facility”, or ”university” for the variable amenity or ”church” or ”public” for
the variable building or the value ”museum” for the variable tourism is considered ”Institutional”; any area with the
value ”apartments” for the variable building is considered ”Multifamily”; any area with the value ”railway” for the
variable landuse or the value ”yard” for the variable service is considered ”Railway”; any area with the values ”cemetery” or ”recreation ground” for the variable landuse, the values ”golf course” or ”park” for the variable leisure, the
values ”beach”, ”water” or ”wood” for the variable natural, the value ”attraction” for the variable tourism, or the value
”river” for the variable water is considered ”Recreational”; and any area with the value ”brownfield” or ”construction”
for the variable landuse or the value ”construction” for the variable building is considered ”Vacant”.
16We consider the 10 meter buffer around linear railway features as ”Railway”, a twelve meter buffer surrounding
centerlines of roadways tagged ’trunk’ or ’motorway’ to be ”Highway”, and a six meter buffer surrounding all other
road centerlines tagged ’primary’, ’secondary’, or ’tertiary’ to be ”Roadway” type land use.
33

fairly consistent with those measured using the GLUS (compare the ”Sum of Disamenity Shares”
rows in Panels A and B in Table 2.2). This is good news for transport and land use researchers and
planners, as OSM offers an attractive, inexpensive, consistent source of data that can be used when
performing comparisons across and within different metropolitan areas worldwide (Boeing, Higgs,
Liu, Giles-Corti, Sallis, Cerin, et al., 2022). At the same time we recommend that researchers use
local datasets such as GLUS wherever available when studying more local scale questions.
2.7 Discussion and Conclusion
We use widely available land use data and walking pathway buffers mapped using the publicly
crowd-sourced OpenStreetMap (OSM) to derive scalable, easy-to-reproduce measures of station
area land uses associated with walking disamenities. To illustrate the potential for these kinds of
scalable measures, we simply report proportions of land uses around stations. Our approach can
also be used as a first step toward derivation of more complex measures of station area disamenities
or specific thresholds at which an area might be flagged by planners as ”unwalkable.” 17 Even
our simple measures, however, capture the realities of pedestrian experience near stations more
accurately than circular buffers, as they weight parcels according to the degree to which they are
potentially experienced by the populations surrounding the station.
The literatures on mode choice and on transit ridership have indicated that the built environment
surrounding stations and transit stops affects ridership. For the particular transit line studied in this
paper, other researchers have previously noted that the presence of industrial facilities in station
areas adversely affects walkers’ experiences and may act as a deterrent to potential riders (Cao
and Schoner, 2014). Our pathway buffer measure of land use complements existing measures, and
could add precision to the estimation and forecasting of boardings such as that performed by Jun,
Choi, Jeong, Kwon, and Kim, 2015 or Gutierrez, Cardozo, and Garc ´ ´ıa-Palomares, 2011.
17For instance, our method could be combined in tandem with a walkability audit aimed to determine the severity
of disamenity posed by different land uses.
34

Table 2.2: GLUS versus OSM Land Use Measures for Blue Line Station Pathway Buffers, 800
Meters Distance
35

Indeed, our measures of station area land use could improve estimation of any relationship
between a transit line and its surroundings. For example, our approach captures characteristics of
pathway buffers along the routes between stations and destinations in catchment areas that influence buyers’ willingness to pay premiums for homes nearby. Papers that estimate the effect of
new transit lines on home values that find evidence of heterogeneous effects across station areas
(e.g. Mulley, Tsai, and Ma, 2018; Yang, Chu, Gou, Yang, Lu, and Huang, 2020) may very well
find that our measures explain a good deal of this heterogeneity — the more unpleasant land use
is along routes from a station to homes, the less we should expect home values to appreciate upon
introduction of a transit station. Our findings may therefore explain why previous research found
that the Minneapolis Blue Line has had no measurable long term effect on home values (C. A. Pilgram and West, 2018), and our land use measures may be one of the omitted variables explaining
instability of parameters recovered by hedonic specifications (Redfearn, 2009). Finally, our land
use and pathway based measures could be used to update the effects of Twin Cities light rail on
transit ridership (Cao and Schoner, 2014).
In addition, our method complements those used to generate walkability indices such as Walk
Score or the National Walkability Index (developed by the U.S. Environmental Protection Agency)
that characterize areas in terms of their access to opportunities (Thomas and Reyes, 2021). These
indices also permit cross-sectional comparisons between large areas with relatively little labor
effort. But unlike our measures, these measures are not pathway- or land-use based.
Our method may also aid practitioners in identifying heavily trafficked corridors within transit catchment areas for the sake of pedestrian improvements – projects for which, at least in the
United States context, funding already exists: The Federal Transit Administration (FTA) considers all pedestrian improvements within one-half mile and all bicycle improvements within three
miles of a public transit stop or station have a “de facto physical and functional relationship to
public transportation” and therefore eligible for funding (Federal Transit Administration, 2011, p.
52046).18 Such improvements include those that mitigate the negative experience associated with
18In addition, improvements beyond these distances may be eligible for funding if applicants demonstrate that “the
improvement is within the distance that people will travel by foot or by bicycle to use a particular stop or station.”
36

detrimental land uses such as sidewalk maintenance, road ”diet” reconstruction, or improvements
in lighting. They may also involve broader rezoning and permitting efforts that induce transitoriented land use change in station areas.
Our method, of course, is not perfect: First, it does not consider all forms of obstacles that may
be detrimental to the pedestrian experience. For example, we are unable to account for sidewalk
presence or condition, as such characteristics are not recorded in local sources like the GLUS and
are inconsistently tagged in OpenStreetMap.19 Similarly, since our method focuses on generating
scalable land-use based measures of disamenities, we do not address issues such as block lengths
or monotony of land use along routes that may also affect pedestrian comfort, as doing so would
require adding layers of parcel-data. Second, land uses within the pathway buffer may not actually
be visible from the route - or may be intentionally concealed to improve the pedestrian experience.
For example, walking paths to the East of 50th Street Station - while immediately next to Highway
55 - are separated from the highway by hedgerows and sound barrier walls. While this specific
barrier is in fact indicated on OpenStreetMap, such interventions are not consistently recorded, and
taking them into account in generating viewsheds would require a far more complex computational
approach.
Ultimately, however, our measure of land use within population-weighted pathway buffers describes land uses surrounding the walks to stations in a manner that more closely resembles the
pedestrian experience than a circular buffer-based measure, while remaining scalable and computationally simple. As such, it complements existing ways of describing pedestrian quality such as
environmental audit measures. Indeed, nothing about our method requires the destination to be a
station; it could equally be applied for other destinations where substantial foot traffic is expected
or desirable – such as schools or grocery stores – and can accommodate any weighting scheme
to reflect different routes’ potential traffic volumes. It is recomputable for any location for census, land use, and OpenStreetMap data are available. While we attach the label ”disamenity” to
For more information on funding opportunities for pedestrian and bicycle improvements in transit station areas, see
Federal Transit Administration, 2017.
19For example, OSM records only about 42% of residential streets in Minneapolis as having sidewalks even though
sidewalk presence is virtually universal along those streets.
37

specific land uses, our method can accommodate whatever land-use mix desirability parameters
a user prefers, enabling them to incorporate community-specific preferences about the pedestrian
experience. And, our method draws attention to specific corridors traversed by potential station-todestination paths, properly weighting and highlighting specific parcels with ”ugly” land uses that
fall along those corridors. Finally, we find that land use measures derived using OpenStreetMap
are broadly consistent with those using local land use survey data, suggesting that researchers can
use OSM to conduct broad, global, cross-city land-use-based comparisons.
38

Chapter 3
Pathway buffer measures of pedestrian quality for the entire
United States
Abstract
I offer a scalable pathway buffer-based tool for assessing how metropolitan areas are faring at
making access to public transit or to elementary schools possible or pleasant on foot, and for
identifying highly visible disamenity land uses such as highways or industrial land that impede
access. I thus complement existing conceptions of pedestrian quality such as audit-based methods.
Using this tool, I estimate disamenity land use shares encountered in each metropolitan census
block in the United States when accessing, bus stops, rail stations, and schools. I lean on associations from prior research on pedestrian route choice, and use land use and street network data
from OpenStreetMap. In effect, I ask two questions: “Are there stops (schools) within half a mile
of walking distance?,” and “If it is possible to walk to a station/bus stop/school, what share of land
uses along the way are likely to cause discomfort?”. Using this framework, I demonstrate the existence of racial inequities in the ability to access transit on foot, and in the quality of the pedestrian
environment along the way. I show that associations between racial and socioeconomic variables
are not the same for the two questions: While wealthier, whiter households are less likely to live
within walking distance of transit, wealthier or whiter households within walking distance of bus
39

stops, rail stations, or schools experience fewer disamenity land uses along their ways, constituting
an environmental justice problem at the national scale.
40

3.1 Introduction
A modal shift toward walking, cycling, and micromobility would be desirable for achieving both
environmental goals - such as achieving cuts in carbon emissions - and public health goals - such
as reducing the prevalence of obesity. The benefits of pedestrian-friendly environments are well
documented by a sprawling public health literature. More walkable environments and greater use
of active transportation are associated with lower emissions production and superior health outcomes (Frank, Sallis, Conway, Chapman, Saelens, and Bachman, 2006). Further, more pedestrianfriendly places may improve mental health outcomes, including those of adolescents (Buttazzoni
and Minaker, 2022). However, such a modal shift is hampered by that the built environment as it
exists can be hostile for active transportation: Insufficiently dense urban form may result in destinations and public transit stops simply being too far away to realistically access on foot. Further,
unpleasant features such as highways or industrial parcels may stand in the way (Millard-Ball,
Silverstein, Kapshikar, Stevenson, and Barrington-Leigh, 2024).
Questions such as ”Who does public transit serve?”, ”Who should it serve?”, and ”Where is it
lacking?” are commonly addressed in a literature studying the characteristics of who uses public
transit (Garrett and Taylor, 1999), the spatial coverage coverage of public transit service areas
(Karner, 2018; Welch and Mishra, 2013), or what transit is available for different demographic
groups (Griffin and Sener, 2016). Other writers coin terms such as ”transit desert”, analogous to
the term ”food desert” - that is, an area lacking in grocery stores or other food outlets (Jomehpour
Chahar Aman and Smith-Colin, 2020). What is undeniable is that both the mode share of public
transit and the absolute number of trips taken on public transit have dramatically decreased since
their peak over half a century ago (Garrett and Taylor, 1999).
To date, few academic studies explicitly study what built environments would-be transit riders
must traverse to access transit at scales beyond small study areas. While evaluating walkability at
such scales has recently become possible (Aparicio, Arsenio, Santos, and Henriques, 2024), to my
knowledge, none delve into the first/last mile problem, or into racial or socioeconomic inequities
in the walking experience. This is not for lack of attention to the first/last mile problem: Public
41

transit agencies expend significant amounts of attention and resources planning around this issue,
compiling plans for first/last mile trip portions that often explicitly discuss equity issues (Mohiuddin, 2021). Furthermore, under the banner of ”transit oriented development”, cities rezone station
areas with the expressed intent of increasing population density within station areas, and of offering disadvantaged and transit-dependent populations housing options close to transit (Dawkins and
Moeckel, 2016; Zhu, Burinskiy, De la Roca, Green, and Boarnet, 2021). At the same time, there
are fears that public transit may also act as an amenity that can drive gentrification and displacement, though evidence for this remains limited (Padeiro, Louro, and da Costa, 2019).
Schools - also a common topic for pedestrian planning - are a secondary object of this study.
Substantial efforts - by practitioners and by academics - exist to better understand what makes for
a pleasant and safe walk to school. It is of special importance since such trips may form children’s
perception of urban environments at a particularly impressionable stage, of greater importance for
habit formation. As such, there are policies and projects around ensuring safe routes to school,
including physical improvement projects. 1 However, as is the case with public transit use, the
share of students walking or biking to school has steadily declined over the past half century
(McDonald, 2007; Omura, Hyde, Watson, Sliwa, Fulton, and Carlson, 2019), to the point where a
majority of school children in the United States are dropped off at school via private automobiles
(Van Dam, 2024).
In this study, I will employ the parsimonious, pathway buffer-based geospatial simulation approach developed in C. Pilgram and West, 2023 - or in Chapter 2 of this dissertation - to walks to
bus and to rail transit, as well as to schools. This study will demonstrate not only that there are
broad population-level inequities in access to those amenities, but also in the quality of the walks,
measured at a hyper-local level. Further, it will demonstrate that stark differences exist between
premium transit - such as rail - and basic public transit service such as buses, in both what populations are served and in quality of access: While 54.5% of the population in the MSAs studied lives
within a half mile walk of their nearest bus stop, this is the case for over two thirds of Nonhispanic
1For examples, see https://www.transportation.gov/mission/health/Safe-Routes-to-School-Programs,
https://ladotlivablestreets.org/programs/safe-routes-to-school, and and https://solanosr2s.ca.gov/safety-projects/
42

Black or Latino people in those areas, but only for 43.2% of Nonhispanic Whites. At the same
time, over five percent of land use within the buffer around those walks for Nonhispanic Black
populations consists of unpleasant land uses such as industrial or highways, while Nonhispanic
White populations within a half mile walk of their bus stop experience approximately half as much
of those likely unpleasant land uses. Similar associations and magnitudes exist for walks to elementary schools, while walks to rail transit tend to include far more unpleasant land uses within
the pathway buffers. Further, these associations observed from descriptive statistics hold up ceteris
paribus in a multivariate regression framework when tested jointly. As such, this paper contributes
to a novel field of quantitative environmental justice studies.
This paper proceeds as follows: Section 3.2 reviews the various related literatures on defining
and measuring walkability, on the first-last mile problem and on equity in public transportation,
and on walking to schools, and outlines the contributions this particular study makes. Section 3.3
discusses the data employed in this study and how they are processed, while Section 3.4 describes
the analyses applied to said data. Section 3.5 presents findings, while Section 4.6 discusses them
and concludes.
3.2 Literature Review
3.2.1 Assessing Walkability
It has long been known that relationships exist between land use and walking behavior, in excess
of self-selection into or out of walking (Schoner and Cao, 2014). In light of such associations,
assessing whether or not a particular place is walkable based on its built environment surroundings
is a common endeavor; yet, there is little to no agreement over what is the correct way to do
so, or what aspects of walkability should be explicitly considered. 2 Walkability indices such as
2For example, aspects of pedestrian comfort that one could consider but that are not explicitly discussed by the
literatures on measuring walkability include thermal comfort, or public safety. To the extent such topics are discussed,
it is usually done implicitly rather than explicitly.
43

Walkscore measure availability of destinations such as retail, restaurants, or transit stops. 3 Such
indices have become popular for cross-sectionally comparing locations in terms of their suitability
for pedestrian lifestyles, and have become widely adopted in the context of promoting real estate,
and have also been adopted for some academic studies (Hall and Ram, 2018; WalkScore.com,
2020). However, those scores – which intend to reflect likelihood of pedestrian activity – differ
from literature that explicitly focuses on actual pleasantness of walking in a particular area or on a
particular path.
As of today, the literature remains divided into at least three approaches for how to assess
whether whether a particular area or route is friendly to pedestrians: One stream of literature –
including the Irvine-Minnesota-Index of Walkability and its many offshoots – employs environmental audit methods, in which research assistants or volunteers score an area in terms of several
variables that relate to pedestrian comfort (Boarnet, Day, Alfonzo, Forsyth, and Oakes, 2006;
Boarnet, Forsyth, Day, and Oakes, 2011; Brown, Werner, Amburgey, and Szalay, 2007). This
literature is largely qualitative and at a relatively local scale due to the amount of labor required.
Another, more recent set of studies employ computer vision methods to score locations (Cetintahra
and Cubukcu, 2015; Ki and Chen, 2023; Naik, Raskar, and Hidalgo, 2016; Yin, Cheng, Wang, and
Shao, 2015; Yin and Wang, 2016; F. Zhang, Zhou, Liu, Liu, Fung, Lin, and Ratti, 2018). Given
enough computing power, those methods can become scalable, but often do not offer intuitive explanations for why locations receive particular scores.4 More recently, the literature has moved in
the direction of using open source data and broader study areas, often by relying on sources such
as OpenStreetMap (”OSM”) (Boeing, Higgs, Liu, Giles-Corti, Sallis, Cerin, et al., 2022; S. Liu,
Higgs, Arundel, Boeing, Cerdera, Moctezuma, et al., 2022). Finally, a nascent literature on urban
vitality attempts to operationalize ideas put forth by Jane Jacobs in the 1960s (Garau, Annunziata,
and Yamu, 2024; Gross, 2024; Jacobs, 1961; Sung, Lee, and Cheon, 2015).
3Open-source software such as Pandana allows for relatively easy generation of such availability/accessibility
surfaces, again emphasizing the number of certain kinds of destinations that are within reach (Foti, Davis, Fernandez,
and Maurer, 2023)
4For a more in-depth review of the pedestrian quality literature, consult C. Pilgram and West, 2023 or Chapter 2 of
this dissertation.
44

C. Pilgram and West, 2023 complements the existing literature by presenting a scalable, recomputable method of estimating land use exposure that is more specific to particular paths: Taking
likely network-based pedestrian paths from origins to destinations, observing what land uses fall
within 75 meters of those paths, and calculating what share of land uses within the buffer are
of types that have been associated with displeasure to pedestrians based on prior research (Basu,
Haque, King, Kamruzzaman, and Oviedo-Trespalacios, 2022; Guo and Ferreira, 2008; MillardBall, Silverstein, Kapshikar, Stevenson, and Barrington-Leigh, 2024; Tribby, Miller, Brown, Werner,
and Smith, 2017).
3.2.2 Equity of Public Transit Systems
Studying equity considerations in public transit is nothing new: In 1999, Garrett and Taylor, 1999
observed that transit riders in the United States were disproportionately low income, and disproportionately Hispanic or African American. Much of the subsequent literature on equity in public
transit focuses on the service area coverage of transit systems, asking what share of the population - or of any specific sub-population - is covered by public transit (Karner, 2018; Welch and
Mishra, 2013). In this, a common way of measuring transit equity is to calculate scores akin to
Gini coefficients, evaluating the difference between the distribution of populations (or jobs) and
that of transit coverage (Delbosc and Currie, 2011), allowing for direct comparisons between different places or forms of coverage. The literature has also developed terminology akin to that of
the ”food desert” literature: Jomehpour Chahar Aman and Smith-Colin, 2020 combine zone-based
measures of transit dependency with several measures of transit availability to draw attention to
what they call ”Transit Deserts” - that is, areas where public transit is both in high demand and
lacking in service quality.
Another common lens for studying equity in public transit is to study equity in access - that is,
studying how access to opportunities varies across space and socioeconomic dimensions (Wachs
and Kumagai, 1973). Extensions of this work introduce the idea of a spatial mismatch between
45

workers and jobs. Some studies study how access to opportunities varies across space for a particular mode (Jomehpour Chahar Aman and Smith-Colin, 2020; Karner, 2018), across different mode
options (Boarnet, Giuliano, Hou, and Shin, 2017; Cui and Levinson, 2020), or across different
demographics such as different income groups (Boarnet, Giuliano, Hou, and Shin, 2017; Karner,
2018) or racial groups.
3.2.3 The First/Last Mile Problem
There is also some prior work specifically on the ”first/last mile problem”, the question of how
riders travel to or from public transit stops at either end of their trips. For instance, Boarnet,
Giuliano, Hou, and Shin, 2017 document that within a set permissible travel time, public transit
generally affords access to far fewer jobs than private automobiles, but also that the number of
jobs accessible within a set time via public transit depends on what mode is used for the first/last
mile. Further, access may also vary within the same built environments: Meng, Koh, and Wong,
2016 document that elderly people are willing to walk or bike shorter distances to stops than the
non-elderly.
People are willing to walk different distances depending on land use around stations, with
pedestrian-friendly environments such as those created by transit-oriented development increasing
riders’ willingness to take public transit (K. Park, Ewing, Scheer, and Tian, 2018), and increasing
the distances they are willing to walk to access it (Jiang, Christopher Zegras, and Mehndiratta,
2012; O’Sullivan and Morrall, 1996; J. Wang and Cao, 2017). Similarly, public safety concerns
- both about traffic safety and about safety from crime - influence willingness to walk to transit
stations (S. Kim, Ulfarsson, and Todd Hennessy, 2007; K. Park, Farb, and Chen, 2021).
3.2.4 Walking to School
A body of research evaluates what impacts children’s perceptions of safety on their walks to school,
finding that children are perfectly capable of perceiving and identifying factors that influence their
safety. Besides social factors, these include land uses, which can impact perceived safety in both
46

positive and negative manners. (Banerjee, Uhm, and Bahl, 2014; Eisenlohr, Jamme, Bahl, and
Banerjee, 2023; Webb Jamme, Bahl, and Banerjee, 2018).
Schools often have catchment areas - that is, areas in which resident children are assigned
to attend a particular school. Ikeda, Mavoa, Cavadino, Carroll, Hinckson, Witten, and Smith,
2020 introduce the idea of ”pedsheds” - a measurement comparing the size of buffered pedestrian
street networks surrounding schools to an overall study area, with the goal of measuring built
environment variables for areas relevant to each school. However, students still often live too far
away from schools for walking to be feasible, or face other land use barriers that prevent walking
to school (Omura, Hyde, Watson, Sliwa, Fulton, and Carlson, 2019).
3.2.5 Contribution of this paper
This paper presents a computational approach for measuring both the ability to access to common
errands such as those discussed in the previous sections – accessing food outlets or schools – on
foot, as well as the the presence of obstacles along such routes at the same time, in a manner that
is replicable and recomputable by not involving any location-specific or proprietary data sources.
Using this approach, I present and discuss high level metropolitan-area level findings produced
by this approach, and explores associations with socioeconomic variables such as race and income
to expose inequities in the pedestrian experience. Unlike prior studies that are almost inherently at
local or at best at citywide scales, I study the built environment at far broader a geographic scale.
While I make several assumptions about what land uses are pleasant/unpleasant, these are taken
from the environmental psychology literature.
3.3 Data Sources
I do so using land use data from OpenStreetMap (”OSM”), store locations from OSM and destinations / population data from the Census. As such, the method is replicable / recomputable for
almost any study area; I demonstrate it for a wide variety of MSAs in the United States. While the
47

data sources employed in this paper are available for any location in the United States, I limit my
study to the 200 largest metropolitan areas in terms of population.
3.3.1 US Census
The United States Census Bureau reports information on the United States’ population, counting
the number of residents every ten years via the Decennial Census in addition to providing several
other, higher frequency data products. As the ”trip origins” in this study, I rely upon the centroids
of census blocks - the most granular spatial unit at which data from the 2020 Decennial Census
are reported. I choose census blocks as they are a useful level of aggregation; maintaining a
high of spatial resolution while dramatically reducing computing time versus even more granular
alternatives such as tax parcels. 5 Further, census blocks allow for the use of US Census Bureau
data such as nighttime populations (from the same Decennial Census), or an estimate of workday
daytime populations from LODES WAC.6
The Census reports census block level nighttime populations - that is, residents of any census
block - by racial group, allowing for highly spatially granular analyses. For some other demographic variables that were not collected as part of the 2020 Decennial Census, I lean on the
Census Bureau’s American Community Survey (”ACS”). Due to limitations in the ACS’s spatial
granularity, those variables are only reported at the census block level.
In addition to census blocks’ centroids and populations, I also lean on the Census for Metropolitan Statistical Area (”MSA”) boundaries.
3.3.2 Public Transit Locations: Transitland
The locations of public transit stops - for both bus stops and rail stations - are taken from Transitland. Transitland is a web service that aggregates General Transit Feed Specification (”GTFS”)
5An additional reason not to use tax parcels is that their use would require collecting parcels data from the tax
assessor’s office for each county in the study areas for this study. Given the goal of performing a near-nationwide
study, this would not be a feasible effort.
6
I access census geographies and populations via the ”tigris” and ”tidycensus” R packages (Walker, Herman, and
Eberwein, 2023; Walker and Rudis, 2023)
48

feeds from transit operators around the world (Internline.io, 2024). 7
I query this service for all
transit stops that fall within the boundaries of the metropolitan areas, querying separately for bus
stops and for rail stations.
While Transitland makes efforts to include schedules for every transit operator on the planet,
there are some operators for whom no GTFS feeds exist, or for which GTFS feeds are not queryable
by location. These are primarily smaller, bus-only transit operators, located in MSAs with less
transit service. 8 Nonetheless, Transitland remains the most comprehensive single data source for
transit coverage - any more comprehensive measure of service would require mixing and matching
information from more sources than would be feasible for the geographical scope of this study.
3.3.3 Schools
For my second type of ”trip destination”, I obtain the locations of public elementary schools from
the National Center for Education Statistics’ (”NCES”) Common Core of Data (”CCD”) nonfiscal
data files (National Center for Education Statistics (NCES), 2023). The NCES CCD reports names,
addresses, and school characteristics such as the grades offered or the share of students eligible for
free lunch - a proxy for poverty - for each public school in the United States. 9
From the NCES CCD’s school directory table, I extract the addresses of all schools that are
labeled as elemetary schools or otherwise offering first through fourth grade instruction. I translate those human-readable addresses into lat/lon coordinates using the Pelias geocoding software
(Pelias, 2022).
7GTFS feeds are an open standard data format developed by Google for reporting transit schedules, split into
separate files for stops (that is, locations from which transit can be accessed), routes (scheduled services that serve
those stops), and departures (specific times at which routes serve any given stop) (General Transit Feed Specification,
2024; Google Transit, 2022). Transitland collects static GTFS feeds - that is, transit service as scheduled, as opposed
to the realtime feeds also provided by some transit operators (Internline.io, 2024)
8An example for a metropolitan area with no GTFS feeds on Transitland that are queryable by loacation is the
Corpus Christi, Texas MSA: While that city has a local bus operator, it does not provide GTFS feeds describing its
service. As a result, querying Transitland for bus stops in that region returns only the stops of long distance services
such as Greyhound. Future revisions of this work will make efforts to include additional operators such as this one.
9While similar data are available for private schools, I omit those from this study.
49

Being a U.S. government dataset, this data source is of course only available for the United
States - though equivalent data on school locations may be available for other countries. 10
3.3.4 OpenStreetMap
Both the street networks along which paths are generated and the land uses along those pathways
are taken from OpenStreetMap - a volunteer-generated source of geoinformation for the entire
planet. (DESCRIBE OPENSTREETMAP SOME MORE)
3.3.4.1 Street Networks
OpenStreetMap’s street network data for the United States is of high quality, as OpenStreetMap
incorporates information on every road in the United States from TIGER Line shapefiles (OpenStreetMap.org, 2024). I obtain the portions of the street network that are available to pedestrians
using the OSMnx Python package (Boeing, 2017). 11
3.3.4.2 Land Use
I generate land use data from OpenStreetMap using the same process as described in C. Pilgram
and West, 2023, or in Chapter 2 of this dissertation:12 I download all OSM features with tags synonymous with various land uses.13 Next, I buffer linear road and railway features to approximate
10I wanted to rely on open source data wherever possible. However, while OpenStreetMap does report the locations of schools, it seldom reports what type of school any given school is. As such, it would only be possible to
identify elementary schools via name - an approach that would miss many schools that do in fact offer primary school
instruction.
11Specifically, I query OSM with the custom filter ’[”highway”] [”area”!∼”yes”][”highway”!∼”abandoned—construction—planned—platform—proposed—raceway—
motorway—motorway link—trunk”] [”service”! ∼”private”]’.
12See C. Pilgram and West, 2023 for a comparison of OpenStreetMap versus a local government-supplied land use
dataset. While local governments’ own land use data - where available - are likely to be more complete than OpenStreetMap in terms of the features we use to generate land use datasets, they frequently use idiosyncratic definitions
of land uses and would require extensive harmonization efforts to enable comparisons between different study areas.
13I define different land uses as follows: Any area with a value for the variable aeroway is considered part of an Airport; any area with a value of ”commercial” or ”retail” for the variable landuse or a value of ”commercial” for the variable building or any value for the variable office is considered ”Commercial/Office”; any area with a value of ”industrial” for the variables landuse or building, ”substation” for the variable power, or ”wastewater plant” for the variable
man made is considered ”Industrial”; any area with the values ”college”, ”hospital”, ”library”, ”place of worship”,
”school”, ”social facility”, or ”university” for the variable amenity or ”church” or ”public” for the variable building or
the value ”museum” for the variable tourism is considered ”Institutional”; any area with the value ”apartments” for the
50

areas perceived as railways or roadways. 14 To resolve overlaps and generate a land use dataset
in which any given place is assigned one land use (that is, there are no overlapping land uses),
I assume a hierarchy of land uses, from least to most dominant in perception: Recreational land
is the least dominant, followed by Multifamily Housing, Institutional, Commercial/Office, Industrial, Airport, Roadway, Highway and Railway. This data is intended to closely resemble land
use datasets available from different local governments such as the Minneapolis General Land
Use Survey (”GLUS”) (Minnesota Geospatial Commons, 2020), but with the same categories and
definitions of land use in all locations as to allow for both within- and cross-MSA comparisons.
3.3.5 Data Processing
Data processing and pathway buffer generation follow the same approach described in C. Pilgram
and West, 2023 or in Chapter 2 of this dissertation:
First, I generate all origin-destination pairs - that is, pairs of census blocks and food outlets or
census blocks and elementary schools - for which we want to generate pathway buffers. To reduce
processing power demands, I only generate routes for block-destination pairs that are no more than
half a mile apart in terms of straight line distance as these are likely to exceed walking distance for
most people.15
variable building is considered ”Multifamily”; any area with the value ”railway” for the variable landuse or the value
”yard” for the variable service is considered ”Railway”; any area with the values ”cemetery” or ”recreation ground”
for the variable landuse, the values ”golf course” or ”park” for the variable leisure, the values ”beach”, ”water” or
”wood” for the variable natural, the value ”attraction” for the variable tourism, or the value ”river” for the variable
water is considered ”Recreational”; and any area with the value ”brownfield” or ”construction” for the variable landuse
or the value ”construction” for the variable building is considered ”Vacant”.
14We consider the 10 meter buffer around linear railway features as ”Railway”, a twelve meter buffer surrounding
centerlines of roadways tagged ’trunk’ or ’motorway’ to be ”Highway”, and a six meter buffer surrounding all other
road centerlines tagged ’primary’, ’secondary’, or ’tertiary’ to be ”Roadway” type land use.
15This step and the subsequent routing are performed using the ”OSMnx” Python
package (Boeing, 2017). As described in Section 3.3, I define the street network available to pedestrians using the custom filter ’[”highway”] [”area”!∼”yes”]
[”highway”!∼”abandoned—construction—planned—platform—proposed—raceway—motorway—
motorway link—trunk”] [”service”! ∼”private”]’, and then perform routing between origins and destinations along
that network. All other data processing steps are performed in R using the ”sf” R package for spatial computations.
51

Next, I identify the shortest route for each origin-destination pair, using the portions of the
street network that are available to pedestrians.16 I then generate pathway buffers of each route by
buffering the route’s constituent road segments by an arbitrarily chosen 75 meters, and then identify
the areas within those buffers occupied by different land uses by intersecting the buffer layers
with the land use data generated from OSM. 17 Finally, I calculate the share of pathway buffers
occupied by each land use for each pathway that does not exceed half a mile in network distance,
18 and use census populations – both overall population and the sizes of specific sub-populations
such as different racial groups – to assign weights to individual census blocks’ pathways when
generating statistics for entire metro areas. In doing so, I group together several land uses that I
consider ”unpleasant”: I consider Industrial, Railway, Highway, Roadway, and Vacant land to be
detrimental to the walking experience, and label these as ”disamenity land uses”.
The resulting data consist of area shares for each land use for the routes from each census
block’s centroid to its respective nearest bus stop, station, or school. One possible use for this data
is that they can being presented in cartographic form. For some examples, see Figures 3.1 and
3.2, displaying the disamenity land use shares encountered when walking to bus stops and to rail
stations from every census block in the Minneapolis-Saint Paul metropolitan area, respectively, in
which the share of disamenity land uses encountered along the walk is displayed in a color scale
from blue and green via yellow to red, while portions of the urban area in which no bus stop or
rail station can be accessed on foot within half a mile are rendered in grey. 19 More importantly,
however, they allow us to make population level statements about the ability and the pleasantness
16Note that for pathways to schools, I do not rely upon schools’ catchment areas, as no such polygon data are
consistently available for all school districts. Instead, I make a simplifying assumption that children in a given census
block are likely assigned to attend their respective closest elementary school. This assumption may also reflect an
ideal desire for students to be able to attend their respective closest schools; to overcome what are frequently racist
and classist planning mistakes that have led to divergence from this ideal.
17See C. Pilgram and West, 2023 for discussion on choosing 75 meters as the buffer size.
18The distance band choice of half a mile is inspired by the frequent use of a half-mile as a proxy for an acceptable
walkable distance. For more discussion of half a mile as an acceptable walking distance for accessing or egressing
transit, see Weinstein Agrawal, Schlossberg, and Irvin, 2008 and Larsen, El-Geneidy, and Yasmin, 2010.
19Similar maps for other metropolitan areas are available upon request; in future work, I hope to build a web
mapping service around this.
52

of running common errands on foot, and to compare both access and presence of obstacles between
different segments of the population (racial, income, urban/suburban/exurban portions of MSAs).
3.4 Analyses
3.4.1 Experienced Disamenity Shares
To obtain experienced disamenity shares by metropolitan area and by racial group, I calculate
population-weighted averages of the shares of each land use across pathway buffers in each MSA,
taking the pathway buffers for the closest bus stop, rail stop, and school for each census block by
network distance if there are multiple options within a half-mile of straight line distance. As such,
this is a summary statistics exercise of sorts.
To estimate the different experienced disamenity shares of different sub-populations within
each metro area, I also calculate population-weighted averages using sub-populations defined by
race, by income group, education level, age group, and vehicle availability.
3.4.2 Regression Analyses
To test for within-metropolitan area associations between pleasantness of accessing amenities such
as transit stops or schools on foot, I estimate two cross-sectional regression models. First, I model
whether or not it is possible to access transit stops (or elementary schools) on foot within one mile
using a logit regression, modeling this binary measure of access as a function of a block’s racial
shares, education levels, median income, and distance from the metropolitan area’s central business district. In a second regression, I model for all blocks where such access is possible on foot
the share of ”unpleasant” land uses encountered on such trips as a function of the same socioeconomic variables of race, education, and income, this time using an ordinary least squares regression
model. In both regressions, I include metropolitan area dummies as well as interactions between
metropolitan area dummies and distance to the central business district to account for differences
53

Table 3.1: Population shares that can walk to amenities across all Metropolitan Areas Studied
between cities in spatial extent, walkability, and OSM completeness. In these regressions, each
observation represents one census block, and I cluster standard errors at the blockgroup level.20
3.5 Findings
3.5.1 Nationwide Results
As is presented in Table 3.1, across all metropolitan areas studied, about 54.5% of people live
within half a mile walking distance of a bus stop, compared to about 3.3% living within half a
mile walking distance of a rail station. 21 Conditional on living within half a mile of a bus stop
or station served by rail, 10.1% and 21.5% of land uses within the pathways are among those
assumed unpleasant. 22 By contrast, 19.3% people live within half a mile of walking distance
of an elementary school, with 7.1% of land uses inside the pathway buffers being unpleasant.
20Standard errors are clustered at the census blockgroup rather than the census block level as two variables - education level and median household income - are not available at the census block level. As a robustness check, I
re-restimate the same models with population weights, and obtain comparable results.
21Note that not all metropolitan areas have rail transit, and in some, the only rail stations are Amtrak or similar long
distance trains. As such, the MSAs entering this comparison are not the same for rail transit stops as it is for bus stops
and elementary schools.
22Note that surface level rail corridors are themselves one of the land uses assumed to be unpleasant, driving much
of the difference between bus stops and rail.
54

Expanding the tolerable distance to a school to a mile, 52.2% of the population can walk to school,
with 7.5% of land uses acting as disamenities.
However, substantial differences exist in between racial groups both in terms of whether such
destinations exist within walking distance, and what land uses are encountered along the way:
Compared to Nonhispanic Black, Latino, or Asian populations, Nonhispanic White people are
about two thirds as likely to live within walking distance of a bus stop, and are also less likely
to live within walking distance of rail stations. Nonetheless, conditional on being within walking
distance of a bus stop, they experience a full percentage point fewer disamenity land uses along
the way. For walks to elementary schools, the disparity between Nonhispanic White and minority
populations is even greater: While 24.9% and 28.9% of Black and Latino populations respectively
live within half a mile of an elementary school, the same is true for only 13.6% of white people
- yet the share of unpleasant land uses within the pathway buffers is a full 2.3 percentage points
smaller for White than for Black populations.
3.5.2 MSA-Level Results
As is presented in Tables 3.2 and 3.3 for a selection of metropolitan areas, there is substantial
variation between metropolitan areas both in terms of the share of populations for whom it is
possible to access transit on foot (left columns), and in what population-weighted share of land
uses encountered are unpleasant to walk through or past (right columns). For example, while
86.2% of San Francisco/Oakland residents live within half a mile of their nearest bus stop and only
9.4% of the land uses encountered on those walks are unpleasant, just under one in three residents
of the Oklahoma City MSA are located within half a mile network distance of a bus stop, and one
eighth of all land uses encountered along the way are unpleasant.
Table 3.2 also shows how both whether walking to the nearest transit stop is possible and
whether it involves a lot of unpleasant land uses vary substantially between different racial groups
within the same metropolitan areas. To interpret and call out a particularly extreme example, look
at the Philadelphia metropolitan area: While 64.6% of the overall population live within a mile’s
55

Table 3.2: Experienced Disamenity Land Use Shares Walking to Bus Stops in Select Metropolitan
Areas
56

walk of their nearest bus stop and 8.4% of land uses encountered within 75 meters of those walks
are unpleasant, there is an almost 35 percentage point difference between Philadelphia’s Black and
White Non-Hispanic populations in terms of ability to walk: Only around half of Non-Hispanic
White people live in areas where walking to a bus stop is possible within a mile, compared to over
87% of the same metropolitan area’s Black population or almost 80% of Latinos. Yet, for those
Non-Hispanic White populations that do live within a mile’s walk of their nearest bus stop, the
proportion of land uses encountered on those walks that are unpleasant is almost one fifth lower
that encountered by Latino populations in the same metro area, and is also substantially lower
than for any other minority group. While Philadelphia presents a particularly egregious example,
this pattern is not unique to Philadelphia; across most metropolitan areas white populations are
generally located in areas with less public transit service, but conditional on living near a bus stop,
are exposed to fewer ugly land uses when walking to it.
As bus service is far more common than rail service, the share of the population within a mile
rail transit is far smaller than that within a mile of bus services. Nonetheless, as is shown in Table
3.3, the discrepancies between racial groups - both in terms of ability to walk to stations and in
terms of land use exposure along the way - are similar overall. Only two of the metropolitan areas
evaluated in this study have at least ten percent of their population living within half a mile walking
distance of rail service of any kind, with another seven MSAs having between five and ten percent
of their populations within that same distance band. 23
As is presented in Table 3.4 for a selection of metropolitan areas, there is also substantial
variation between metropolitan areas both in terms of the share of populations for whom it is
possible to access schools on foot (left columns), and in what population-weighted share of land
uses encountered are unpleasant to walk through or past (right columns). While more than three
quarters of San Francisco/Oakland residents live within a mile of their nearest elementary school
and only 6.5% of the land uses encountered on those walks are unpleasant, less than a third of Flint
23Expanding the permissible walking distance to a full mile, only 17 of the metropolitan areas evaluated have at
least ten percent of their population living within walking distance of rail service of any kind, with another 20 MSAs
having between five and ten percent of their populations within one mile of rail service
57

Table 3.3: Experienced Disamenity Land Use Shares Walking to Rail Stations in Select Metropolitan Areas
58

Table 3.4: Experienced Disamenity Land Use Shares Walking to Elementary Schools in Select
Metropolitan Areas
59

residents are within a mile of an elementary school, and almost an eighth of land uses encountered
are unpleasant.
Table 3.4 also shows how both whether walking to the nearest elementary school is possible and
whether it involves a lot of unpleasant land uses vary substantially between different racial groups
within the same metropolitan areas. 24 To interpret and call out a particularly extreme example,
look at the Baltimore metropolitan area: While 54% of the overall population live within a mile’s
walk of their nearest elementary school and just under 7% of land uses encountered within 75
meters of those walks are unpleasant, there is a 30 percentage point difference between Baltimore’s
Black and White Non-Hispanic populations in terms of ability to walk: The Non-Hispanic White
population is largely in areas where walking to school is not possible within a mile, unlike the
same metropolitan area’s Black, Latino, and Native populations. Yet, for those Non-Hispanic
White populations that do live within a mile’s walk of their nearest school, the share of land uses
encountered on those walks that are unpleasant is less than two thirds that encountered by Black
populations in the same metro area, and is also substantially lower than for any other minority
group. While Baltimore presents a particularly egregious example, this pattern is not unique to
Baltimore; across most metropolitan areas white populations are generally located further away
from their nearest schools, but conditional on living near a school, are exposed to fewer ugly land
uses when walking to their school.
3.5.3 City-Level Maps
While the aggregate statistics presented in previous sections demonstrate the existence of inequities
in land use exposure across demographics, the greatest variation is across space. Such variation is
best communicated via maps; for which Figures 3.1 and 3.2 are examples. Additional maps are in
an Appendix. 25
24Note that population shares presented here are out of the entire population of any given block. In future iterations
of this work, I intend to re-calculate population shares within walking distance of schools based on the elementary
school age population of each block.
25In the future, I hope to provide a web map service presenting maps for all metropolitan areas studied.
60

3.5.4 Regression Models
Next, as was previewed in Section 3.4, I perform regressions, taking the block-level intermediate
data underlying the numbers presented in Table 3.4 to test for statistical relations with socioeconomic variables such as race and income.26 As is presented in Table 3.5, all else held equal, census
blocks with whiter and more affluent populations are generally less able to access elementary
schools on foot within a mile. This is not entirely surprising given the fact that such populations
are more commonly located in low-density suburban portions of metro areas compared to nonwhite or poorer populations. Interestingly, both people with less than a high school education
and those with more than a high school education are more likely to live in places where errands
such as walking to school can be performed within a mile, compared to those with only a high
school education. However, conditional on being able to conduct such errands on foot, there exists
a negative association between whiter and more affluent populations and ”unpleasant” land uses:
All else held equal, whiter, more educated, and more affluent populations who live within a mile
of elementary schools face fewer land uses such as industrial parcels, railways, or freeways when
accessing them on foot.
3.6 Discussion and Conclusion
Some caveats remain necessary: OpenStreetMap data quality is not uniform across space, and
variation in completeness is likely correlated with demographics due to the digital divide. Since
OSM is maintained by volunteer contributors, it likely better reflects real world - both in terms of
street network or land use variables and in terms of destination features such as food outlets - in
places that are more educated and possess higher incomes (Borkowska and Pokonieczny, 2022;
Herfort, Lautenbach, Porto de Albuquerque, Anderson, and Zipf, 2022; Zhou, Wang, and Liu,
2022).
26Future versions of this paper and the ACSP presentation will include the same regression models for walks to
grocery stores beside the regressions for walking routes to schools. While I have performed the same analysis for food
outlets, I omit it in this version of the paper for brevity’s sake.
61

Table 3.5: Regression Analyses for Walking Routes to Bus Stops and Elementary Schools
62

The method employed here – first developed to study first/last mile access to public transit (C.
Pilgram and West, 2023) – is not limited to the particular topics discussed in this paper; indeed
it could be applied to any local accessibility topic, and can use weighting schemes other than the
nighttime populations employed here to reflect other populations. 27 The method’s usefulness is
drawn from its simplicity; by abstracting away from detail it can be applied at very broad scales,
without requiring too much computing power. Importantly, it is not intended to replace environmental audit methods such as the Irvine-Minnesota Index, which remain far more thorough and
complete a measure of pedestrian quality, especially when the area of interest is at a scale that can
in fact be assessed in such detail.
Nonetheless, it allows for some insights that would not be possible otherwise, allowing for
comparisons between metropolitan areas and unearthing population-level associations between
pedestrian quality and socioeconomic variables. In doing so, I bring to attention what is ultimately
an environmental justice issue: Populations that are more likely to be dependent on travel via
public transit or on walking for their errands are more likely to face land uses such as industrial
land, railways, or highways, which may act as barriers to performing errands on foot or at minimum
cause discomfort along the way.
Specifically, I find that whiter, more affluent people tend to live in lower density environments
with little transit service, thus often being located outside of walking distance to the nearest bus or
rail stop; and that conditional on being within walking distance, the land use mix encountered by
whiter and wealthier populations while walking to bus stops or transit stations tends to include far
less ”disamenity” land uses such as industrial areas, highways, or railways.
The first finding is primarily a reflection of residential settlement patterns in North American
cities. To the extent low-density suburban environments are served by rail transit at all, stations
are often designed around access and egress by private automobile, be it through park-and-ride
parking lots or ”kiss-and-ride” dropoff zones. By contrast, central cities of metropolitan areas and
27An example for this would be to use a data source like LODES to simulate commute routes and estimate disamenity land use shares encountered on commutes.
63

otherwise denser areas – which often house greater minority populations – have far denser transit
service, with more of an expectation of access and egress occurring on foot.
The second finding is perhaps of greater importance. While not as visceral as the issues commonly discussed in the environmental justice literature, one might argue that at a nationwide scale,
this observation constitutes an environmental justice issue. 28 Perhaps it should not come as a
surprise that at the metropolitan area level, experienced disamenity shares are virtually the same
no matter the destination: Since walking from residential starting points to piublic transit stops
requires walking through much of the same environment as walking to other destinations such as
schools or transit stops, variation across destination types is almost inherently hyperlocal. It is
worth noting that the land use shares as reported in this study likely understate inequities in exposure to unpleasant land uses: The difference in experienced land uses is exacerbated by who is
transit dependent, as non-transit-dependent populations have the option not to experience ”bad”
land uses up close.
Can this be solved by planners and policymakers? At the scale of this study, perhaps no: The
inequities discussed here are the cumulative observation of seemingly minute differences across
tens of thousands of blocks across the entire United States. However, at a more local scale, planners
can be mindful of land uses within approaches to new transportation infrastructure, such as new
light rail lines. Light Rail or Bus Rapid Transit (”BRT”) lines are often constructed within existing
major roadways or even in freeway medians for the sake of easier planning (Harris, 2020). While
this may simplify the building corridors in the first place, it also makes for stations whose access
requires traversing those same roadways, or entering what Jane Jacobs may refer to as a ”border
vacuum”.
Both urban form and land uses are remarkably persistent of time, to the point where preColumbian and Roman roads still explain settlement patterns to this day (Garcia-L opez, ´ Holl,
and Viladecans-Marsal, 2015). In the short term, interventions such as sound barriers, planters,
and other urban design elements can serve as mitigation strategies to reduce the unpleasantness of
28A large body of environmental justice literature studies the locations of polluting plants.
64

otherwise harmful land uses. On longer time frames, cities can use zoning and other housing policy levers as tools to place more housing units – and thus more people – within areas from which
accessing public transit infrastructure is pleasant, as part of Transit-Oriented Development. Cityor local-level maps presented here such as the one in Figures 3.1 and 3.2 may draw attention to
areas of particular concern, and could assist in prioritizing highly visible disamenity land uses for
improvements to the pedestrian environment.
This work leans heavily on assumptions borrowed from the environmental psychology literature. To help calibrate interventions, future work should test – experimentally or observationally –
to what extent exposure to different land uses acts as an impediment to accessing public transit.
65

Figure 3.1: Share of Disamenity Land Uses Encountered on Walks to Bus Stops in MinneapolisSaint Paul, Minnesota
66

Figure 3.2: Share of Disamenity Land Uses Encountered on Walks to Rail Stations in MinneapolisSaint Paul, Minnesota
67

Chapter 4
Developer’s Choice: The Evolving Provision of New Housing
Supply & Declining Elasticity
Clemens A. Pilgram and Christian L. Redfearn
Abstract
It is becoming clear that aggregate housing supply within US metropolitan areas (MSAs) has become less elastic over the last 20 years. This finding is the result of individual developers making
different choices about what types of housing to build and where to build it. In order to learn
more about how housing supply is changing, we impose structure on the nature of new supply by
introducing differing channels through which new housing supply is added. While it is common
to treat development as a largely “black box” process, the reality is that there are significant differences across developers and development firms in terms of scale, use of construction technology,
and the capital structures used to finance development. By organizing new housing supply along
these dimensions, we are able to uncover systemic changes in how house metropolitan housing
markets have accommodated new supply over the last 20 years. Although exurban development
remains the primary channel through which new housing is supplied in US urban housing markets,
its share of new supply has fallen sharply. An increasing share of new housing units can be found
within “built-out” suburban submarkets, in and around denser urban subcenters, and – in many
68

metropolitan areas – amid post-industrial locations. We note that housing built in infill locations
requires both higher-density construction technology and higher break-even rents. These highercost developments may be playing a contributing role lower supply elasticity and in higher house
prices. Our results suggest that the internal dynamics of a metropolitan’s housing market should
be considered in developing policy responses to higher prices.
69

4.1 Introduction
In this paper, we address how US metropolitan areas have accommodated new housing supply
by organizing new supply into developer types. Rather than treating the industry as a single,
competitive industry with an aggregate, upward-sloping supply curve, we study new supply by
the developer types we see in practice. We focus on individual developer types because the vast
majority of the US housing stock has been built by private developers, and each year the new
supply is the outcome of their choices. While the home-building industry is heavily regulated in
terms of zoning and building codes, it is most often a private developer who makes choices about
what to build and where to build – both in response to market pressures and in light of constraints
imposed by the local regulatory context. In practice, we see large-scale detached single-family
home developers building in the exurbs at the same time we see massive mixed-use developers
build high-density housing in post-industrial sites near the traditional core. These developments, of
course, are part of an aggregate housing supply, but typically occupy distinct housing submarkets,
built by distinct firms, using different building technology, and for different types of households.
While there is likely a continuum of builder types, we settle on four new housing supply “channels” for this analysis. In addition to exurban, low-density developers and high-density, postindustrial developers, we also see multifamily developers in and around existing areas of higher
density, and smaller-scale infill developers who find excess land amid low-density suburban sites.
This four-way taxonomy is coarsely defined, but the basic groups match our intuition and provide
compelling results. By organizing new supply in this way, we are able to document marked differences in the distributions of new supply and stark differences in how different metropolitan areas
have accommodated new growth between 2000 and 2020.
Based on our typology into four distinct channels, we call to attention four key findings: First,
in most metropolitan areas, exurban development still accounted for a majority of new housing
supply – but this channel’s share of new supply fell sharply from 60 percent to 44 percent from
the 2000s to the 2010s. This is a remarkable shift from one decade to the next, from 6.0 million
new net units during the 2000s to just 3.0 million new net units during the 2010s. Second, the
70

balance of new supply arrived into infill locations. We call this out because infill development is
generally more costly to develop than building in the exurbs. Third, we find remarkable variation
in way different MSAs have accommodated their growth. Younger, smaller MSAs build the vast
majority of their new net units in the exurbs; older and larger MSAs saw the most decline in
exurban development in favor of much more urban infill and post-industrial areas. But perhaps the
most striking finding is a broad pattern of “suburban sclerosis.” Once Census block groups achieve
housing densities from 1 to 4 units per acre, the share of new supply falls well below its share in
the stock as a whole.
The primary contributions we make in this paper are twofold. First, by disaggregating our
analysis of housing supply, we find striking local patterns within metropolitan housing markets.
And second, by providing a more realistic treatment of the house building industry, we are better
able to see that housing supply has changed in significant ways. The aggregate nature of much
housing supply research, and the rather “black box” nature of housing production functions appears
inadequate to explain housing dynamics within metropolitan area housing markets. Inferences
drawn from aggregate analyses may miss important internal housing dynamics about changing
elasticity by “channel” and likely fail to inform policy makers about the particular challenges for
new housing and land-use policies aimed at increasing supply in light of these dynamics. The
fact that the building industry is not a generic black box is obvious, but our results make clear
the benefits of incorporating more structure into our treatment of development firms and how they
contribute to new supply.
While our goal in this paper is to understand housing markets better, we are ultimately interested in understanding whether policy interventions are likely to succeed or not. That is, there
are now numerous conversations about housing affordability, about its many linkages to quality
of life and productivity, and about what to can or should be done about it. Among these, scholars often look either at aggregate results across metropolitan areas or at highly-local studies that
focus on neighborhoods as their unit of analysis. As already discussed, aggregate analysis may
miss important dynamics, but highly local analysis may also miss many larger forces at work with
71

a metropolitan area. Our aim is to examine local dynamics within an entire metropolitan area
(MSA) across a broad comparison group of MSAs. The results suggest that policy makers should
consider the variety of underlying mechanisms at play as they consider new policies in response to
rising housing prices.
We proceed as follows: Section 4.2 summarizes the existing literatures on the supply elasticity
of housing and on what policy experiments have been attempted to encourage urban infill housing
development. Section 4.3 discusses the data sources and methods employed. Section 4.4 describes
findings at a high level. Section 4.5 outlines possible mechanisms at work, and how local amenities
might play a key role. Section 4.6 concludes.
4.2 Framing Our “Channels”
While there are both large and mature literatures on housing supply and housing policy, we argue
that both may be served by actually documenting what has been changing within metropolitan
housing markets over the last two decades. We organize new supply into four distinct types of
housing development - we call “channels”. These channels are defined by land uses and housing
density at the beginning of each decade we study. We extend the literature on elasticity by discussing what changes in the relative sizes of housing channels mean for overall housing supply
elasticity. In particular, we use our four channels to examine various policy levers relative to those
designed with aggregate dynamics in mind.
Housing affordability is increasingly an issue everywhere: In 2022, the Pew Research Center
found that fully 85 percent of participants in a national survey responded that housing affordability
was a problem. While 91 percent of urban respondents thought housing affordability was a problem, so too did 82 percent of rural respondents. 1 Prior to the most recent decade, a largely-elastic
building industry provided enough new housing units to limit price increases to just above inflation. Through the end of the 20th century, episodes of rapid house price appreciation were largely
1https://www.pewresearch.org/short-reads/2022/01/18/a-growing-share-of-americans-say-affordable-housing-isa-major-problem-where-they-live/
72

Figure 4.1: Inflation-adjusted housing costs and Per Capita Housing Starts
constrained to local markets – notably Boston in the mid-1980s, Los Angeles in the late-1980s,
and San Francisco in the late-1990s (Case, Shiller, et al., 1994; Case and Shiller, 2003; McCarthy
and Peach, 2004). There were others, but – as Figure 4.1 shows – those too were exceptions: aggregate US house prices essentially followed inflation from 1980 to 2000. Figure 4.1 also shows
that the early correlation between CPI and aggregate sale prices disappears temporarily after 2000
with the mid-2000s real estate bubble, only to weaken markedly in the 2010s: During the expansion between 2012 and the emergence of COVID in the first quarter of 2020, inflation rose only
13 percent, while US house prices rose 45 percent. This striking divergence was broad-based,
with metropolitan areas that were previously known for their elastic supply response experiencing similarly large price increases. Indeed, even in Houston – the poster-child for supply-elastic
housing markets with low regulatory barriers to construction, house prices rose by 55 percent – far
exceeding inflation and even surpassing the national average of house price increases. Such high
housing cost has several undesirable consequences, reducing residential mobility and thus access
to opportunity (Stawarz, Sander, and Sulak, 2021), stunting overall economic growth (Hsieh and
Moretti, 2019), and rendering government housing programs needlessly expensive (Corinth and
73

Irvine, 2023). While there are many demand fundamentals behind outcomes in housing markets,
our focus in this paper is on documenting if and how housing supply changed.
Despite their recovery from lows following the Great Financial Crisis, per capita housing starts
remain far below historical levels (see Figure 4.1). While there is a broad consensus that land-use
regulation constrains development at the margin, there are other potential contributors to lower
supply. Several recent papers have found declining housing supply elasticity (Baum-Snow and
Han, 2019; Orlando and Redfearn, 2023). These papers found marked variation in falling elasticity
both across US counties and within metropolitan areas, but agree that overall elasticity fell. Despite
the concern about insufficient supply, gentrification has also been a topic of interest. Rather than
seen as a bright spot of new supply, the interest has been largely related to equity, and to who
benefits from new development (Asquith, Mast, and Reed, 2023; Been, Ellen, and O’Regan, 2019;
Mast, 2023). That debate is not a primary interest of this paper, but we note that increasing supply
through infill development – and gentrification – has been rising for 20 years or more (Couture
and Handbury, 2023) while the share of new units through suburban and exurban channels have
broadly declined.
These papers – at least indirectly – raise the issue about where and through what channels new
units are being added to the stock. Gentrification is generally studied in highly local geographies,
while much of what we know about housing supply is aggregate in nature – often using metropolitan areas as the unit of analysis. And yet while a house’s proximity to local amenities is a bedrock
principle for urban and housing economics, we often study housing supply using aggregate geographies like MSAs. In doing so, we implicitly assuming all of the units within an MSA are
substitutes. We know that housing submarkets exist (Pryce, 2013), are imperfect substitutes for
each other (Ren, Wong, and Chau, 2023), experience significantly asymmetric price movements
over time (Bogin, Doerner, and Larson, 2019), influence automated valuation models (AVMs)
(Bourassa, Hoesli, and Peng, 2003) and potentially bias aggregate price estimates as well (Malone
and Redfearn, 2022). With local prices varying within a metropolitan area, we might expect new
74

supply to arrive to different locations within metropolitan areas as developers discover potentially
profitable locations in the presence of new pricing.
Baum-Snow, 2023b documents that recent changes in the housing stock are decidedly nonsymmetric, with low-density suburbs being the destination for a large share of the new units over
a period from 1980 to 2018. Orlando and Redfearn, 2024a recently compared the changing spatial density of new net units among large and growing metropolitan areas in Texas and California.
Among eight metropolitan study areas, they find that infill locations and higher-density housing
comprised a growing share of new housing supply. Both papers suggest an evolving supply function, in which different locations – and likely different building technologies – are changing how
new units are provided. This pattern in conjunction with differing unit construction costs suggest
that overall costs of new construction is rising. This means that the “break-even” rents required to
justify new construction is rising as well (Eriksen and Orlando, 2022).
It is for these reasons that we organize our analysis of new housing supply around developer
choices: their choices result in the new housing that we are trying to understand. While their
role in new supply is central, little is known about development because it is often assumed to
be a competitive and atomistic industry, one through which larger fundamental determine housing outcomes (Somerville, 1999). A small number of papers address the industrial structure of
home building (Quintero, 2022; Wissoker, 2016). We simply assert that developers’ choices are
influenced by dynamics within the metropolitan area, and that the spatial distributions of incomes,
preferences, local amenities, and the housing structures, themselves all act to change what and
where developers choose to build.
Of course, housing developers seek good risk-adjusted returns. If there is a surge in demand
for entry-level housing, simple “stick” or wood-framed construction at the periphery is the lowest
cost way to provide it. If these prospective new buyers are willing to commute, then there may
be a profit opportunity for developers. Likewise, for households who are looking to live within
walking distance to high-end amenities, their traditional choice has been in downtowns. Land here
is expensive, requiring high-density housing construction types and significantly higher per unit
75

costs. If there are more households willing to pay for the access to these amenities, there is a
chance for profitable development. We arrived at our four channels by observing broad categories
of developers, and in doing so we find striking dynamics within the metropolitan aggregates.
For example, while both a low-density, exurban channel and a high-density infill channel contribute to aggregate housing supply, single-family residential (SFR) units in the periphery differ
from condominium towers in post-industrial core areas. The new units offer different housing
characteristics, different local amenities, and, importantly, imply different land and construction
costs. We find that exurban, or “greenfield,” development of new low-density housing remains the
low-cost way channel, and indeed it remains the dominant source of new housing supply among
our sample MSAs. That said, the share of new units built at the periphery varies widely across
MSAs. We find that suburban and urban infill are distinctly different channels, in which local
amenities are central to the location of new supply. We also find an emerging channel in the development of post-industrial land near urban cores. The differential growth rates in new housing units
by channel raise obvious challenges to some policies aimed at making housing more affordable.
Often touted as a solution to the ongoing housing shortage, the relaxation of zoning restrictions
by upzoning urban areas or eliminating rules entirely is perhaps the most frequently discussed and
well-studied policy tool for encouraging infill development (Gray, 2022; Kazis, 2023; M. Kim,
2024). Upzoning can be either blanket upzoning - where an entire municipality does away with
restrictions – or targeted – such as allowing for more intense development in specific corridors or
in areas with better public transit service. Such upzonings are generally found to result in property
value increases reflecting the potential for future development. However, this potential is rarely
realized (Freemark, 2020, 2023; Kuhlmann, 2021). Stacy, Davis, Freemark, Lo, MacDonald,
Zheng, and Pendall, 2023 find using a panel of several cities that while upzoning does translate to
modest increase in housing supply, these effects are entirely concentrated in units at the higher end
of the rent price distribution.
One issue perhaps dampening the effect of upzonings and related policies to encourage housing
production is that zoning rules are themselves a reflection of underlying local attitudes regarding
76

housing construction (Fang, Stewart, and Tyndall, 2023). Such attitudes may constrain housing
production via regulatory pathways other than just zoning laws (Manville and Monkkonen, 2021;
Monkkonen, Lens, and Manville, 2022).
While rarely stated explicitly, the literature on upzoning focuses almost entirely on existing,
already relatively built-out residential portions urban areas; it rarely studies greenfields or brownfields. This focus on already built-out residential areas is also a common thread for the nascent
literatures on permitting specific types of infill housing, such as “missing middle” multifamily
(MF) housing (Dong, 2021, 2023), townhomes (Wegmann, Baqai, and Conrad, 2023), or accessory dwelling units, also known as “ADUs” (Marantz, Elmendorf, and Kim, 2023b). While each
of the studies cited here evaluates a policy that intends to fit more housing units into the landscape,
the specific policies – even if form-based – are implicitly specific to particular kinds of urban area.
In light of the broad concern about housing prices, local, state, and national policy makers
have been introducing regulation and legislation aimed at providing more affordable housing. Rent
control and eviction moratoria are more prevalent now than before COVID (Benfer, Koehler, Mark,
Nazzaro, Alexander, Hepburn, et al., 2023). At the same time, genuine efforts to streamline the
development entitlement process are also gaining momentum. Our interest is in understanding how
different metropolitan areas have met (or not met) housing demand, and how various land-use and
housing policies might influence affordability looking forward.
4.3 Channel Data
This study relies on data from three sources: The US Census Bureau’s 2000, 2010, and 2020
Decennial Censuses, and LODES WAC.2
The Decennial Census counts the number of housing units on any given block. We obtain those
census block-level housing unit counts for all states for each of the decennial censuses conducted
from 2000 until 2020 via IPUMS NHGIS, as well as census block group boundaries and land
2We also use OpenStreetMap for some supporting analyses.
77

area sizes via the tigris R software package.3 To reconcile changes in census boundaries between
censuses, we take census block-level housing unit counts for each of the four surveys and identify
which census block group their block’s centroid would fall into, allowing us to compare unit counts
over time for block groups.
LEHD LODES WAC is a data product by the US Census Bureau reports the number of jobs
by industry in any given census block for each year since 2002. 4 We aggregate the number of
manufacturing, transport, and warehouse jobs in the first year for which data are available up to the
block group level, choosing those industries because they are particularly land-intensive. Further,
we rely on LODES WAC to identify the locations of central business districts, which we define as
the census tract with the greatest employment density within and MSA, as well as to identify areas
with substantial manufacturing or warehousing employment in the first year for which the data are
available.
We then classify block groups into the four groups described below based on their residential
unit density, their distance to the central business district, and based on their number of manufacturing jobs in 2002.
We classify block groups into four types - “exurban”, “suburban residential”, “urban residential”, and “post-industrial” based on their housing unit density at the start of a decade (in 2000
for the 2000s, or in 2010 for the 2010s), their distance from the central business district, and the
number of manufacturing, transportation, and warehousing jobs as measures of land-intensive employment:
1. Exurban type block groups are block groups with a housing unit density of less than one
housing unit per acre at the start of a given decade.
2. Suburban residential type block groups are block groups with a housing unit density of
between one and four housing units per acre at the start of a given decade.
3While other sources on housing construction – such as the Census Bureau’s Building Permits Survey – offer far
more granular a temporal resolution, their spatial resolution – often limited to the county or municipality level – is
insufficiently granular for constructing the typology suggested in this paper.
4For most years, LODES WAC data series begin in 2002. In some states, the data series begins in later years. For
states where no LODES WAC data are available as early as 2002, we use the respective earliest year.
78

Table 4.1: Channel Shares, All MSAs combined, 2000 and 2010
3. Post-industrial type block groups are block groups located within the bottom quarter of
the distance distribution from their respective MSA’s central business district that had at
least 50 manufacturing, transportation, or warehousing-related jobs in 2002 in 2002 (or the
respective earliest year for which LODES WAC is available), and that have fewer than four
housing units per acre at the start of a given decade.
4. Urban residential type block groups are all block groups with a housing unit density of
more than four housing units per acre at the start of a given decade.
We apply the criteria in the above order to resolve ties. The above criteria divide metropolitan
areas into channels whose area, housing unit, and population shares across all MSAs are shown
in Table 4.1. Perhaps unsurprisingly, we observe that urban and suburban channel blockgroups
account for outsize proportions of existing housing units and populations, whereas exurban blockgroups account for over just under a quarter of metropolitan housing units and populations despite
comprising over 90 percent of metropolitan land. Missing from this table is how new supply is
added among the four channels. We address this in the next section.
4.4 The Dynamics Across Four Supply Channels
It will be useful to compare the evolution of our supply channels against changes in the aggregate
housing stock across our 52 metropolitan areas. As benchmarks, the US population from 2000 to
2010 grew by 9.7 percent and by 7.4 percent from 2010 to 2020. By comparison, the population
within our 52 MSA sample grew by more than the US, growing by 10.6 percent from 2000 to
2010, and 9.7 percent from 2010 to 2020, reflecting a continued trend in US population from rural
to more urban locations. The total housing stock within our sample grew by 14.3 percent from 2000
79

Figure 4.2: Housing Supply Over Two 10-Year Periods
to 2010, and 8.7 percent over the decade from 2010 to 2020. So, from 2000 to 2010, the housing
supply growth rate far exceeded both the population in our sample MSAs and the population of
the US as a whole. But from 2010 to 2020, the housing stock growth exceeded the US population
growth rate, but not that of our gross MSA population. While this is consistent with higher house
prices – and lower elasticities, we will next show why the aggregate figures tell an incomplete story
of how housing supply changed over the two decades we study.
Using the decennial Census, we plot the housing stock for our 52 metropolitan areas across our
three cross sections: 2000, 2010, and 2020. Figure 4.2 shows the pair-wise comparisons to show
the relative growth in our 52 MSA housing stocks between the two decades. The dashed lines are
forty-five degree lines, making clear that housing supply growth was the norm. The blue regression
line reflects both the consistent growth among the sample MSAs and the shift to slower growth –
from 14.3 percent growth from 2000 to 2010 to the 8.7 percent growth in the subsequent decade.
There appears to be more heterogeneity in the growth rates in the first decade relative to the second.
Despite the apparent heterogeneity, the correlation in MSA growth rates is 0.76: MSAs that grew
rapidly from 2000 to 2010, were more likely to grow faster over 2010 to 2020. At this level of
aggregation, there is little to suggest any marked shift in metropolitan new housing supplies.
80

The one MSA that did not grow in terms of housing supply is New Orleans, and then only
between 2000 and 2010. Exploring this exception offers a good way to explain our measure of
new housing supply and highlight what is lost in aggregation. Note that the changing housing
stock in New Orleans and the number of new net units are not the same thing. We are focused on
studying how developers choose to add new housing units and less interested in the total change in
the housing stock.
Hurricane Katrina and the flooding that followed was devastating to New Orleans. According
to Vigdor, 2008, the housing stock in New Orleans was 215,000 units in 2000, but had fallen
to 106,000 by 2006. Of this smaller stock, 32,000 were likely damaged and not habitable. But
Vigdor also reports 57,000 building permits were issued in the years after the storm. This would
be less than half of the estimated damaged or destroyed, but also makes clear how our accounting
differs from aggregate totals. Though the total housing stock of New Orleans fell over the decade,
developers still built thousands of new housing units during the same time period. We are interested
in seeing what and where the new units were added.
To do this, we examine the housing stock at the Census block group level. We note when
the housing count within a block group increases and define this as new net housing supply. Our
measure of total new supply is the sum of the positive changes in housing counts within the block
groups; we do not subtract from this total new net supply when housing unit counts fall in other
block groups. The choice of developers to build more supply and the decision to tear down a unit
– for any purpose – are two distinct decisions. As such, New Orleans may have lost units and
population after Hurricane Katrina in aggregate, but developers still entered the market and built
large amounts of new housing. The net aggregate housing supply for the metropolitan area fell
from 556,000 to 546,000 between 2000 and 2010, but the sum of the block groups that added new
net units grew by over 47,000.
While these thousands of units were added within the New Orleans MSA, Table 4.2 makes
clear that new supply was decidedly not symmetric within that particular metropolitan housing
market. The table reports the shares of new supply by our four supply channels. For New Orleans,
81

Table 4.2: Share of Net New Units by Supply Channel, 2000-10 and 2010-20
82

59 percent of all new supply between 2000 and 2010 occurred in exurban areas, and more than 25
percent via the suburban channel. Only twelve percent and less than five percent of new housing
units appeared in urban and post-industrial areas, respectively. In other words, almost 85 percent
of all new housing for that decade occurred outside the more urban channels. Contrast this with
the composition of the new supply over the following decade: only 46 percent of new supply
was exurban in nature. The rest of the new supply largely reflects a change for more density:
29 percent in suburban areas, and approximately 25 percent between both the urban and postindustrial channels. So while aggregate housing supply would correctly record a net loss in supply
from 2000 to 2010 and a modest increase over 2010 to 2020, our use of channels allows us to see
significant growth via some channels and, importantly, a notable shift in the composition of new
supply between decades.
The summary statistics across the top of the table shows broader trends in changing housing
supply channels across our 52 MSAs. The first is the broad aggregate decline in the share of new
supply through the exurban channel – construction largely of single-family homes at the periphery
of a metropolitan area. The average share of new housing supply via exurban channel across
our MSAs from 2000 to 2010 was 60 percent. That figure from 2010 to 2020 was 44 percent.
Suburban shares rose from 23 percent to just 28 percent. Against this large exurban decline and
only modest suburban growth, urban infill and post-industrial growth showed remarkable growth:
from 14 percent to 24 percent via the urban channel, and from 3 percent to 4 percent via the
post-industrial channels, respectively. These dynamics suggest a great deal of change that was not
suggested by the aggregate stock numbers in Figure 4.2. As is demonstrated in Table 4.3, much of
this shift – and of the reduction in aggregate supply presented in Figure 4.2 comes from a dramatic
decline in units contributed by the exurban channel and a modest decrease in the number of units
delivered by the suburban and post-industrial channels; only partially offset by an increase in units
delivered by the urban channel.
Table 4.2 also reveals a marked heterogeneity amid several broad trends. It is abundantly clear
that new housing supply varied broadly by channels across MSAs. There are MSAs that provide
83

Table 4.3: Changes in Housing Stock by Supply Channel, 2000-10 and 2010-20
84

the large majority of their new housing by building in the greenfields at their periphery, and there
are a handful of MSAs that provide more than 40 percent of their new housing supplies via the
combined urban and post-industrial channels. Some add significant new supply from urban and
post-industrial locations. And while the suburban channel adds less new supply than its share of
the existing housing stock, some MSAs saw marked increases in their shares of NNUs via the
suburban channel. Indeed, the places added most by adding new supply in suburban channels are
not defined by geography – and by extension, not by broad patterns in land-use regulation. Smaller,
but vibrant MSAs add a lot of supply via exurban and suburban channels. Larger, more mature
MSAs accommodate more of their growth via the urban channel. This is true of “static” MSAs –
Chicago, Baltimore, Boston, Milwaukee, and Philadelphia, and true of long built out MSAs – Los
Angeles, San Francisco, and San Diego. Even recent success stories, like Denver and Seattle, are
turning to more urban infill development.
The dynamism across the MSAs is a result of disaggregating. Consider the inferences that are
made on aggregate housing prices and aggregated new supply, these would measure something average in nature, but not reflect how responsive particular types of developers might be in particular
locations. This set of shifting compositions of channels strongly suggests that supply elasticities
are likely quite varied by channels, and possibly suggest that housing and land-use policies might
better be designed with these channel dynamics in mind. This is discussed in our conclusion,
but one common trend seems apparent: suburban densities appear to impede further development.
Land-use regulation has been viewed as a primary causal factor in falling elasticities, but the consistently low new supply from the suburban channel suggests a different margins at work, beyond
regulation and topographical limits. Neither of these would explain the dynamics we document
here.
Table 4.4 presents each of our four channel’s rate of growth between the two decades, in addition to evidence of the heterogeneity in changes in housing supply within MSAs. 5 We find that
5Growth rates are calculated by dividing the number of new units added within blockgroups of a given channel by
the existing housing stock at the beginning of that decade. In order to avoid bias areas that are in fact losing units as
is the case in large portions of Rust Belt cities such as Detroit, we only include the change in units for blockgroups
where the change in units is positive.
85

Table 4.4: Growth Rates by Supply Channel, 2000-10 and 2010-20
86

growth is still accommodated largely in exurban locations, but the growth rate of in new supply via
the exurban channel has fallen (27 percent to 12 percent). The post-industrial channel continues to
have the highest growth rates (15 percent versus 38 percent in the 2000s) – if starting from a very
low stock – but already shows signs of slowing. Moreover, the share of post-industrial land within
our MSAs is quite limited. Once built out, it will likely provide much less new supply.
The growth rate of the urban channel – which lagged behind the suburban channel in the 2000s
– has surpassed the growth rate of the suburban channel in the 2010s. To be clear, the shares of all
four vary significantly, but the change in their growth rates is significant.
And while there is an apparent broad shift in how housing supply growth in accommodated,
the table makes clear how much variation there is across MSAs. And perhaps most interesting,
are the patterns among the 52 MSAs. Though many of the large, vibrant, and rapidly growing
MSAs – such as Denver, Portland, Salt Lake City, San Jose, Seattle, Washington, DC – show
a marked increase in new supply via the urban channel channels, these patterns are echoed in
Austin, Charlotte, Dallas, Nashville, Raleigh, and others. This shared pattern may be surprising,
if land uses are significantly constrained by regulation – the former are thought of as more-highly
regulated and restrictive relative to the latter.
Figures 4.3 through 4.6 make clear that housing supply appears to be changing within our
MSAs. Figure 4.3 illustrates the asymmetric shifts in shares of new net housing via the exurban
channel over the two decades we study. If MSAs were undertaking growth by scaling up in a
symmetric manner, the dots in both panels would be located along the 45 degree line – the share
of new net units arriving via the exurban channel would match the share of the exurban housing
stock at the beginning of each decade. The left panel of Figure 4.3 reports that the in all 52 MSAs
from 2000 to 2010 had shares of their new supply via the exurban channel that far exceeded their
share of the exurban housing stock. This is particularly true of the 2000 to 2010 period. And while
still true the following decade, the entire regression line shifted toward the 45 degree line. The
levels and patterns via the suburban channel, shown in Figure 4.4, are quite different. First, note
how little new supply is built in the suburbs relative to its share of the housing stock. Moreover,
87

Figure 4.3: NNU Shares vs Stock by Channel: Exurban
Figure 4.4: NNU Shares vs Stock by Channel: Suburban
88

Figure 4.5: NNU Shares vs Stock by Channel: Urban
there is a clear association between MSAs with higher shares of their stock in suburban locations
and a declining share of new net supply. While the data reflect some migration further from the 45
degree line over the two periods, the regression lines are largely similar.
Figure 4.5 and Figure 4.6 make clear where the declining share of new exurban supply is being
built. Figure 4.5 shows a marked change both in the share of the stock of housing in our urban
locations and in their relative shares of new net units. First note that there is a very large dispersion
in the shares of the existing housing stock with regard to the amount of urban housing. Of course,
New York City is highly urbanized, but most MSAs are not. Over the two decades, there was
a significant and broad-based rotation toward provision of new housing via the urban channel in
proportion to the underlying stock of urban housing stock. Whereas urban locations may have
earlier been seen as built out, they now appear to be areas in which profitable new supply can be
entitled and built. Furthermore, urban infill is becoming a more significant channel for all new
housing: only six of the 52 MSAs provided more than 25 percent of their new supply via the urban
channel between 2000 and 2010, but 14 did in the following decade.
The turn to more infill locations is particularly striking in Figure 4.6, which shows the transformation of the post-industrial supply channel. Across large and small MSAs, MSAs that are
89

Figure 4.6: NNU Shares vs Stock by Channel: Post-Industrial
dynamic and less so, all have turned to these types of infill locations to build more of their new
supply over the second decade of our data. For some MSAs, the post-industrial channel was largely
an afterthought in the first decade of our sample, but most delivered a non-negligible share of new
units over the second decade. It is likely that this change in supply composition toward infill channels will have implications for costs and break-even rents/price. Infill development is generally
more costly than single-family detached wood construction (Eriksen and Orlando, 2022).
To explore these supply dynamics, we re-plot the data in two ways. We bifurcate the data
coarsely – splitting the MSAs into groups based on their growth rates in new net housing supply.
We wanted to know if housing markets under pressure to build more built differently from those
under less pressure. Over the next several figures, the panel on the left is comprised of the MSAs
whose new net housing supply grew at faster than the median rate growth, while the slower growth
MSAs are plotted in the right-hand panel. In both panels, we label the MSAs so that we can begin
to develop a sense of which MSAs are experiencing the change and how they are accommodating
new supply.
Figure 4.7 is interesting for several reasons, but first because all of the MSAs are below the
forty-five degree line – all of them saw a fall in their exurban shares of new net supply. This is true
90

Figure 4.7: Exurban Channel Shares
of large and vibrant MSAs, as well as those who are barely growing at all. It might be useful to use
other covariates to explain the variation in the shares and growth rates among the four channels, but
this is left for future research. The panel on the right is also interesting because of its composition.
MSAs with slower growing new supply include a large number of Rust Belt cities, but also quite
their complements: larger and vibrant MSAs are among the slow growing MSAs with regard to
new net supply. New York City, San Jose, Los Angeles, and Sacramento are in this group because
of other constraints and because of high prices. This will be addressed in the next section.
Relative to the exurban channel figures above, the suburban results in Figure 4.8 betray less
change. To be clear, more infill housing within suburban neighborhoods occurred in the faster
growing MSAs, but the data clusters around the forty-five degree line and a relative stability in
their shares. If there is any pattern that might warrant further examination it is that both regression
line slopes decline: the places that built more suburban housing in the previous decade, build less
in the following decade.
The urban and post-industrial channels are where changing new supply is most clear and where
we believe more can be learned about housing supply by looking within metropolitan aggregates.
Figures 4.9 and 4.10 reveal a strong rotation toward more infill housing supply. This is true of
91

Figure 4.8: Suburban Channel Shares
Figure 4.9: Urban Channel Shares
92

all MSAs. But across the two panels, it is clear that those MSAs with vibrant urban cores have
experienced the most growth in infill. And those MSAs with very little urban housing stock in
either decade add very little in absolute numbers over either decade.
Figure 4.9 shows first how little of the total new net supply was built in urban locations in
the 2000s. During our 20-year sample, the faster growing MSAs were the smaller, secondary and
tertiary MSAs. Among these, few faced land constraints that would have necessitated infill development. Seattle, Washington, Denver, and Portland saw significant investment and new housing
construction in their cores, but the urban channels was a very small contributor among the smaller
MSAs. Among the larger – and in many ways constrained – MSAS, there was a notable shift in
new housing built via the urban channel. 10 of these MSAs built more than 20 percent of their new
net housing this way.
Figure 4.10 is perhaps more remarkable. The post-industrial channel was not a material source
of new supply in the 2000s. A decade later the aggregate story changed largely because of the
low base. For many MSAs, the post-industrial channel remained a very small share of new net
housing supply. But for more than 20 MSAs, post-industrial housing was at least five percent of
new supply, and, for a handful, growth in post-industrial areas outstripped growth in their urban
channel.
Taken together, these figures strongly suggest that local housing supply dynamics within a
metropolitan area should not be overlooked. Estimating housing supply elasticity by using MSAs
as the units of analysis is likely to draw incomplete inferences about new supply. The set of results
in this section suggests that each MSA faces a unique set of fundamentals, both across MSAs and
the channels within them. That is, the large and fastest growing MSAs in the 2000s accommodated
their growth not be scaling up its prior distribution of housing density, but rather by spreading out.
Austin, Las Vegas, Orlando, and Phoenix experienced phenomenal growth in their housing stocks
while suburban, urban, and post-industrial channels in these MSAs changed only modestly. All
three of these examples also grew rapidly in the 2010s, but did so though a much different mix of
channels. By contrast, more static MSAs were far from static with regards to development. New
93

Figure 4.10: Post-Industrial Channel Shares
net housing supply in the periphery of Buffalo, Cleveland, and Cincinnati all grew at double digit
rates during the 2000s, while population and units fell in their cores. Detroit’s aggregate population
has been stable for decades, but the MSA achieved a growth rate of new net housing units in the
exurbs of 13 percent.
4.5 Changing Channels: Why?
Thus far we have attempted to make two contributions, but in doing so an avoided an obvious
question: why? What explains both the broader patterns across the 52 MSAs and the marked
heterogeneity within them? The results in the previous section suggest that disaggregation is highly
useful to understand how housing supply changed. In addition to motivating disaggregated analysis
of housing supply, we also try to contribute by adding structure to the nature of development.
That is, our channels are largely based on the types of developers we see in practice – the many
smaller-scale developers who build at lower densities, larger scale home builders in the exurbs, as
well as medium- and large-scale builders of for-sale and rental multifamily properties in denser
neighborhoods. This attempt to shed light on an often used “black box” has helped us understand
some of the local dynamics. As a descriptive paper, it is beyond the scope of this effort to develop
94

and test formal theses about the changing supply. That said, we have attempted to learn something
more about changing housing supply by mapping the changes and comparing them to several
natural candidates. In this section, we only sketch these theses – leaving future work to those with
better data.
At the heart of our theses is a highly-adaptable building industry that will construct new housing
whenever there is an acceptable risk-adjusted return. The question then becomes why developers
are finding infill development a better risk-adjusted return than in the past. We suspect that changing spatial amenities, changing income distributions, and changing preferences all play a role in
change demand for location. We further believe that these fundamentals have implications for the
changing composition of building technologies and average unit costs; while higher-density housing development economized on expensive land, infill development is generally more costly on a
per unit than low-density, simple wood-frame construction.
The basic urban model employs distance to the CBD as the sole spatial amenity, which drives
– depending on the model – land rents, housing density, household sorting, and among many more
results, the basic organization land uses. But most employment in the MSAs in our sample lies
outside their CBDs and the vast majorities of trips from home are not to work but rather trips to
visit other spatial amenities. We suggest that the willingness to pay for more spatial amenities –
either because of rising incomes, stronger preference for better retail opportunities, and/or perhaps
better amenities themselves could explain why developers are finding better risk-adjusted returns in
infill locations over the period from 2010 to 2020. We lack detailed data in either cross-section or
time series to undertake a rigorous analysis, but can still explore the thesis using the data available.
To test informally whether housing unit growth might be related to access to consumer amenities, we assign each metropolitan area’s block measures of access to employment and a proxy for
local consumption amenities. Access to employment is an obvious and key factor for demand for
housing at any location. By extension, the demand for access to other spatial amenities is relevant
95

for rent and whether a new unit could be profitable given land prices. By the same logic, we simply
calculate access to non-chain restaurants as our proxy for access to consumption goods. 6
This proxy is purely cross-sectional and based on 2022 amenity locations reported in OpenStreetMap. While decidedly incomplete, Figures 4.11 and 4.12 illustrate how – qualitatively –
the map of non-chain restaurants proxies adequately for clusters of higher-end retail. Figure 4.11
betrays the ubiquity of fast food restaurants. Everywhere there are people, there are fast food esFigure 4.11: Map of Fast Food Restaurants in the Los Angeles MSA
tablishments. Given their pervasiveness, it seems unlikely that someone would pay a premium on
the basis of access to fast food locations.
By contrast, Figure 4.12 plots the non-chain restaurants. Clusters in and around downtown Los
Angeles are apparent, but so too are the clusters that span the higher-end and densely developed
neighborhoods from Silver Lake to Santa Monica. Pasadena, Warner Center, Long Beach, Newport
6Restaurant locations are taken from OpenStreetMap, querying OpenStreetMap using the OSMnx Python package
(Boeing, 2017). We separate restaurants into chain- and non-chain restaurants by cross-referencing their names against
a list of the 250 largest restaurant chains in the United States (Foodservice Database Company, 2022).
96

Figure 4.12: Map of Non-Chain Restaurants in the Los Angeles MSA
Beach, and Dana Point are all visible in the figure. All of these locations are associated with
abundant retail and higher housing prices.
Combined with the access to employment, we can then plot the new net housing supply from
2010 to 2020. Figure 4.13 shows the highly accessible tracts in red (those tracts that are in the top
five percentiles in terms of 30-minute assess to jobs) and yellow (those tracts that are in the top five
percentiles of in terms of 15-minute access to non-chain restaurants). The dots reflect the new net
housing supply along two dimensions. The size of the dots indicate the number of new units, while
the color of the dots shades to darker as the age of the housing stock trends younger. In this way,
large, dark dots are large-scale developments of new housing. This outcome is generally found at
the far periphery of the metropolitan area where large parcels of open land can be found.
The map shows several regularities. First, access to employment and access to consumption
are not perfectly correlated. Second, the vast majority of the tracts have neither short commutes to
ample employment nor access to a choice of non-chain restaurants. And in these locations, there
is very little new supply being built. Scattered across the map are some tracts with non-trivial
97

Figure 4.13: Map of Access & New Net Supply – Los Angeles MSA
new supply, but the larger dots are found at the periphery – via our exurban channel, found in the
post-industrial locations in the core, or found close to consumption opportunities.
The next several figures repeat this mapping exercise for a representative set of MSAs: Phoenix,
Austin, Chicago, and Detroit. The first two are among the fastest growing metropolitan areas over
the last 10 years. Phoenix is much larger and faces different constraints, while Austin is still
relative small, with plentiful land proximal to the CBD. Chicago and Detroit are large and mature
MSAs, but face different challenges.
Figure 4.14 maps the measures of access for the Phoenix MSA. As expected, the rapidly growing MSA has accommodated the new supply via the exurban channel. The large, dark dots are
found consistently at the far periphery and represent largely new stock. But the secondary tracts
of new supply – the intermediary-size dots are heavily clustered in the core, in Tempe, and in
Scottsdale. These are the among the more expensive housing locations and contain – or are proximal to tracts with multiple non-chain restaurants. The suburban tracts with neither good access to
employment or consumption saw little new housing supply built during this decade.
98

Figure 4.14: Map of Access & New Net Supply – Phoenix MSA
While Phoenix has been growing faster than average, Austin has grown the fastest in terms of
rates. It remains small relative to most of our sample MSAs. It began 2010 ranked 36th in the
country, but reached 26th by the end of the decade. Figure 4.15 maps Austin’s housing growth
during this decade. Again, given the easy access to greenfield development, the large majority of
new housing development occurs in the periphery. But it is also clear that a lot of development
has occurred in more infill locations. Indeed, during this decade Austin was a primary market
for apartment development – despite the ready access to land that is still close to amenities in the
downtown. While this dynamic likely has many factors, as least part of this narrative has been
supported by renters who want to live within walking distance to spatial amenities (H. Lee, 2020;
Moos, 2016).
The remaining example MSAs are much larger, with long histories of their periods of rapid
growth well behind them. While the population aggregates are near flat, the three maps suggest
significant internal changes in housing development and new supply. Figure 4.16 shows a map
of the central area of the Chicago MSA. While not visible in this zoomed-in map, there is some
99

Figure 4.15: Map of Access & New Net Supply – Austin MSA
development at the periphery. The more interesting dynamic in Chicago is the return to the core.
The figure shows an astonishing concentration of new net units located among both employment
and consumption. While there are many location in the region that offer ready access to employment, the single largest concentration of retail amenities in the region is the downtown and near
north. The larger dots on the map are comprised largely of our two infill channels – the urban and
post-industrial channels. The dots to the north of downtown Chicago are largely urban infill sites,
while a significant number of tracts to the south represent post-industrial channel development.
Among the most surprising results of our disaggregation is found in Detroit. In 1958, Detroit
was, perhaps, the most productive city on the planet, building almost 60 percent of all cars built
100

Figure 4.16: Map of Access & New Net Supply – Chicago MSA
in the world that year. In the years leading up to that point, job growth had been high, incomes
had risen, and the supply of housing had grown dramatically. But as the car industry decamped
to non-union locations in the decades that followed, the local economy suffered. Despite this, the
metropolitan area population remained largely constant: in 1960 it was 3.48 million people, and
by 2021 it was 3.52 million people. The large negative shock to the region’s dominant industry
and the apparent stability of its population coexist only as a function of the aggregated geography
of the analysis. Within the MSA, the city of Detroit’s population fell from 1.85 million in 1950 to
632,000 in 2021. The flight to the suburbs produced and continues to produce a large increase in
new housing supply on the exurban fringe – even as thousands of housing units remained vacant in
the core. This does not mean that no development is occurring at the core. Indeed, a small handful
of block groups – mostly those with good access to urban consumer amenities – are staving off
decline, if not producing substantial numbers of new housing units. These dynamics are apparent
101

Figure 4.17: Map of Access & New Net Supply – Detroit MSA
in Figure 4.17. Despite the flat aggregate statistics, there is new housing development both in the
far periphery and in the very core of Detroit. And where most of the tracts within the City of
Detroit show no new net units, there is a clear cluster of new development – near both jobs and
consumption opportunities.
4.6 Discussion and Conclusion
It is always noteworthy whenever house prices rise well above inflation. But where previous price
increases have made news, they generally fell and prices returned to some long-term path. This past
102

decade has been remarkable because the prices increases have been large, sustained, and pervasive.
There are many questions to be asked to understand these dynamics, but in this paper, we focus
on housing supply. In particular, we try to document and understand how housing supply evolved
over a 20 year period. We – like many – suspected that housing supply behaved differently over
this housing cycle than in the past. This paper documents that, for 52 MSAs in our study, housing
supply has indeed behaved differently.
Our paper is largely descriptive. The patterns we find became apparent due to an informal
taxonomy of the Census tracts in our 52 metropolitan areas. Each tract is assigned a channel based
on the housing density at the beginning of the decade, and – in the case – of the post-industrial
channel, Census tracts must have contained manufacturing jobs at the very start of our sample
period. While it is common to take a rather generic, “black box” view of housing development, the
reality is that there are broad – and sometimes overlapping – firms that build housing at different
densities and using different building technologies. These are the basis for our channels. And by
defining each tract in this way, we could test whether development really could be approached with
an aggregate lens in mind. And, our four-channels allowed us to see quite varied ways in which
different MSAs accommodated growth.
The heterogeneity across MSAs is remarkable, but around the variation there are several broad
patterns. First, there has been a pronounced change in the shares of new units built at the periphery. These must largely be single-family detached homes. This is the low-cost way to produce
housing – the per square foot of construction costs for “stick” units is half that of per square foot
of higher-density multifamily construction. We offer that this pivot from the low-cost technology
to a more costly construction technology implies that rents/prices much be higher to generate new
development. But to command higher rents/prices, tenants and owners will do so only when local
amenities can be capitalized into location. Our second broad patter is that infill development is
more likely to occur around the amenities we tested – jobs and non-chain restaurants. These are
far from a complete set of spatial amenities. But, as incomplete as they are, they seemed to help
explain some of the new infill units we see in our data.
103

Together, these dynamics begin to raise issues around what gets built and where. At the heart
of the broad pattern of higher housing prices is a falling supply response. But aggregated data
cannot identify these dynamics, and so cannot be used to tease out their independent contributions
to higher prices. Our primary contributions are the set of results that show the benefits of allowing
more local supply being considered. We also feel that providing more structure to the nature of
housing development is also a contribution. While it is not novel, we feel that much of why house
prices have risen at rates well above inflation will have their roots in the specific industrial structure
of home building, not a generic, “black box.” While we hope these efforts will be helpful, the
promise of disaggregation was only introduced in our preliminary analysis of amenities and how
developers choose where to invest and build new housing supply.
It was clear that the choices developers made changed significantly between 2000-10 and 2010-
20. If we are to understand what caused developers to make these choices, we will need to study
housing supply at a much lower level of geography. And if we are to inform policy makers as to
how they might induce developers to build more – and different types of – housing, we will need
to understand the choice developers make.
104

Chapter 5
Online Housing Listings and Preferences
Abstract
Over the past decades, online housing listings have become the primary way of marketing homes
to buyers. Realtors and sellers write housing listings to appeal to the type of buyer they expect to
be most interested in their property. In their free text sections, listings draw attention to features of
both the homes themselves, as well as to those of its surrounding areas, such as nearby amenities
or infrastructure. Listings have become the object of scholarly research as a form of so called “Big
Data”, due to their coverage of market segments not recorded otherwise, or the ability to reflect
features not covered otherwise. Recent studies have demonstrated text in housing listings can be
employed for tracking neighborhood dynamics over the course of decades, or changes in consumer
preferences for different housing attributes during the Covid-19 pandemic.
However, no study to date has employed online housing to evaluate how the marketing of
residential properties reflects legal changes in what an owner can do with a lot. By applying Natural
Language Processing methods to a monthly panel of free text sections of online housing listings
in Los Angeles and Orange counties, California spanning from April 2020 until June 2024, I track
references to accessory dwelling units (“ADUs”), as well as to new two state laws that liberalized
zoning – SB9 and SB10, both of which entered law during the observation period. I find that within
months, housing listings change to reflect the legal possibilities offered by these legal changes. In
particular, the share of listings referencing ADUs increases substantially; however, adoption is not
105

even across space, and many references merely refer to the ability to construct an ADU on a given
lot, rather than to the presence of an ADU.
As a secondary object of study, I track references to new public transit infrastructure within
the same data to demonstrate that references to transit infrastructure are responsive to changes in
actual nearby infrastructure.
106

5.1 Introduction
As has been the case with many other aspects of modern life, searching for housing has moved
from an offline, broker-driven process to taking place almost entirely online over the past two
decades: As of 2023, two in five recent home buyers begin their search for a home by searching for
properties online on web sites such as Zillow, Redfin, or Apartments.com, and essentially all recent
home buyers used the internet at some point in their search (National Association of Realtors, 2023;
Sparber, 2023. This shift has improved search efficiency for all parties (Kroft and Pope, 2014),
and also allowed sellers to capture additional value for their properties (Sing and Zou, 2024).
Such housing listings also present an opportunity for policy researchers and other academics, as
they can serve as a window through which they can directly observe how residential properties - and
specific attributes of them - are being marketed. Text inside housing listings is not entirely novel
as a data source: Various forms of free-text remarks have been a part of housing listings for a long
time, including but not limited to listings in newspapers or on physical noticeboards. Accordingly,
they have been studied by researchers for decades - for example, Chambers, 1984 uses Torontoarea newspaper rental ads to create a house price index for the early 20 th century. However, the
shift toward promoting housing units online and the advent of web scraping techniques and APIs
has made it easier to access information on how houses are promoted at metropolitan or even
national scales, allowing for them to be used in entirely new forms of scientific inquiry and policy
research such as building price indices, or observing changes in consumer preferences (Boeing and
Waddell, 2017; J. Lee and Lee, 2023; X. Wang, Li, and Wu, 2020).
This paper asks whether marketing materials for residential properties - in the form of online
housing listings - can be used by researchers and policymakers to track responses to and uptake of
policies that affect residential property markets. Specifically, I use information from the free-text
portion of housing listings collected in Los Angeles and Orange Counties between April 2020 and
June 2024 to track responses to changes in land use laws, responses to changes in neighborhood
amenities, as well as coronavirus pandemic-era housing market trends previously documented in J.
Lee and Lee, 2023. I look at responses to California’s Senate Bill 9 (”SB9”) and related laws that
107

relaxed restrictions on constructing Accessory Dwelling Units (”ADUs”), as well as to new rail
transit stops that opened during my observation period. I observe a substantial increase over time
in references to ADUs - albeit one that is somewhat concentrated in certain submarkets defined
along both demographic and geographic lines. By contrast, legal novelties from SB9 and new
public transit infrastructure are seldom referenced in marketing material for for-sale residential
properties.
Via this analysis, this paper presents a novel way of applying basic text analysis and exploratory
data analysis to track prospective homebuyers’ reception of policy changes or of new infrastructure.
This method may allow researchers and policymakers insights into preferences on a shorter time
frame than is possible with other data sources, rendering it useful for policy dashboarding purposes:
Given that real estate agents appear to be sufficiently in tune with consumer preferences to respond
to shifts in such preferences, frequent analysis of housing listings may allow a timely glimpse into
such trends.
This paper is organized as follows: Section 5.2 reviews the academic literature on the use of
online housing listings in urban studies research, while Section 5.3 describes the online listings
landscape and theorizes why listings language may change over time. Section 5.4 describes the
study area and specific interventions studied; 5.5 describes the data sources employed, and Section
5.6 outlines the analyses performed on said data. Finally, Section 5.7 presents findings, and Section
5.8 discusses them and concludes.
5.2 Literature Review
With the ever-growing amount of information on the internet and much of it concerning life in
cities, Natural Language Processing (”NLP”) as a method is beginning to gain traction in urbanist
research (Fu, 2024; A. M. Kim and Kang, 2023; Schweitzer, 2014), often cited as one of the
various forms of ”Big Data” research. Recently, a new stream of literature has applied natural
language processing and large language models (”LLMs”) such as ChatGPT to understand topics
108

related to land use and housing policy: Mleczko and Desmond, 2023 and Bartik, Gupta, and Milo,
2023 apply such methods to zoning codes to create national zoning indices, while Blanco and
Song, 2024 evaluate how online rental housing listings respond to changes in non-discrimination
laws.
Online housing listings and other free-text portions of listings have so far played a relatively
minor if growing role in the research on housing markets and demand for amenities. While they
do not offer the same institutional guarantees of data quality or the same historical depth of data,
online housing listings processed using natural language processing methods can offer insights into
sectors of the housing market that are either not captured by other sources, or where such sources
are not available to researchers (Annamoradnejad, Annamoradnejad, Safarrad, and Habibi, 2019;
Boeing and Waddell, 2017; Harten, Kim, and Brazier, 2021; X. Wang, Li, and Wu, 2020). Parsing
listings also allows for featurization - that is, capturing housing attributes seldom recorded in other
data sources by scanning text for key words or phrases (Foster, Liberman, and Stine, 2013; Nowak
and Smith, 2017; Su, He, Sun, Zhang, Hu, and Kang, 2021). Finally, housing listings may contain
references to place names (Kordjamshidi, Van Otterlo, and Moens, 2011; Leidner and Lieberman,
2011), contain hints of who the target buyer or renter is that they are attempting to speak to (L.
Wang, He, Su, Li, Hu, and Li, 2022); or of who they are explicitly seeking to exclude (Adu and
Delmelle, 2022; Blanco and Song, 2024; Oliveri, 2009).
Besides their spatial and topical coverage, another desirable feature of online housing listings -
and advantage over traditional housing market data sources - is the timing of their availability: At
any given point, housing listings reflect market conditions of that point in time, while traditional
sources’ publication schedules and inherent publication lag render it impossible to draw conclusions about present market conditions (Blanco and Song, 2024; Boeing and Waddell, 2017; Chen,
Liu, Li, Liu, and Xu, 2016).
When evaluating the text portion of housing listings, it is important to consider who writes the
listings, and for what audience (Boeing, 2020b). Similarly, listings may not be distributed evenly
across space or across the socioeconomic spectrum (Besbris, Schachter, and Kuk, 2021); an issue
109

that can be addressed with appropriate weighting schemes (Lopez Ochoa, 2023). Further, listings’
veracity can sometimes a challenge, as listings can act as a way to get potential buyers/tenants’
attention and establish contact (Harten, Kim, and Brazier, 2021).1
Collected over a sufficient time frame, the text contained in housing listings allows for tracking
changes over time. Nillson and Delmelle use MLS comments to observe how neighborhoods
change demographically and socioeconomically over time (Nilsson and Delmelle, 2022), as well
as to observe housing over the course of its life cycle (Nilsson and Delmelle, 2023). J. Lee and
Lee, 2023 document how over the course of the Covid-19 pandemic, real estate agents shifted
their focus from location - such as calling out proximity to public transport stops or to the central
business district - to features of the unit itself, such as which floor it is on or whether it has a
view. Listings can also be used to track how housing policies are adopted: Blanco and Song, 2024
document both the extent of discrimination against housing vouchers in online rental listings, and
how quickly new laws against source-of-income discrimination are reflected in the language of
listings.
This paper contributes to this literature that applies natural language processing methods to
housing listings over a longer time frame by demonstrating that housing listings not only mirror
the preferences of prospective buyers, but also change rapidly to reflect non-structural changes to
the properties themselves. Specifically, I demonstrate that listings reflect changes in properties’
legal characteristics - such as when land use laws become more permissive in how the land can be
used, and in the properties’ neighborhood amenities - such as when new infrastructure is built in
their vicinity; and thus afford academics and policymakers a glimpse into preferences with regards
to those attributes.
1Harten, Kim, and Brazier, 2021 document this problem in listings for bed spaces in Shanghai; however, the issue
of imprecise listings used to capture potential buyers/tenants’ attention may also be present in other markets.
110

5.3 The Online Housing Listing landscape
Online listings have become the dominant way of marketing real estate: As of today, virtually all
home buyers consult listings web sites as part of their home buying process (Sparber, 2023). A
variety of listings web sites compete for prospective home buyers’ attention. Industry giants such
as Zillow, Realtor.com, Redfin, and Trulia account for by far the most web traffic; the top ten home
search sites alone account for a market share of 98 percent (Fisher, 2023). Nonetheless, virtually all
major listings web sites advertise the same set of housing listings, sourced from Multiple Listings
Services (National Association of Realtors, 2012). While the market for online rental housing
listings is more fragmented across different players than that for for-sale properties, each of which
caters to a particular niche of the market, it substantially moved online, albeit not to the same extent
as is the case with the market for for-sale properties (Besbris, Schachter, and Kuk, 2021; Boeing,
2020b; Costa, Sass, Kennedy, Roy, Walter, Acolin, et al., 2021).2
Regardless of the exact venue in which they are advertised online, the process by which online
housing listings come into being is substantially the same: A seller - or a sellers’ agent - posts
information about the property they are listing to their local Multiple Listings Service, alongside
information on how to contact them if one is interested in buying the property. Such information
varies between property types, sellers agents, and submarkets, but usually consists of basics such
as the list price, address, and various measures of the size of the property such as its lot size,
square footage, or the number of bedrooms and bathrooms. Generally, such information are based
on information from county assessor offices. However, agents have more flexibility with regards to
other portions of the listing: Often, sellers agents also include a short text describing the property
and its location, as well as photos or floor plans, in the hopes of capturing prospective buyers’
attention. 3 While - at least in the United States - contents are subject to regulations such as
2As an example for an online housing listings website targeting a particular niche audience, GoSection8 lists rental
units available to housing voucher recipients. Compared to mainstream listings web sites, the distribution of listings
on GoSection8 skews towards inexpensive units in higher-poverty neighborhoods (Hess, Walter, Kennedy, Acolin,
Ramiller, and Crowder, 2023).
3More recently, virtual property tours based on 360-degree photography have joined the range marketing materials
sellers may use to promote their properties online (Xiong, Cheung, Levy, and Allen, 2024; Yu, Ma, Pant, and Hu,
2021), however, this remains a niche phenomenon.
111

the Fair Housing Act, sellers and their agents enjoy substantial artistic liberty in formulating text
portions of listings (Oliveri, 2009).
The set of features or amenities that listings draw attention to may change over time for a variety of different reasons: First, there may be population-wide (or local) changes in preferences.
Examples for this include the pandemic-induced decline in preference for proximity to public transit or to central business districts documented by J. Lee and Lee, 2023. Second, the language of
listings may reflect changes in the prospective buyer population for a pool of properties. Sellers and
their agents may choose to draw attention to attributes and amenities that appeal to whoever they
anticipate is likely to bid the most for their property, rather than reflect the preferences of the incumbent population. We would expect such demographically-induced changes in listing language
to be most common in places undergoing rapid neighborhood changes such as gentrification or an
influx of foreign buyers. Finally, listings may reflect changes in what exactly prospective buyers or
renters are looking at, or the terms of the transaction. Rules concerning rentals can change, as can
land use regulations, or at a more local scale, neighborhood amenities, such as when new infrastructure is constructed. In all of these instances, listing language may evolve for reasons other than
shifts in consumer preferences or in the pool of buyers - but only if sellers or their agents believe
that drawing attention to this change is helpful to their cause given consumer preferences.4
5.4 Study Context
5.4.1 Study Area: Los Angeles and Orange Counties
The primary area of my study is Los Angeles and Orange Counties, California. I situate my study
in this particular area for three reasons: First, because it experienced several changes in land use
laws during the study period - some of which are statewide, others are locally specific changes -
each of which makes it easier for property owners to construct additional housing units such as
4Another situation in which real estate agents may adapt listing language in response to a change in regulations is
if it is legally required of them, as is the case with the nondiscrimination rules documented by Blanco and Song, 2024.
112

Accessory Dwelling Units on their property. Second, because the region experienced substantial
investments into new public transit infrastructure during the study period. Finally, I choose to study
Greater Los Angeles due to personal familiarity with the region.
5.4.2 New Legal Possibilities: ADUs and Upzoning
During the time period studied, there were several changes to land use laws that facilitate infill
development of new housing units, effectively expanding the options of property owners for what
they can do with their land.
Accessory Dwelling Units - sometimes referred to as ”ADUs,” ”granny flats,” or ”in-law units”
- are secondary, smaller housing units placed on the same parcel of land as another - ”primary”
- housing unit. While ADUs have existed in California in some form for over 40 years, several
reforms enacted between 2016 and 2020 limited municipalities’ ability to restrict their construction, effectively making their construction by-right statewide, thus dramatically speeding up and
simplifying approvals (Marantz, Elmendorf, and Kim, 2023a). As a result, construction of ADUs
has boomed throughout the state, including in Los Angeles (Brueckner and Thomaz, 2024). 5
Senate Bills 9 and 10 (”SB9” and ”SB10”) are two California statewide housing bills that were
signed into law by Governor Newsom in September 2021 as part of a broader effort to combat a
statewide housing shortage, entering effect on January 1st, 2022. SB9 allows for splitting most
existing single-family lots into two lots or for the construction of a duplex on the land previously
occupied by a single-family home, while SB10 allows even greater numbers of housing units to
be built on previously single-family lots if they are located in areas near public transit or that are
otherwise prioritized for urban infill development (State of California, 2021). While SB9 and SB10
are just two of many housing bills passed during this time, they are of particular interest because
they offer fairly straight-forward redevelopment options to property owners.
5Brueckner and Thomaz, 2024 evaluate in what kinds of places within Los Angeles ADUs have been particularly
successful, finding that they are particularly common in Latino areas. This is an observation I am able to corroborate
with the approach presented in this paper.
113

In addition to the aforementioned statewide programs and housing bills, there are local incentive programs for development. Two prominent examples for these are the City of Los Angeles’ ”Transit-Oriented Communities” (”TOC”) - available since 2016 - and Executive Directive 1
(”ED1”), which was enacted by Mayor Karen Bass in late 2022. These programs offer relaxed
zoning rules and streamlined approval processes in exchange for projects meeting certain criteria:
In the case of TOC, developers can build larger buildings than otherwise allowed if a certain share
of units are deed-restricted to lower income tenants and the project is within a set distance of public
transit services; in the case of ED1 developers are granted by-right approval so long as 100% of
units in the building are deed-restricted to lower-income tenants (”100% affordable”) (Los Angeles
City Planning, 2024a, 2024b).
5.4.3 New Amenities: Public Transit
While known mostly for its extensive freeway network, Greater Los Angeles is served by several
rail transit lines ranging from light rails and subways to heavy commuter rail - and contains a wide
variety of public transit service levels across different areas in the metropolitan area. The study area
contains several public transit projects that entered service during the study period, or will enter
service in the next years: Initial portions of the the Crenshaw Light Rail Line (”K-Line”) entered
service in October 2022, and the Downtown Los Angeles Regional Connector entered service in
June of 2023. Further, the Orange County Streetcar and Purple Line Subway Extensions were
actively under construction during the study period (Los Angeles Metro, 2020a, 2020b; Orange
County Transportation Authority, 2020). Several additional projects are on the horizon for the
more distant future, including an expansion of the Green Line Light Rail beyond Norwalk, or
a Light Rail in the Eastern San Fernando Valley. As such, this Southern California study area
lends itself for studying how listings refer to both existing transit infrastructure, as well as for new
infrastructure that remains in planning or is under construction.
114

5.5 Data Collection and Processing
In this particular paper, I rely on data collected from MLSlistings.com, a website with for-sale
housing listings in California. 6 The web site contains listings for all types of properties, listed
through Multiple Listing Services (MLS), ranging from condominiums and townhomes to single
family homes and multifamily investment properties, to vacant land. 7 The reason I choose MLSListings.com over web sites with a larger market share such as Zillow or Redfin is that unlike
those web sites, MLSListings.com does not deter web scraping via use of captchas or similar antibot protections.
I obtained listings via web scraping, using a program8
that looks up what listings are available
for each ZIP code within the study area, and then visits and saves the web page for each listing.
This program was run on a near-monthly basis from April 2020 until June of 2024, 9
for the same
set of Los Angeles and Orange County ZIP codes each time. 10
After collecting listings, I parse free text components such as the list price, property description,
property characteristics (such as the square footage or the number of bedrooms and bathrooms of
a given listing), and listing characteristics (such as the name and license of the agent who posted
the listing) out of the saved copy of each listing. In addition to the property address displayed to
a visitor, listings on MLSListings.com generally contain the latitude and longitude for each listing
within each listing page’s metadata, allowing for spatial analyses based on precise coordinates as
well as for joining with data from explicitly spatial sources such as Census data.
6While MLSlistings.com claims to specialize in Northern California, the website reliably lists enouogh properties
in Southern California to study Southern California markets.
7For more information on Multiple Listing Services, see National Association of Realtors, 2012.
8Scraping programs employed in this study are written in the R programming language, using the rvest and RSelenium software packages.
9Near-monthly due to the highly manual nature of running the required programs and the amount of supervision
required.
10Since ZIP codes do not perfectly align with counties, this approach also collected a small handful of listings in
neighboring counties.
115

Due to the large variety of listing types, I group listings into four types of properties: SingleFamily, Multi-Family, Condominiums and Townhouses, Land, and Manufactured Homes.11
Table 5.1: Observation Counts by Property Type
Property Type Observations Listings Unique Texts
Single-Family Residences 258,402 135,604 159,512
Multi-Family Residences 63,885 22,501 28,390
Condominiums and Townhouses 94,977 52,864 61,462
Land 107,360 19,666 24,363
Manufactured Home 22,687 9,138 11,359
The resulting data consist of 547,311 unique listing-date observations, across 238,737 unique
listings and 47 different dates of observation. Sample sizes by property type is displayed in Table
5.1, and sample sizes by obseration date are displayed in Table 5.2.12 Some listings remain online
for long enough to be observed in more than one month. I include those listings in each month
in which they are observed, as to generate statistics that cross-sectionally reflect the market as
observed on MLSListings.com at any given time of observation. 13
The region-wide spatial distribution of properties by property type is displayed in Maps 5.1,
5.2, 5.3, 5.4, and 5.5. The patterns in these maps essentially follow land use patterns of the greater
region: Single-Family properties (Map 5.1) exist everywhere outside of central business districts
and industrial areas, while Multifamily properties (Map 5.3), or condominiums (Map 5.2) are
concentrated in central areas. 14 Manufactured homes (shown in Map 5.4) are highly clustered in
a small handful of trailer parks - most of which are relatively far from regional centers. Finally,
11Specifically, I collapse listings of the types ”Cabin,” ”Single Family Residence,” and ”Residential” into SingleFamily; ”Duplex,” ”Triplex,” ”Quadruplex,” and ”Residential Income” into Multi-Family; ”Townhouse,” ”Condominium,” and ”Stock Cooperative” into Condominiums and Townhomes, ”Manufactured Home” and ”Manufactured
In Park” into Manufactured Homes, and ”Land” into Land. This categorization is based upon manual review of a
sample of a handful of listings of each type, and while not perfect, helps group listings into internally similar groups. I
discard a small number of listings of niche types such as ”Timeshare,” ”Business Opportunity,” or ”Commercial Sale”
as they are too different from other categories and do not have sufficient sample size to warrant a group of their own.
12A change in the structure of search results page for each ZIP code may have resulted in missing some listings in
late 2021/early 2022, resulting in smaller sample sizes for those months.
13Note that the free text portion of listings can change over the duration of time that the listing is online. Over the
238,737 unique listings observed in this study, I observe 284,801 unique combinations of a listing identifier and free
text section, suggesting that making edits to the listing text is a common practice.
14The difference in spatial patterns between condominiums/townhomes and multifamily properties appears to correlate with wealth.
116

Table 5.2: Sample Sizes by Date and Property Type
Observation Date Single-Family Multi-Family Condominiums Land Manufactured Total
Residences Residences Homes
2020-04-20 6, 153 1, 038 2, 263 1, 144 594 11, 192
2020-05-13 6, 196 1, 068 2, 355 1, 133 558 11, 310
2020-06-15 6, 479 1, 081 2, 603 1, 095 582 11, 840
2020-07-13 5, 991 1, 121 2, 581 1, 055 580 11, 328
2020-08-13 5, 499 1, 202 2, 476 1, 044 578 10, 799
2020-09-14 5, 815 1, 312 2, 751 1, 184 557 11, 619
2020-10-12 6, 549 1, 408 2, 916 1, 383 581 12, 837
2020-11-13 6, 508 1, 444 2, 896 1, 639 604 13, 091
2020-12-15 5, 458 1, 410 2, 491 1, 831 601 11, 791
2021-02-08 4, 460 1, 235 1, 992 2, 003 585 10, 275
2021-03-15 4, 222 1, 115 1, 706 2, 094 550 9, 687
2021-04-12 4, 667 1, 168 1, 791 2, 201 525 10, 352
2021-05-09 4, 748 1, 216 1, 681 2, 317 494 10, 456
2021-08-09 5, 624 1, 427 2, 001 2, 595 462 12, 109
2021-09-09 5, 272 1, 390 1, 887 2, 612 425 11, 586
2021-10-11 5, 003 1, 319 1, 746 2, 521 413 11, 002
2021-11-08 4, 513 1, 179 1, 519 2, 475 394 10, 080
2021-12-08 3, 143 984 1, 180 859 317 6, 483
2022-01-10 2, 617 852 940 921 310 5, 640
2022-02-07 2, 843 866 1, 067 888 313 5, 977
2022-03-07 3, 103 900 1, 087 869 306 6, 265
2022-04-11 3, 531 936 1, 105 823 275 6, 670
2022-05-09 5, 261 1, 281 1, 558 2, 663 338 11, 101
2022-06-13 6, 793 1, 393 2, 036 2, 758 345 13, 325
2022-07-12 8, 499 1, 560 2, 625 2, 842 401 15, 927
2022-08-08 8, 901 1, 550 2, 780 2, 831 441 16, 503
2022-09-13 8, 146 1, 560 2, 543 2, 872 456 15, 577
2022-09-26 8, 410 1, 590 2, 604 3, 005 462 16, 071
2022-10-10 8, 442 1, 606 2, 660 2, 960 465 16, 133
2022-11-07 8, 454 1, 634 2, 740 3, 009 506 16, 343
2023-02-06 5, 509 1, 310 1, 955 2, 764 464 12, 002
2023-03-08 4, 860 1, 357 1, 778 2, 779 492 11, 266
2023-04-03 4, 519 1, 365 1, 710 1, 954 495 10, 043
2023-05-09 4, 571 1, 410 1, 576 2, 498 498 10, 553
2023-06-05 4, 906 1, 454 1, 626 3, 060 509 11, 555
2023-07-15 5, 241 1, 523 1, 817 3, 152 506 12, 239
2023-08-07 5, 896 1, 766 2, 065 3, 319 561 13, 607
2023-09-05 5, 381 1, 568 1, 893 3, 169 517 12, 528
2023-10-06 5, 523 1, 653 1, 988 3, 106 540 12, 810
2023-11-13 5, 940 1, 707 2, 152 3, 261 552 13, 612
2023-12-04 5, 359 1, 580 1, 982 2, 521 517 11, 959
2024-01-11 4, 617 1, 424 1, 773 2, 957 518 11, 289
2024-02-04 4, 504 1, 446 1, 857 3, 005 521 11, 333
2024-03-04 4, 533 1, 521 1, 893 2, 962 498 11, 407
2024-04-01 4, 510 1, 541 1, 855 3, 034 479 11, 419
2024-05-06 5, 245 1, 631 2, 088 3, 007 483 12, 454
2024-06-02 5, 988 1, 784 2, 389 3, 186 519 13, 866
Entire Sample 258, 402 63, 885 94, 977 107, 360 22, 687 547, 311
117

Figure 5.1: Spatial distribution of listings for Single-Family Properties
118

Figure 5.2: Spatial distribution of listings for Condominiums and Townhomes
119

Figure 5.3: Spatial distribution of listings for Multi-Family Properties
120

Figure 5.4: Spatial distribution of listings for Manufactured Homes
121

Figure 5.5: Spatial distribution of listings for Vacant Land
122

while vacant lots (shown in Map 5.5) exist throughout the entire region, many listings for vacant
land are clustered in locations such as the hillsides of Lincoln Heights, or in the Mojave Desert.
I use additional sources in addition to the aforementioned listings: For demographic information, I rely on the US Census Bureau’s American Community Survey (”ACS”). While the ACS
is reported at many spatial levels of aggregation and most urban social science research uses it at
the Census Blockgroup or Census Tract level, I use a different level of aggregation in this study,
namely, I rely on ACS data reported at the Zip Code Tabulation Area (”ZCTA”) level. 15 This
level of aggregation approximates ZIP codes as used by the US Postal Service, and is convenient
for two reasons: First, listings almost always include their postal code as part of the address . Second, listings data are sparse enough for many Census Tracts to only contain one or two listings per
observation date - ZCTAs on the other hand provide a convenient level of spatial resolution that
is substantially more granular than county level, more evenly sized than municipalities, while still
remaining granular enough to capture local variation.
Besides listings and the ACS, I rely on a GIS Shapefile with the locations of rail stations from
which Los Angeles Metro - the main public transit operator in the region - operates (Los Angeles
Metro, 2024).16 I calculate the distances between each listing and the stations contained in this
shapefile to obtain the distance from each listing to its respective nearest Metro station.
5.6 Data Analysis
A variety of NLP approaches for evaluating listings data have emerged over the past years, from
simple tabulations and visualization of common terms17 to fully automated identification of themes
or summarization of text contents. Researchers employ a variety of approaches to identify (”label”)
15I access the ACS via the tigris and tidycensus R packages, as well as via IPUMS NHGIS. All subsequent analyses
are performed in R, using the stargazer and ggplot packages to generate visually appealing outputs.
16Another operator - Metrolink - operates commuter rail in the area. I choose not to include Metrolink stations due
to their infrequent service.
17For example, J. Lee and Lee, 2023 display word clouds showing common terms in Singapore housing listings
both before and after the beginning of the Covid-19 pandemic.
123

listings containing references to particular concepts, ranging from manual review to hybrid supervised deep learning methods (Lore, Harten, and Boeing, 2024) or use of large language models
(”LLMs”) such as ChatGPT (Blanco and Song, 2024).
For the analyses in this paper, I perform three steps: First, I attach labels to listings - that
is, identify whether any given listing discusses a particular concept of interest - using key words
identified via manual review of a subset of listings. Next, I calculate the shares of listings referring
to each concept of interest on any given observation date, for any given property type, for any given
ZCTA, or for combinations of these aggregations. Finally, I generate plots and maps to visualize
trends in the data generated via this approach.
5.6.1 Labeling
The approached employed in this paper for attaching labels to listings is based on keywords identified via manual review of a subset of listings: First, I reviewed several hundred randomly selected
housing listings to develop an understanding of how listings are written in general, taking note of
concepts I encountered and what words were used to refer to them. Next, to identify what words
and phrases are used to discuss specific amenities with clear, known spatial locations, I reviewed
the text of several dozen listings near known amenity locations such as metro stations. Finally,
to identify key words and phrases used to refer to accessory dwelling units and the housing law
changes of interest, I rely largely on the literal names of such policies or their abbreviations (e.g.
”Executive Directive 1” and ”ED1”), as well as words describing what the laws do (e.g. ”lot split”
for Senate Bill 9 in addition to ”Senate Bill 9” and ”SB9”, as the ability to split lots is the perhaps
most salient feature of this law). Based on these manual reviews, I compile lists of key words
relevant to each concept of interest.
Relevant terms for accessory dwelling units identified via this approach are ”accessory dwelling
unit,” ”granny flat,” ”in-law unit,” ”ADU,” and ”A.D.U.”.Terms I track for zoning policy include
”senate bill 9,” ”SB9,” ”lot split,” ”TOC,” ”Transit Oriented Communities,” ”ED1,” ”Executive Directive 1,” and variants thereof. Similarly, relevant terms for identifying references to public transit
124

include ”metro,” ”subway,” ”transit,” ”rail,” ”bus stop,” and ”station”, while terms for identifying
references to highways include ”freeway,” ”highway,” ”Interstate,” and ”I-” followed by one to
three digits. Additional concepts of interest - such as covid-era trends such as remote work - are
flagged via a similar process. 18 Using those lists of keywords, I flag listings containing those
keywords by searching for exact, non-case-sensitive matches for entire words. 19
In an additional analysis, I evaluate whether references to accessory dwelling units are referring
to a real, existing ADU, or referring to the possibility of constructing and ADU. To differentiate
between these two types of references, I flag sentences that both contain a reference to ADUs via
the aforementioned words and words such as ”convert,” ”plan,” or ”potential” that signify that the
ADU is purely hypothetical. 20
5.6.2 Generating Descriptive Statistics
After having attached labels to listings indicating whether they refer to any of the aforementioned
concepts of interest, I calculate what share of particular sub-markets contain references to terms to
generate several forms of descriptive statistics: shares of listings across different property types,
across different dates of observation, or across space at the ZCTA level in the form of maps. I
18The words and phrases I use to flag references to telework are ”telework,” ”telecommute,” ”work from home,”
”working from home,” ”home office,” ”zoom meeting,” and ”remote work”. I flag references to commuting via the
words ”commute” and ”commuting;” and references to virtual tours via the phrases ”virtual tour” and ”3D tour”. I
capture references to countertops via the words ”countertop” and ”backsplash”; references to appliances via the words
”appliances,” ”fridge,” ”stove,” and ”range,” and to a unit being spacious via the word ”spacious”. References to
nearby shopping are captured via the terms ”shops,” ”shopping,” ”retail,” and ”stores”, and references to dining are
captured via the words ”dining,” ”restaurant,” and ”cafe”. Finally, references to schools are captured via the word
”school”.
19While this form of flagging presence of key words is rather simple, more involved NLP methods could be applied
to the same kind of source materials. For excellent examples of more sophisticated parsing of housing listings, see
Blanco and Song, 2024, in which the authors use ChatGPT to identify discussion of housing vouchers; or Lore, Harten,
and Boeing, 2024, who employ a hybrid deep learning approach to flag the presence or absence of concepts in the text
content of listings, training a BERT model with a small manually labeled training dataset.
20To identify a list of words referencing a hypothetical rather than an actual ADU, I reviewed a sample of several
hundred randomly chosen sentences that each contained at least one reference to an ADU. The complete list of words
and phrases I use in this particular analysis to identify references to a hypothetical ADU (as opposed to an existing one)
is ”potential,” ”possible,” ”possibly,” ”possibilities”, ”likely,” ”allow,” ”converting,” ”conversion,” ”convert,” ”could,”
”turn into,” ”ready for,” ”add,” ”build,” and ”permits”. Given the wide variety of vocabulary that could be used to
describe hypothetical nature of an unbuilt ADU, this analysis in particular is one where an LLM-based approach may
perform better than the semi-manual one employed here.
125

also combine listings with demographic information for their surroundings as reported by the ACS
at the census tract level to plot frequencies of concepts against demographic information such as
income. Further, I map spatial differences in changes over time in how frequently a given concept
is mentioned.
5.7 Findings
First, I evaluate how often listings reference common concepts discussed in listings, such as interior features - namely, appliances, countertops, or discussing how spacious a unit is, or proximity
to urban consumer amenities such as shopping, dining, or schools. All of these are property characteristics that are not part of the numeric measurements that are part every listing - such as the
number of bedrooms and bathrooms or the size of a property - but are nonetheless commonly
discussed by sellers or their agents when listing a property.
Table 5.3: References to common attributes
Reference Land Condominiums Single Family Multifamily Manufactured
Countertops 0.1% 39.2% 32.5% 12.4% 21.5%
Appliances 1.5% 47.1% 39.6% 13.4% 31.7%
Spacious 1.6% 44.8% 39.6% 17.9% 35.2%
Public Transit 2.8% 7.0% 4.0% 6.8% 2.8%
Highways 13.0% 10.7% 8.3% 10.2% 9.2%
Shopping 15.9% 41.7% 26.6% 32.3% 31.0%
Dining 10.0% 61.9% 53.1% 30.3% 45.2%
Schools 10.6% 22.6% 25.3% 17.5% 13.1%
As is shown in Table 5.3, such references are very commonplace, though they vary substantially
between different types of listings: Listings for condominiums/townhomes and for single-family
home listings more frequently discuss interior features, nearby dining, or access to schools than
listings for other types of properties - perhaps due to the fact that most prospective buyers for
such properties are likely buying for their own use. On the other hand, listings for vacant land
more frequently note proximity to highways than any other type of listing, perhaps to signal the
potential of their locations.
126

5.7.1 Tracking Changes to References Over Time
In addition to this cross-sectional variation, we find longitudinal signals in real estate listings:
References to both ADUs and to public transit do indeed change over time.
5.7.1.1 Covid-Era Preferences
To demonstrate that the text content of housing listings is responsive to changes in societal context,
I track references to two pandemic-era phenomena: Working from home, and virtual tours - in
addition to references to commuting.
Figure 5.6: References to Covid-Era Phenomena: Telework (dashed), Virtual Tours (dotted), Commuting (solid)
127

As I demonstrate in Figure 5.6, I observe that virtual tours - a relative novelty in 2020 in the
early days of the Covid-19 pandemic - became less of a differentiating factor over the course of
2020, and ceased being called out explicitly by 2022. On the other hand, references to telecommuting, home offices, or otherwise the ability to work from home experienced a first peak in early
2021. After this initial peak, such references declined over the course of 2021 as vaccines against
the coronavirus became available, only to become more common again afterwards as hybrid work
became more commonplace. By contrast, less than one percent of listings in 2020, 2021, or 2022
referenced commuting, with such references gradually increasing to around two percent of all listings over the course of 2023 and 2024.
5.7.1.2 Housing Laws
Figure 5.7 shows references to Accessory Dwelling Units over time, by type of property. We see a
steady increase in references over time, though this increase is almost entirely confined to SingleFamily and Multifamily Residential Properties, among which references increased from around 3
percent of all listings in mid-2020 to over 12 percent by mid-2024. The increase in references to
ADUs is relatively continuous, and does not exhibit any discernible breaks indicative of events.
However, a mention of ”accessory dwelling unit” or similar does not necessarily mean that a
property has an ADU on site at the time of listing. To determine whether such references refer to
an actual, existing ADU or merely discuss the potential for building one, I look at whether sentences with such references also contain words or phrases such as ”potential”, ”could construct”,
or similar. I find that throughout the period of observation, around 50 to 60 percent of references to
ADUs talk about a hypothetical ADU rather than an actual, existing additional unit on the property
(see, Figure 5.9).
References to Senate Bills 9 and 10 or to their effects on the other hand do exhibit a clear event:
As is shown in Figure 5.8, references to either the bills themselves or to the possibility of splitting
a lot spiked around the time SB9 and SB10 entered law at the beginning of 2022, only to decline
afterwards. However, even at their peak, references to these laws are far less common than for
128

Figure 5.7: References to Accessory Dwelling Units by Property Type
129

Figure 5.8: References to Senate Bills 9 and 10 by Property Type
130

Figure 5.9: Share of ADU references that discuss hypothetical vs. actual ADUs
ADUs. More importantly, references to SB9 and SB10 are almost entirely confined to listings for
undeveloped land or for multifamily properties; despite the laws targeting areas zoned for singlefamily residences only a small handful of listings for such properties make note of SB9/10 or the
new possibilities afforded by those laws.
These observations may hint at a challenge to the effectiveness of SB9 - the average person in
the market for a house may not be interested in redevelopment at all, as they are merely shopping
for a residence to live in. Compared to such legal changes, ADUs are likely far more salient a
concept to the average home buyer. Similarly, development incentive programs such as TOC and
ED1 are referenced only in a small fraction of listings (see, Figure 5.10), as they are likely relevant
to only a very small audience of potential buyers.
5.7.1.3 Public Transit
Contrary to what J. Lee and Lee, 2023 observe in Singapore, the share of listings across the entire
sample referencing public transit (displayed in Figure 5.11) is relatively stable over time, with no
obvious pandemic-related trends over the period of observation. This may be due to the fact that
131

Figure 5.10: References to TOC and ED1 over time
132

Figure 5.11: References to Public Transit by Property Type
133

public transit use is relatively low within the study area of this study, at least compared to the
Singaporean context.
5.7.2 Differences across Submarkets and Space
In addition to variation over time, I observe cross-sectional signals: Different submarkets - defined
via area demographics, via property characteristics, or geographically - reference different amenities differently in a manner that is consistent with everything else we know about preferences.
5.7.2.1 Housing Laws
Figure 5.12: References to Accessory Dwelling Units by Lot Size
134

Figure 5.13: References to Accessory Dwelling Units by ZCTA Income
135

Figure 5.14: References to Accessory Dwelling Units by ZCTA Latino Population Share
136

Figure 5.15: References to Accessory Dwelling Units by ZCTA Non-Hispanic White Population
Share
137

Figure 5.16: Listings in 2020 (Orange: References ADUs, Black: Does not reference ADUs)
138

Figure 5.17: Listings in 2024 (Orange: References ADUs, Black: Does not reference ADUs)
139

Figure 5.18: Percentage Point change in references to ADUs, 2020 to 2023
140

Maps 5.16 and 5.17 show the spatial patterns of where listing texts reference ADUs in 2020
and in 2024, respectively. Evaluating where - and for what kind of lots - listings reference ADUs,
we find that references to ADUs are the most common among roughly 1/8-acre sized lots - which
is also approximately the modal lot size in the study area (see, Figure 5.12). By contrast, ads
for larger lots rarely highlight the presence of or the potential for an ADU. References to ADUs
are more common in places with relatively lower median incomes (Figure 5.13), in places with a
high share of Latino or Hispanic residents (Figure 5.14), and a low share of Non-Hispanic White
residents (Figure 5.15).21 Comparing more recent listings to those from 2020 in Map 5.18, we do
not observe signs of a broadening of in what kinds of places and for what kinds of properties online
listings reference ADUs - instead, the same associations observed in 2020 intensified over time.22
Taken together, these observations suggest that ADUs are not a universal selling point for a
property: References to them are concentrated among relatively smaller properties. While larger
residential lots may have greater potential for ADUs in the form of more space, larger lots likely
appeal to a different kind of buyer, who likely is not as interested in building additional housing
units on the same land.
5.7.2.2 Public Transit
While references to public transit remained essentially at the same level throughout the study
period , marked differences exist across different property types: As is seen in Figure 5.11, listings
for condominiums, townhomes, and other multifamily properties are more than twice as likely to
refer to public transit than are listings for single family homes; with listings for vacant land or for
manufactured homes rarely ever making references to public transit. Plotting references to transit
against distance to rail transit stops in Figure 5.19, we see that this may largely be a composition
effect: Conditional on being a certain distance from transit, the share of listings referencing public
transit is comparable between condominiums and single-family residences. Within approximately
21Plots are for Single-Family Residences; however, the same associations hold for Multifamily Residences.
22I plot changes from 2020 to 2023 rather than to 2024 due to the larger sample size stemming from that I have a
full year of observations for 2023.
141

Figure 5.19: References to Public Transit by Property Type and Distance to nearest LA Metro Rail
station
142

Figure 5.20: Share of Sample by distance to nearest LA Metro Rail station
143

a quarter mile of stations, around 20 percent of both single-family and of condominium listings
make reference to the availability of public transit, dropping to around five percent for singlefamily homes and eight percent for condominiums at a distance of a full mile, plateauing around
three to four percent further out. 23 However, the share of all condominiums and of all multifamily
buildings that are located in close proximity to transit stations than is the case for single-family
homes (Figure 5.20).
Mapping references to public transit over time by property type, we see - perhaps unsurprisingly - that listings in places with more public transit service refer to it more often (see, Map 5.21).
However, we again find little evidence for increases over time in references to transit as an attempt
to market residential properties in response to new public transit infrastructure - the overall share
of property listings noting proximity to transit has remained essentially the same over time if not
modestly declined.
5.8 Discussion and Conclusion
In this paper, I explored the language real estate agents use in housing listings to promote properties online. Unsurprisingly, I find that sellers and their agents tailor listings to cater to particular audiences: Listings for single-family or for condominium properties emphasize attributes and
amenities that would-be residents are likely to use themselves, while listings for vacant land or
for multifamily properties often highlight the potential for return on investment, for cash flows,
or for development potential. Further, I demonstrate that online housing listings are responsive to
changes to their neighborhoods, and to the legal and social context in which the properties exist.
It is important to remember that the information captured in online housing listings is merely
one representation of the housing market, rather than a snapshot of the market itself. Several
challenges exist in the interpretation of such information: Listings can be aspirational - not only
in the list prices sellers attempt to achieve, but also in the content in descriptions. For example,
23While this is not the primary purpose of this paper, the gradients depicted in Figure 5.19 also contribute to a
literature on how far stations’ effects reach (Guerra, Cervero, and Tischler, 2012; Li, Yang, Zhang, Ling, Xiong, and
Li, 2019).
144

Figure 5.21: Spatial distribution of references to public transit
145

listings for vacant land in the Antelope Valley is frequently advertised as close to Los Angeles
based on straight-line distances or screenshots of Google Maps directions during off-peak hours;
other listings may highlight proximity to more desirable nearby neighborhoods rather than drawing attention to their actual surroundings. Compared to the property characteristics taken from
institutional data sources such as a property’s size, the content of online housing listings’ free text
portions is not subject to any such quality control. Indeed, sellers and their agents frequently include the phrase ”buyer to verify” - or variants thereof - in listings to insulate themselves from legal
liability should claims - both on measurable attributes such as unit sizes, and on legal possibilities
such as the possibility of building an ADU - turn out not to be entirely accurate.
Interpreting observations as reflecting consumer preferences requires some caution: While we
assume that listings are tailored to the tastes of whoever the sellers’ agent assumes is willing to
pay the most for a given property, we are relying on that the agents’ assumptions are correct: To
the extent we can observe consumer preferences, it is merely by proxy; we do not observe directly
who those particular customers are and what preferences they have.
Another challenge to interpretation is the thickness - or, more often, the thinness - of the sample.
The sample presented in this study is perhaps sufficient to study events and trends that impact the
entirety of the study area, as is the case with ADU liberalization or with SB9/SB10. However,
at more local scales, the sample size may become insufficient: Only so many homes are listed
on the market in any given time; rendering the methods presented here very sensitive to outliers
at local scales. Another issue affecting thickness or thinness of a set of listings for a given topic
is the target audience of the listings sites considered: Websites reporting for-sales listings such
as MLSListings.com may be useful to researchers for capturing useful cross-sectional variation in
references to ADUs. However, their target audience - home buyers - may not be all that interested
in other amenities such as public transit; reception of new public transit may be better captured by
other listings web sites such as Craigslist, apartments.com or GoSection8 if their audiences have
greater overlap with the target audience for a particular policy.
146

Evaluating neighborhood characteristics through listings may also introduce a special form
of bias: We only observe units that are on the market: Listings not available for units that are
not on the market, may not reflect preferences of incumbents. Interpreting trends in the data is
further complicated by shifts in underlying preferences due to other events that happened during
the same time frame: The time frame covered by this study coincides with the Covid-19 pandemic,
which likely impacted society-wide preferences regarding living space, taking public transport,
and regarding living in cities.
Nonetheless, the examples presented in this paper show that information contained in the free
text portions of housing listings can be of interest to planners and policymakers to better understand housing preferences. I demonstrate that there are indeed longitudinal signals in this data;
that repeatedly collecting listings for the same area can allow for tracking how adoption of policies
and amenities occurs over time. Listings can provide valuable insights into what local amenities
and infrastructures are of interest to home buyers, and serve as a barometer for the adoption of
new housing laws. A great advantage over other sources is the timeliness of information in listings
data versus official data sources: Listings reflect current market conditions, while institutional data
sources such as home sales often come with a publication lag. Assuming partnerships between policy researchers and Multiple Listings Services are possible, an application for this method would
be to perform dashboarding of policy trends in real time.
Future research may take the ideas presented in this paper a step further, and apply more sophisticated ways of programmatically reading housing listings - such as via large language models
(e.g. ChatGPT or BERT) - to extract references to concepts from listings.24
24Finally, as a future research agenda, I would like to introduce and demonstrate some novel spatial measures of
text: For instance, I would like to present maps of where words - or concepts - appear statistically significantly more
commonly than across the entire sample. For this measure, I perform a two-sample difference of means test, comparing
the share of listings referring to a concept within a spatial unit - such as a neighborhood or a census tract - to the share
of listings referring to that same concept across the entirety of listings of the same type.
147

Chapter 6
Conclusion
This dissertation could just as well have been titled ”Four Essays exploring the Potential and Shortcomings of Publicly Available Big Data in Urban Studies”, capitalizing on the trend of discussing
”Big Data” (C. Wang and Yin, 2023). However, I intentionally chose otherwise - my focus is on
the urban experience; on walkability, housing development, and how homes - and the places they
are situated in - are marketed. The papers explore what neighborhood features prospective buyers
are interested in; whether walking to them is possible and if so - whether it is likely to be a pleasant
experience based on land uses encountered along the way, and what trends exist in the creation of
new housing units, as the locations of new developments have long-term implications for urban
experiences of the future.
As noted in this dissertation’s introduction, I seek to ask questions at the broadest scale possible, but based on bottom-up, highly local analyses that I then scale up to a broader scope. Previously, zooming out on and scaling up existing research questions was previously largely conducted
via literature reviews (for example, Freemark, 2023), or via meta-analysis (for example, Mohammad, Graham, Melo, and Anderson, 2013 or Hamidi, Kittrell, and Ewing, 2016). Beyond the
papers’ individual questions, the four papers jointly do explore a theme of what can - and what
cannot - be done with publicly accessible data sources if a goal is to achieve as broad a geographic
scale as possible.
As I discuss in Chapter 2, my pathway buffer measure could just as well have been constructed
with local, institutional data sources. Indeed, relying on local land use data may be preferable
148

for researching inherently local questions, as data collected by local governments are likely more
consistent in quality than OpenStreetMap. However, scaling up the geographical scope of this
analysis beyond one metropolitan area - as I do in Chapter 3 - would have required collecting land
use data that are collected differently from dozens of sources and standardizing data across those
many studies. While not impossible, such an approach would require human time beyond the scope
of a solo effort, and raised questions of comparability of measures across places. OpenStreetMap
cannot be used entirely free of disclaimers - its level of completeness varies from place to place
(Zhou, Wang, and Liu, 2022) - but the fact that it nominally records features in the same way
across the entire planet renders it a highly useful data source for analyses that require geodata at
such broad scales.
I encountered a different tradeoff when selecting data for studying housing development for
chapter 4 of this dissertation - between being granular spatially, temporal fidelity, and enjoying
broad geographic coverage. It appears that any given source can posses at most two of these desirable characteristics, and that policy researchers have to choose carefully what to sacrifice: Many
researchers - for example, Dong, 2023 or Schuetz, 2020 - lean on local microdata such as assessor
parcel data or building permits, as are typically collected by municipalities or counties. Such data
are extremely fine-grained both spatially - often at the address or parcel level - and temporally,
but do not enjoy broad geographic coverage as that is not part of their administrative purpose.
Another family of commonly used data sources includes nationwide housing starts as well as the
Census Bureau’s Building Permits Survey; these sources enjoy broad geographic coverage and a
fine temporal resolution at the expense of spatial fidelity, with numbers reported at the municipality or county level. By using housing unit counts from Decennial Census in the third chapter
of this dissertation, I chose the third possible combination of desirable characteristics: The data
are geographically granular with nationwide coverage at the census block level, but only count the
number of housing units every ten units.
Finally, in Chapter 5 I work with listing-level data; a form of microdata in a sense. Such listings
are gathered - downloaded - one by one via a web scraping approach. While studies exist in which
149

data collected in such a manner - for example, Blanco and Song, 2024’s evaluation of income
source discrimination laws’ effect on Craigslist rental listings at a national scale - performing web
scraping at this scale requires substantial computational resources and oversight. However, asking
similar questions at a broader geographic scale while maintaining or even improving temporal
coverage could be possible if a researcher was to partner with a host of Multiple Listings Services,
or with a national housing listings site such as Zillow or Redfin to access their entire archive of
listings.
Taken together, this dissertation shows that there is potential for broad geographic scopes in
quantitative urban studies research. However, researchers need to be aware of the presence of
these tradeoffs and limitations - and push back on the notion of ”big data” as a panacea that can
answer all questions in urban governance and policy.
150

References
Adkins, A., Dill, J., Luhr, G., & Neal, M. (2012). Unpacking Walkability: Testing the Influence of
Urban Design Features on Perceptions of Walking Environment Attractiveness [Publisher:
Routledge eprint: https://doi.org/10.1080/13574809.2012.706365]. Journal of Urban Design, 17(4), 499–510. https://doi.org/10.1080/13574809.2012.706365
Adu, P., & Delmelle, E. C. (2022). Spatial Variations in Exclusionary Criteria from Online Rental
Advertisements. The Professional Geographer, 74(4), 704–714. https://doi.org/10.1080/
00330124.2022.2061537
Agustini, K. A. V., & West, S. E. (2022). Redevelopment along arterial streets: The effects of light
rail on land use change. Real Estate Economics, n/a(n/a), 1–40. https://doi.org/10.1111/
1540-6229.12407
Annamoradnejad, R., Annamoradnejad, I., Safarrad, T., & Habibi, J. (2019). Using Web Mining
in the Analysis of Housing Prices: A Case study of Tehran [ISSN: null]. 2019 5th International Conference on Web Research (ICWR), 55–60. https://doi.org/10.1109/ICWR.2019.
8765250
Aparicio, J. T., Arsenio, E., Santos, F. C., & Henriques, R. (2024). Walkability defined neighborhoods for sustainable cities. Cities, 149, 104944. https : / / doi . org / 10 . 1016 / j . cities . 2024 .
104944
Asquith, B. J., Mast, E., & Reed, D. (2023). Local effects of large new apartment buildings in
low-income areas. The Review of Economics and Statistics, 105(2), 359–375.
Banerjee, T., Uhm, J., & Bahl, D. (2014). Walking to School: The Experience of Children in
Inner City Los Angeles and Implications for Policy [Publisher: SAGE Publications Inc].
Journal of Planning Education and Research, 34(2), 123–140. https://doi.org/10.1177/
0739456X14522494
151

Barrington-Leigh, C., & Millard-Ball, A. (2017). The world’s user-generated road map is more
than 80% complete [Publisher: Public Library of Science]. PLOS ONE, 12(8), e0180698.
https://doi.org/10.1371/journal.pone.0180698
Bartik, A., Gupta, A., & Milo, D. (2023, November). The Costs of Housing Regulation: Evidence
From Generative Regulatory Measurement. https://doi.org/10.2139/ssrn.4627587
Basu, N., Haque, M. M., King, M., Kamruzzaman, M., & Oviedo-Trespalacios, O. (2022). A systematic review of the factors associated with pedestrian route choice [Publisher: Routledge
eprint: https://doi.org/10.1080/01441647.2021.2000064]. Transport Reviews, 42(5), 672–
694. https://doi.org/10.1080/01441647.2021.2000064
Baum-Snow, N. (2023a). Constraints on City and Neighborhood Growth: The Central Role of
Housing Supply. Journal of Economic Perspectives, 37(2), 53–74. https://doi.org/10.1257/
jep.37.2.53
Baum-Snow, N. (2023b). Constraints on city and neighborhood growth: The central role of housing
supply. Journal of Economic Perspectives, 37(2), 53–74.
Baum-Snow, N., & Han, L. (2019). The microgeography of housing supply. Work in progress,
University of Toronto.
Been, V., Ellen, I. G., & O’Regan, K. (2019). Supply skepticism: Housing supply and affordability.
Housing Policy Debate, 29(1), 25–40.
Benfer, E. A., Koehler, R., Mark, A., Nazzaro, V., Alexander, A. K., Hepburn, P., Keene, D. E.,
& Desmond, M. (2023). Covid-19 housing policy: State and federal eviction moratoria
and supportive measures in the united states during the pandemic. Housing Policy Debate,
33(6), 1390–1414.
Besbris, M., Schachter, A., & Kuk, J. (2021). The Unequal Availability of Rental Housing Information Across Neighborhoods. Demography, 58(4), 1197–1221. https://doi.org/10.1215/
00703370-9357518
Bhat, C. R., & Guo, J. Y. (2007). A comprehensive analysis of built environment characteristics on
household residential choice and auto ownership levels. Transportation Research Part B:
Methodological, 41(5), 506–526. https://doi.org/10.1016/j.trb.2005.12.005
152

Blanco, H., & Song, J. (2024, May). Discrimination Against Housing Vouchers: Evidence from
Online Rental Listings. https://doi.org/10.2139/ssrn.4819763
Boarnet, M. G., Day, K., Alfonzo, M., Forsyth, A., & Oakes, M. (2006). The Irvine–Minnesota Inventory to Measure Built Environments: Reliability Tests. American Journal of Preventive
Medicine, 30(2), 153–159.e43. https://doi.org/10.1016/j.amepre.2005.09.018
Boarnet, M. G., Forsyth, A., Day, K., & Oakes, J. M. (2011). The Street Level Built Environment and Physical Activity and Walking: Results of a Predictive Validity Study for the
Irvine Minnesota Inventory [Publisher: SAGE Publications Inc]. Environment and Behavior, 43(6), 735–775. https://doi.org/10.1177/0013916510379760
Boarnet, M. G., Giuliano, G., Hou, Y., & Shin, E. J. (2017). First/last mile transit access as an
equity planning issue. Transportation Research Part A: Policy and Practice, 103, 296–310.
https://doi.org/10.1016/j.tra.2017.06.011
Boeing, G. (2017). OSMnx: New methods for acquiring, constructing, analyzing, and visualizing
complex street networks. Computers, Environment and Urban Systems, 65, 126–139. https:
//doi.org/10.1016/j.compenvurbsys.2017.05.004
Boeing, G. (2020a). A multi-scale analysis of 27,000 urban street networks: Every US city, town,
urbanized area, and Zillow neighborhood [Publisher: SAGE Publications Sage UK: London, England]. Environment and Planning B: Urban Analytics and City Science, 47(4),
590–608.
Boeing, G. (2020b). Online rental housing market representation and the digital reproduction of
urban inequality [Publisher: SAGE Publications Ltd]. Environment and Planning A: Economy and Space, 52(2), 449–468. https://doi.org/10.1177/0308518X19869678
Boeing, G. (2021). Street Network Models and Indicators for Every Urban Area in the World
[Publisher: John Wiley & Sons, Ltd]. Geographical Analysis, n/a(n/a). https://doi.org/10.
1111/gean.12281
Boeing, G., Higgs, C., Liu, S., Giles-Corti, B., Sallis, J. F., Cerin, E., Lowe, M., Adlakha, D.,
Hinckson, E., Moudon, A. V., Salvo, D., Adams, M. A., Barrozo, L. V., Bozovic, T.,
Delclos-Ali ` o, ´ X., Dygryn, ´ J., Ferguson, S., Gebel, K., Ho, T. P., . . . Arundel, J. (2022).
Using open data and open-source software to develop spatial indicators of urban design
and transport features for achieving healthy and sustainable cities [Publisher: Elsevier]. The
Lancet Global Health, 10(6), e907–e918. https://doi.org/10.1016/S2214-109X(22)00072-9
153

Boeing, G., Pilgram, C., & Lu, Y. (2024). Urban street network design and transport-related greenhouse gas emissions around the world. Transportation Research Part D: Transport and
Environment, 127, 103961. https://doi.org/10.1016/j.trd.2023.103961
Boeing, G., & Waddell, P. (2017). New Insights into Rental Housing Markets across the United
States: Web Scraping and Analyzing Craigslist Rental Listings [Publisher: SAGE Publications Inc]. Journal of Planning Education and Research, 37(4), 457–476. https://doi.org/
10.1177/0739456X16664789
Bogin, A., Doerner, W., & Larson, W. (2019). Local house price dynamics: New indices and stylized facts. Real Estate Economics, 47(2), 365–398.
Borkowska, S., & Pokonieczny, K. (2022). Analysis of OpenStreetMap Data Quality for Selected
Counties in Poland in Terms of Sustainable Development [Number: 7 Publisher: Multidisciplinary Digital Publishing Institute]. Sustainability, 14(7), 3728. https://doi.org/10.3390/
su14073728
Bourassa, S. C., Hoesli, M., & Peng, V. S. (2003). Do housing submarkets really matter? Journal
of Housing Economics, 12(1), 12–28.
Bright, J., De Sabbata, S., Lee, S., Ganesh, B., & Humphreys, D. K. (2018). OpenStreetMap data
for alcohol research: Reliability assessment and quality indicators. Health & Place, 50,
130–136. https://doi.org/10.1016/j.healthplace.2018.01.009
Brovelli, M. A., & Zamboni, G. (2018). A New Method for the Assessment of Spatial Accuracy
and Completeness of OpenStreetMap Building Footprints [Number: 8 Publisher: Multidisciplinary Digital Publishing Institute]. ISPRS International Journal of Geo-Information,
7(8), 289. https://doi.org/10.3390/ijgi7080289
Brown, B. B., Werner, C. M., Amburgey, J. W., & Szalay, C. (2007). Walkable Route Perceptions
and Physical Features: Converging Evidence for En Route Walking Experiences [Publisher:
SAGE Publications Inc]. Environment and Behavior, 39(1), 34–61. https://doi.org/10.1177/
0013916506295569
Brueckner, J. K., & Thomaz, S. (2024). ADUs in Los Angeles: Where are they located and by
how much do they raise property value? Real Estate Economics, 52(3), 885–907. https :
//doi.org/10.1111/1540-6229.12471
154

Buttazzoni, A., & Minaker, L. (2022). Associations between adolescent mental health and pedestrianand transit-oriented urban design qualities: Evidence from a national-level online Canadian
survey [Publisher: SAGE Publications Ltd]. Urban Studies, 00420980221138311. https :
//doi.org/10.1177/00420980221138311
Calthorpe, P. (1989). The pedestrian pocket. Pedestrian Pocket Book, 350–356.
Cao, X. ( (2015). Residential Preference and Choice of Movers in Light Rail Neighborhoods
in Minneapolis, Minnesota [Publisher: SAGE Publications Inc]. Transportation Research
Record, 2494(1), 1–10. https://doi.org/10.3141/2494-01
Cao, X. (, & Schoner, J. (2014). The influence of light rail transit on transit use: An exploration
of station area residents along the Hiawatha line in Minneapolis. Transportation Research
Part A: Policy and Practice, 59, 134–143. https://doi.org/10.1016/j.tra.2013.11.001
Case, K. E., Shiller, R. J., et al. (1994). A decade of boom and bust in the prices of single-family
homes: Boston and los angeles, 1983 to 1993. New England Economic Review, (Mar), 40–
51.
Case, K. E., & Shiller, R. J. (2003). Is there a bubble in the housing market? Brookings papers on
economic activity, 2003(2), 299–362.
Cerin, E., Sallis, J. F., Salvo, D., Hinckson, E., Conway, T. L., Owen, N., Dyck, D. v., Lowe,
M., Higgs, C., Moudon, A. V., Adams, M. A., Cain, K. L., Christiansen, L. B., Davey, R.,
Dygryn, J., Frank, L. D., Reis, R., Sarmiento, O. L., Adlakha, D., . . . ´ Giles-Corti, B. (2022).
Determining thresholds for spatial urban design and transport features that support walking
to create healthy and sustainable cities: Findings from the IPEN Adult study [Publisher:
Elsevier]. The Lancet Global Health, 10(6), e895–e906. https://doi.org/10.1016/S2214-
109X(22)00068-7
Cervero, R. (2002). Built environments and mode choice: Toward a normative framework. Transportation Research Part D: Transport and Environment, 7(4), 265–284. https://doi.org/10.
1016/S1361-9209(01)00024-4
Cervero, R., & Guerra, E. (2011). Urban Densities and Transit: A Multi-dimensional Perspective.
Cervero, R., & Kockelman, K. (1997). Travel demand and the 3Ds: Density, diversity, and design.
Transportation Research Part D: Transport and Environment, 2(3), 199–219. https://doi.
org/10.1016/S1361-9209(97)00009-6
155

Cetintahra, G. E., & Cubukcu, E. (2015). The influence of environmental aesthetics on economic
value of housing: An empirical research on virtual environments. Journal of Housing and
the Built Environment, 30(2), 331–340. https://doi.org/10.1007/s10901-014-9413-6
Chambers, E. (1984). A New Measure of the Rental Cost of Housing in the Toronto Market, 1890-
1914 [Number: 33]. Histoire sociale / Social History, 17(33).
Chen, Y., Liu, X., Li, X., Liu, Y., & Xu, X. (2016). Mapping the fine-scale spatial pattern of housing
rent in the metropolitan area by using online rental listings and ensemble learning. Applied
Geography, 75, 200–212. https://doi.org/10.1016/j.apgeog.2016.08.011
Corinth, K., & Irvine, A. (2023). JUE Insight: The Effect of Relaxing Local Housing Market
Regulations on Federal Rental Assistance Programs. Journal of Urban Economics, 136,
103572. https://doi.org/10.1016/j.jue.2023.103572
Costa, A., Sass, V., Kennedy, I., Roy, R., Walter, R. J., Acolin, A., Crowder, K., Hess, C., Ramiller,
A., & Chasins, S. (2021). Toward a Cross-Platform Framework: Assessing the Comprehensiveness of Online Rental Listings. Cityscape (Washington, D.C.), 23(2), 327–339.
Couture, V., & Handbury, J. (2020). Urban revival in America. Journal of Urban Economics, 119,
103267. https://doi.org/10.1016/j.jue.2020.103267
Couture, V., & Handbury, J. (2023). Neighborhood change, gentrification, and the urbanization of
college graduates. The Journal of Economic Perspectives, 37(2), 29–52.
Cui, M., & Levinson, D. (2020). Primal and Dual Access. Geographical Analysis, 52(3), 452–474.
https://doi.org/10.1111/gean.12220
Dawkins, C., & Moeckel, R. (2016). Transit-Induced Gentrification: Who Will Stay, and Who
Will Go? [Publisher: Routledge eprint: https://doi.org/10.1080/10511482.2016.1138986].
Housing Policy Debate, 26(4-5), 801–818. https : / / doi . org / 10 . 1080 / 10511482 . 2016 .
1138986
Day, K., Boarnet, M., Alfonzo, M., & Forsyth, A. (2006). The Irvine–Minnesota Inventory to Measure Built Environments: Development. American Journal of Preventive Medicine, 30(2),
144–152. https://doi.org/10.1016/j.amepre.2005.09.017
156

Delbosc, A., & Currie, G. (2011). Using Lorenz curves to assess public transport equity. Journal of
Transport Geography, 19(6), 1252–1259. https://doi.org/10.1016/j.jtrangeo.2011.02.008
Ding, C., Cao, X., & Liu, C. (2019). How does the station-area built environment influence Metrorail ridership? Using gradient boosting decision trees to identify non-linear thresholds.
Journal of Transport Geography, 77, 70–78. https : / / doi . org / 10 . 1016 / j . jtrangeo . 2019 .
04.011
Dong, H. (2021). Exploring the Impacts of Zoning and Upzoning on Housing Development: A
Quasi-experimental Analysis at the Parcel Level [Publisher: SAGE Publications Inc]. Journal of Planning Education and Research, 0739456X21990728. https://doi.org/10.1177/
0739456X21990728
Dong, H. (2023). Higher Density Development for Lower Cost Housing? Understanding the Multifamily Housing Market and the Role of Density in Multifamily Home Prices. Journal of
Planning Education and Research, 43(3), 617–636. https://doi.org/10.1177/0739456X20912829
Eisenlohr, A., Jamme, H.-T., Bahl, D., & Banerjee, T. (2023). The ambivalence of independent
mobility: Balancing perceived risks and expected benefits of walking to school in innercity neighborhoods. Travel Behaviour and Society, 31, 49–62. https://doi.org/10.1016/j.
tbs.2022.10.014
Eriksen, M. D., & Orlando, A. W. (2022). Returns to scale in residential construction: The marginal
impact of building height. Real Estate Economics, 50(2), 534–564.
Ewing, R., & Cervero, R. (2001). Travel and the Built Environment: A Synthesis [Publisher: SAGE
Publications Inc]. Transportation Research Record, 1780(1), 87–114. https://doi.org/10.
3141/1780-10
Ewing, R., & Cervero, R. (2010). Travel and the Built Environment [Publisher: Routledge eprint:
https://doi.org/10.1080/01944361003766766]. Journal of the American Planning Association, 76(3), 265–294. https://doi.org/10.1080/01944361003766766
Fan, Y., Guthrie, A., & Levinson, D. (2012). Impact of light-rail implementation on labor market
accessibility: A transportation equity perspective [Publisher: Journal of Transport and Land
Use]. Journal of Transport and Land Use, 5(3), 28–39.
Fang, L., Stewart, N., & Tyndall, J. (2023). Homeowner politics and housing supply. Journal of
Urban Economics, 138, 103608. https://doi.org/10.1016/j.jue.2023.103608
157

Federal Highway Administration. (2017). 2017 National Household Travel SUrvey.
Federal Transit Administration. (2011). Final Policy Statement on the Eligibility of Pedestrian and
Bicycle Improvements Under Federal Transit Law. Federal Register, 76(1).
Federal Transit Administration. (2017). Manual on Pedestrian and Bicycle Connections to Transit
(Report 0111).
Fisher, A. (2023, March). Zillow dominates the portals, but Homes.com gains ground.
Foodservice Database Company. (2022, May). Top 250 Restaurant Chains in the U.S. 2019 (fsdbco.com).
Foster, D. P., Liberman, M., & Stine, R. A. (2013). Featurizing text: Converting text into predictors
for regression analysis. The Wharton School of the University of Pennsylvania, Philadelphia, PA.
Foti, F., Davis, M., Fernandez, F., & Maurer, S. (2023, July). Pandana — pandana 0.7 documentation.
Frank, L. D., Sallis, J. F., Conway, T. L., Chapman, J. E., Saelens, B. E., & Bachman, W. (2006).
Many Pathways from Land Use to Health: Associations between Neighborhood Walkability and Active Transportation, Body Mass Index, and Air Quality [Publisher: Routledge
eprint: https://doi.org/10.1080/01944360608976725]. Journal of the American Planning
Association, 72(1), 75–87. https://doi.org/10.1080/01944360608976725
Freemark, Y. (2020). Upzoning Chicago: Impacts of a Zoning Reform on Property Values and
Housing Construction. Urban Affairs Review, 56(3), 758–789. https : / / doi . org / 10 . 1177 /
1078087418824672
Freemark, Y. (2023). Zoning Change: Upzonings, Downzonings, and Their Impacts on Residential
Construction, Housing Costs, and Neighborhood Demographics [Publisher: SAGE Publications Inc]. Journal of Planning Literature, 38(4), 548–570. https : / / doi . org / 10 . 1177 /
08854122231166961
Fu, X. (2024). Natural Language Processing in Urban Planning: A Research Agenda [Publisher:
SAGE Publications Inc]. Journal of Planning Literature, 08854122241229571. https://doi.
org/10.1177/08854122241229571
158

Garau, C., Annunziata, A., & Yamu, C. (2024). A walkability assessment tool coupling multicriteria analysis and space syntax: The case study of Iglesias, Italy [Publisher: Routledge
eprint: https://doi.org/10.1080/09654313.2020.1761947]. European Planning Studies, 32(2),
211–233. https://doi.org/10.1080/09654313.2020.1761947
Garcia-Lopez, ´ M.-A., ` Holl, A., & Viladecans-Marsal, E. (2015). Suburbanization and highways
in Spain when the Romans and the Bourbons still shape its cities [Publisher: Elsevier].
Journal of Urban Economics, 85, 52–67.
Garrett, M., & Taylor, B. (1999). Reconsidering Social Equity in Public Transit. Berkeley Planning
Journal, 13(1). https://doi.org/10.5070/BP313113028
Gehrke, S. R., & Clifton, K. J. (2019). An activity-related land use mix construct and its connection to pedestrian travel [Publisher: SAGE Publications Ltd STM]. Environment and
Planning B: Urban Analytics and City Science, 46(1), 9–26. https : / / doi . org / 10 . 1177 /
2399808317690157
General Transit Feed Specification. (2024). General Transit Feed Specification.
Givoni, M., & Rietveld, P. (2007). The access journey to the railway station and its role in passengers’ satisfaction with rail travel. Transport Policy, 14(5), 357–365. https://doi.org/10.
1016/j.tranpol.2007.04.004
Google Transit. (2022, September). GTFS Static Overview — Static Transit.
Gray, M. N. (2022, June). Arbitrary Lines: How Zoning Broke the American City and How to Fix
It [Google-Books-ID: WWlnEAAAQBAJ]. Island Press.
Griffin, G. P., & Sener, I. N. (2016). Public Transit Equity Analysis at Metropolitan and Local
Scales: A Focus on Nine Large Cities in the US. Journal of Public Transportation, 19(4),
126–143. https://doi.org/10.5038/2375-0901.19.4.8
Gross, J. (2024). Insights from pedestrian lingering in California’s urban areas.
Guerra, E., Cervero, R., & Tischler, D. (2012). Half-Mile Circle: Does It Best Represent Transit
Station Catchments? [Publisher: SAGE Publications Inc]. Transportation Research Record,
2276(1), 101–109. https://doi.org/10.3141/2276-12
159

Guo, Z., & Ferreira, J. J. (2008). Pedestrian Environments, Transit Path Choice, and Transfer Penalties: Understanding Land-Use Impacts on Transit Travel. Environment and Planning B,
35(3), 461–479.
Gutierrez, J., Cardozo, O. D., & Garc ´ ´ıa-Palomares, J. C. (2011). Transit ridership forecasting at station level: An approach based on distance-decay weighted regression. Journal of Transport
Geography, 19(6), 1081–1092. https://doi.org/10.1016/j.jtrangeo.2011.05.004
Hall, C. M., & Ram, Y. (2018). Walk score® and its potential contribution to the study of active
transport and walkability: A critical and systematic review. Transportation Research Part
D: Transport and Environment, 61, 310–324. https://doi.org/10.1016/j.trd.2017.12.018
Hamidi, S., Kittrell, K., & Ewing, R. (2016). Value of Transit as Reflected in U.S. Single-Family
Home Premiums: A Meta-Analysis [Publisher: SAGE Publications Inc]. Transportation
Research Record, 2543(1), 108–115. https://doi.org/10.3141/2543-12
Harris, C. (2020, May). The Economics of Urban Light Rail: A Guide for Planners and Citizens.
Harten, J. G., Kim, A. M., & Brazier, J. C. (2021). Real and fake data in Shanghai’s informal rental
housing market: Groundtruthing data scraped from the internet [Publisher: SAGE Publications Ltd]. Urban Studies, 58(9), 1831–1845. https://doi.org/10.1177/0042098020918196
Hartig, T., Evans, G. W., Jamner, L. D., Davis, D. S., & G arling, T. (2003). Tracking restoration ¨
in natural and urban field settings. Journal of Environmental Psychology, 23(2), 109–123.
https://doi.org/10.1016/S0272-4944(02)00109-3
Herfort, B., Lautenbach, S., Porto de Albuquerque, J., Anderson, J., & Zipf, A. (2022, August).
Investigating the digital divide in OpenStreetMap: Spatio-temporal analysis of inequalities
in global urban building completeness. https://doi.org/10.21203/rs.3.rs-1913150/v1
Hess, C., Walter, R. J., Kennedy, I., Acolin, A., Ramiller, A., & Crowder, K. (2023). Segmented
Information, Segregated Outcomes: Housing Affordability and Neighborhood Representation on a Voucher-Focused Online Housing Platform and Three Mainstream Alternatives
[Publisher: Routledge eprint: https://doi.org/10.1080/10511482.2022.2133548]. Housing
Policy Debate, 33(6), 1511–1535. https://doi.org/10.1080/10511482.2022.2133548
Hsieh, C.-T., & Moretti, E. (2019). Housing constraints and spatial misallocation. American Economic Journal: Macroeconomics, 11(2), 1–39.
160

Hurst, N. B., & West, S. E. (2014). Public transit and urban redevelopment: The effect of light rail
transit on land use in Minneapolis, Minnesota. Regional Science and Urban Economics,
46, 57–72. https://doi.org/10.1016/j.regsciurbeco.2014.02.002
Ikeda, E., Mavoa, S., Cavadino, A., Carroll, P., Hinckson, E., Witten, K., & Smith, M. (2020).
Keeping kids safe for active travel to school: A mixed method examination of school policies and practices and children’s school travel behaviour. Travel Behaviour and Society, 21,
57–68. https://doi.org/10.1016/j.tbs.2020.05.008
Internline.io. (2024). Intro to Transitland - the largest and most powerful transit data platform
collecting thousands of GTFS and GTFS Realtime feeds from cities around the world.
Jacobs, J. (1961). The Death and Life of Great American Cities. Random House.
Jane Jacobs. (1961). The Curse of Border Vacuums. In The Death and Life of Great American
Cities. Random House.
Jiang, Y., Christopher Zegras, P., & Mehndiratta, S. (2012). Walk the line: Station context, corridor
type and bus rapid transit walk access in Jinan, China. Journal of Transport Geography,
20(1), 1–14. https://doi.org/10.1016/j.jtrangeo.2011.09.007
Jomehpour Chahar Aman, J., & Smith-Colin, J. (2020). Transit Deserts: Equity analysis of public
transit accessibility. Journal of Transport Geography, 89, 102869. https://doi.org/10.1016/
j.jtrangeo.2020.102869
Jun, M.-J., Choi, K., Jeong, J.-E., Kwon, K.-H., & Kim, H.-J. (2015). Land use characteristics
of subway catchment areas and their influence on subway ridership in Seoul. Journal of
Transport Geography, 48, 30–40. https://doi.org/10.1016/j.jtrangeo.2015.08.002
Karner, A. (2018). Assessing public transit service equity using route-level accessibility measures
and public data. Journal of Transport Geography, 67, 24–32. https://doi.org/10.1016/j.
jtrangeo.2018.01.005
Kazis, N. M. (2023). Learning From Land Use Reforms: Housing Outcomes and Regulatory
Change [Publisher: US Department of Housing and Urban Development]. Cityscape, 25(2),
93–106.
161

Ki, D., & Chen, Z. (2023). Walkability inequity in Los Angeles: Uncovering the overlooked role
of micro-level features. Transportation Research Part D: Transport and Environment, 122,
103888. https://doi.org/10.1016/j.trd.2023.103888
Kim, A. M., & Kang, K. H. (2023). We Are Here: City Signs and Maps in Ethnic Placemaking
[Publisher: Routledge eprint: https://doi.org/10.1080/01944363.2022.2030785]. Journal
of the American Planning Association, 89(1), 2–15. https://doi.org/10.1080/01944363.
2022.2030785
Kim, M. (2024). The Case for Mass Upzoning. Housing Policy Debate, 34(2), 246–251. https :
//doi.org/10.1080/10511482.2023.2234890
Kim, S., Ulfarsson, G. F., & Todd Hennessy, J. (2007). Analysis of light rail rider travel behavior:
Impacts of individual, built environment, and crime characteristics on transit access. Transportation Research Part A: Policy and Practice, 41(6), 511–522. https://doi.org/10.1016/j.
tra.2006.11.001
Kordjamshidi, P., Van Otterlo, M., & Moens, M.-F. (2011). Spatial role labeling: Towards extraction of spatial relations from natural language. ACM Transactions on Speech and Language
Processing, 8(3), 4:1–4:36. https://doi.org/10.1145/2050104.2050105
Kroft, K., & Pope, D. G. (2014). Does Online Search Crowd Out Traditional Search and Improve
Matching Efficiency? Evidence from Craigslist [Publisher: The University of Chicago Press].
Journal of Labor Economics, 32(2), 259–303. https://doi.org/10.1086/673374
Kuhlmann, D. (2021). Upzoning and Single-Family Housing Prices [Publisher: Routledge eprint:
https://doi.org/10.1080/01944363.2020.1852101]. Journal of the American Planning Association, 87(3), 383–395. https://doi.org/10.1080/01944363.2020.1852101
LaJeunesse, S., Ryus, P., Kumfer, W., Kothuri, S., & Nordback, K. (2021). Measuring Pedestrian
Level of Stress in Urban Environments: Naturalistic Walking Pilot Study [Publisher: SAGE
Publications Inc]. Transportation Research Record, 2675(10), 109–119. https://doi.org/10.
1177/03611981211010183
Larsen, J., El-Geneidy, A., & Yasmin, F. (2010). Beyond the Quarter Mile: Re-examining Travel
Distances by Active Transportation [Publisher: Institute of Urban Studies, University of
Winnipeg]. Canadian Journal of Urban Research, 19(1), 70–88.
162

Lee, H. (2020). Are Millennials Coming to Town? Residential Location Choice of Young Adults
[Publisher: SAGE Publications Inc]. Urban Affairs Review, 56(2), 565–604. https : / / doi .
org/10.1177/1078087418787668
Lee, J., & Lee, K. O. (2023). Online listing data and their interaction with market dynamics: Evidence from Singapore during COVID-19. Journal of Big Data, 10(1), 99. https://doi.org/
10.1186/s40537-023-00786-5
Lee, S., & Talen, E. (2014). Measuring Walkability: A Note on Auditing Methods [Publisher: Routledge eprint: https://doi.org/10.1080/13574809.2014.890040]. Journal of Urban Design,
19(3), 368–388. https://doi.org/10.1080/13574809.2014.890040
Leidner, J. L., & Lieberman, M. D. (2011). Detecting geographical references in the form of place
names and associated spatial natural language. SIGSPATIAL Special, 3(2), 5–11. https :
//doi.org/10.1145/2047296.2047298
Li, S., Yang, H., Zhang, G., Ling, Z., Xiong, Y., & Li, Y. (2019). What is the best catchment area of
a metro station? A study based on station level ridership modeling. 2019 5th International
Conference on Transportation Information and Safety (ICTIS), 1239–1244. https://doi.org/
10.1109/ICTIS.2019.8883789
Liu, J., Xiao, L., & Zhou, J. (2021). Built environment correlates of walking for transportation
[Publisher: JSTOR]. Journal of Transport and Land Use, 14(1), 1129–1148.
Liu, S., Higgs, C., Arundel, J., Boeing, G., Cerdera, N., Moctezuma, D., Cerin, E., Adlakha, D.,
Lowe, M., & Giles-Corti, B. (2022). A Generalized Framework for Measuring Pedestrian
Accessibility around the World Using Open Data. Geographical Analysis, 54(3), 559–582.
https://doi.org/10.1111/gean.12290
Liu, Y., Yang, D., Timmermans, H. J. P., & de Vries, B. (2020). The impact of the street-scale
built environment on pedestrian metro station access/egress route choice. Transportation
Research Part D: Transport and Environment, 87, 102491. https://doi.org/10.1016/j.trd.
2020.102491
Lopez Ochoa, E. (2023). Housing price indices from online listing data: Addressing the spatial
bias with sampling weights [Publisher: SAGE Publications Ltd STM]. Environment and
Planning B: Urban Analytics and City Science, 50(4), 1039–1056. https://doi.org/10.1177/
23998083221130713
163

Lore, M., Harten, J. G., & Boeing, G. (2024). A hybrid deep learning method for identifying topics
in large-scale urban text data: Benefits and trade-offs. Computers, Environment and Urban
Systems, 111, 102131. https://doi.org/10.1016/j.compenvurbsys.2024.102131
Los Angeles City Planning. (2024a). Executive Directive 1 (ED 1).
Los Angeles City Planning. (2024b). Transit Oriented Communities Incentive Program.
Los Angeles Metro. (2020a, April). Crenshaw/LAX Transit Project [Library Catalog: www.metro.net].
Los Angeles Metro. (2020b, April). Purple Line Extension [Library Catalog: www.metro.net].
Los Angeles Metro. (2024). GIS Data – Metro Open Data.
Loukaitou-Sideris, A. (2006). Is it Safe to Walk? Neighborhood Safety and Security Considerations
and Their Effects on Walking [Publisher: SAGE Publications Inc]. Journal of Planning
Literature, 20(3), 219–232. https://doi.org/10.1177/0885412205282770
Malone, T., & Redfearn, C. L. (2022). To measure globally, aggregate locally: Urban land, submarkets, and biased estimates of the housing stock. Real Estate Economics, 50(3), 656–
671.
Manville, M., & Monkkonen, P. (2021). Unwanted Housing: Localism and Politics of Housing
Development [Publisher: SAGE Publications Inc]. Journal of Planning Education and Research, 0739456X21997903. https://doi.org/10.1177/0739456X21997903
Marantz, N. J., Elmendorf, C. S., & Kim, Y. B. (2023a). Where Will Accessory Dwelling Units
Sprout Up When a State Lets Them Grow? Evidence From California [Publisher: US Department of Housing and Urban Development]. Cityscape, 25(2), 107–118.
Marantz, N. J., Elmendorf, C. S., & Kim, Y. B. (2023b). Where will accessory dwelling units sprout
up when a state lets them grow? evidence from california. Cityscape, 25(2), 107–118.
Marshall, W. E., Garrick, N. W., & Hansen, G. (2008). Reassessing on-street parking [Publisher:
SAGE Publications Sage CA: Los Angeles, CA]. Transportation Research Record, 2046(1),
45–52.
164

Mast, E. (2023). Jue insight: The effect of new market-rate housing construction on the low-income
housing market. Journal of Urban Economics, 133, 103383.
McCarthy, J., & Peach, R. W. (2004). Are home prices the next bubble? Available at SSRN 634265.
McDonald, N. C. (2007). Active Transportation to School: Trends Among U.S. Schoolchildren,
1969–2001. American Journal of Preventive Medicine, 32(6), 509–516. https://doi.org/10.
1016/j.amepre.2007.02.022
Meng, M., Koh, P. P., & Wong, Y. D. (2016). Influence of Socio-Demography and Operating
Streetscape on Last-Mile Mode Choice. Journal of Public Transportation, 19(2), 38–54.
https://doi.org/10.5038/2375-0901.19.2.3
Millard-Ball, A., Silverstein, B., Kapshikar, P., Stevenson, S., & Barrington-Leigh, C. (2024).
Dividing Highways: Barrier Effects and Environmental Justice in California [Publisher:
SAGE Publications Inc]. Journal of Planning Education and Research, 0739456X241247330.
https://doi.org/10.1177/0739456X241247330
Minnesota Geospatial Commons. (2020, April). Generalized Land Use 2020.
Minnesota Geospatial Commons. (2022). Transit Stops - Minnesota Geospatial Commons.
Mleczko, M., & Desmond, M. (2023). Using natural language processing to construct a National Zoning and Land Use Database [Publisher: SAGE Publications Ltd]. Urban Studies,
60(13), 2564–2584. https://doi.org/10.1177/00420980231156352
Mohammad, S. I., Graham, D. J., Melo, P. C., & Anderson, R. J. (2013). A meta-analysis of the
impact of rail projects on land and property values [Publisher: Elsevier]. Transportation
Research Part A: Policy and Practice, 50, 158–170.
Mohiuddin, H. (2021). Planning for the First and Last Mile: A Review of Practices at Selected
Transit Agencies in the United States [Number: 4 Publisher: Multidisciplinary Digital Publishing Institute]. Sustainability, 13(4), 2222. https://doi.org/10.3390/su13042222
Monkkonen, P., Lens, M., & Manville, M. (2022, November). Built Out Cities? A New Approach
to Measuring Land Use Regulation.
165

Moos, M. (2016). From gentrification to youthification? The increasing importance of young age in
delineating high-density living [Publisher: SAGE Publications Ltd]. Urban Studies, 53(14),
2903–2920. https://doi.org/10.1177/0042098015603292
Mulley, C., Tsai, C.-H. (, & Ma, L. (2018). Does residential property price benefit from light rail
in Sydney? Research in Transportation Economics, 67, 3–10. https://doi.org/10.1016/j.
retrec.2016.11.002
Myers, D. (2016). Peak Millennials: Three Reinforcing Cycles That Amplify the Rise and Fall
of Urban Concentration by Millennials. Housing Policy Debate, 26(6), 928–947. https :
//doi.org/10.1080/10511482.2016.1165722
Naik, N., Raskar, R., & Hidalgo, C. A. (2016). Cities Are Physical Too: Using Computer Vision
to Measure the Quality and Impact of Urban Appearance. American Economic Review,
106(5), 128–132. https://doi.org/10.1257/aer.p20161030
National Association of Realtors. (2012, January). Multiple Listing Service (MLS): What Is It.
National Association of Realtors. (2023). Highlights From the Profile of Home Buyers and Sellers.
National Center for Education Statistics (NCES). (2023). Common Core of Data (CCD) [Publisher:
National Center for Education Statistics].
Nilsson, I., & Delmelle, E. C. (2022). Understanding Residential Sorting through Property Listings: A Case Study of Neighborhood Change in Charlotte, NC 1993–2018 [Num Pages:
23]. In Economies, Institutions and Territories. Routledge.
Nilsson, I., & Delmelle, E. C. (2023). An embedding-based text classification approach for understanding micro-geographic housing dynamics. International Journal of Geographical
Information Science, 37(12), 2487–2513. https://doi.org/10.1080/13658816.2023.2209803
Nowak, A., & Smith, P. (2017). Textual Analysis in Real Estate. Journal of Applied Econometrics,
32(4), 896–918. https://doi.org/10.1002/jae.2550
Oliveri, R. C. (2009). Discriminatory Housing Advertisements On-Line: Lessons from Craigslist
Symposium. Indiana Law Review, 43(4), 1125–1184.
166

Omura, J. D., Hyde, E. T., Watson, K. B., Sliwa, S. A., Fulton, J. E., & Carlson, S. A. (2019). Prevalence of children walking to school and related barriers—United States, 2017. Preventive
Medicine, 118, 191–195. https://doi.org/10.1016/j.ypmed.2018.10.016
OpenStreetMap.org. (2024). TIGER - OpenStreetMap Wiki.
Orange County Transportation Authority. (2020, April). Construction.
Orlando, A. W., & Redfearn, C. L. (2023). Housing supply elasticities: A structural vector autoregression approach [Working Paper].
Orlando, A. W., & Redfearn, C. L. (2024a). Houston you have a problem: How large cities accommodate more housing. Real Estate Economics, Forthcoming.
Orlando, A. W., & Redfearn, C. L. (2024b). Houston, you have a problem: How large cities accommodate more housing [ eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/1540-
6229.12490]. Real Estate Economics, n/a(n/a). https://doi.org/10.1111/1540-6229.12490
O’Sullivan, S., & Morrall, J. (1996). Walking Distances to and from Light-Rail Transit Stations
[Publisher: SAGE Publications Inc]. Transportation Research Record, 1538(1), 19–26.
https://doi.org/10.1177/0361198196153800103
Padeiro, M., Louro, A., & da Costa, N. M. (2019). Transit-oriented development and gentrification: A systematic review. Transport Reviews, 39(6), 733–754. https://doi.org/10.1080/
01441647.2019.1649316
Park, K., Ewing, R., Scheer, B. C., & Tian, G. (2018). The impacts of built environment characteristics of rail station areas on household travel behavior. Cities, 74, 277–283. https :
//doi.org/10.1016/j.cities.2017.12.015
Park, K., Farb, A., & Chen, S. (2021). First-/last-mile experience matters: The influence of the built
environment on satisfaction and loyalty among public transit riders. Transport Policy, 112,
32–42. https://doi.org/10.1016/j.tranpol.2021.08.003
Park, S., Deakin, E., & Lee, J. S. (2014). Perception-Based Walkability Index to Test Impact of
Microlevel Walkability on Sustainable Mode Choice Decisions [Publisher: SAGE Publications Inc]. Transportation Research Record, 2464(1), 126–134. https://doi.org/10.3141/
2464-16
167

Pelias. (2022). Pelias Geocoder.
Pilgram, C., & West, S. (2023). Transit station area walkability: Identifying impediments to walking using scalable, recomputable land-use measures [Number: 1]. Journal of Transport and
Land Use, 16(1), 267–292. https://doi.org/10.5198/jtlu.2023.2303
Pilgram, C. A., & West, S. E. (2018). Fading premiums: The effect of light rail on residential
property values in Minneapolis, Minnesota. Regional Science and Urban Economics, 69,
1–10. https://doi.org/10.1016/j.regsciurbeco.2017.12.008
Pryce, G. (2013). Housing submarkets and the lattice of substitution. Urban Studies, 50(13), 2682–
2699.
Quintero, L. (2022). Fewer players, fewer homes: Concentration and the new dynamics of housing
supply. Johns Hopkins Carey Business School Research Paper, (18-18).
Redfearn, C. L. (2009). How informative are average effects? Hedonic regression and amenity
capitalization in complex urban housing markets. Regional Science and Urban Economics,
39(3), 297–306. https://doi.org/10.1016/j.regsciurbeco.2008.11.001
Ren, R., Wong, S. K., & Chau, K. W. (2023). A revisit of supply elasticity and within-city heterogeneity of housing price movements. The Journal of Real Estate Finance and Economics,
1–28.
Renne, J. L., & Appleyard, B. (2019). Twenty-five years in the making: TOD as a new name for an
enduring concept [Issue: 4 Pages: 402–408 Publication Title: Journal of Planning Education
and Research Volume: 39].
Schoner, J., & Cao, X. (2014). Walking for Purpose and Pleasure: Influences of Light rail, Built
Environment, and residential Self-Selection on Pedestrian Travel [Publisher: SAGE PublicationsSage CA: Los Angeles, CA]. Transportation Research Record. https://doi.org/10.
3141/2464-09
Schuetz, J. (2020). Teardowns, popups, and renovations: How does housing supply change? [ eprint:
https://onlinelibrary.wiley.com/doi/pdf/10.1111/jors.12470]. Journal of Regional Science,
60(3), 459–480. https://doi.org/10.1111/jors.12470
168

Schweitzer, L. (2014). Planning and Social Media: A Case Study of Public Transit and Stigma
on Twitter [Publisher: Routledge eprint: https://doi.org/10.1080/01944363.2014.980439].
Journal of the American Planning Association, 80(3), 218–238. https://doi.org/10.1080/
01944363.2014.980439
Sing, T. F., & Zou, Y. (2024). Improving search efficiency in housing markets using online listings [ eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/1540-6229.12482]. Real Estate Economics, n/a(n/a). https://doi.org/10.1111/1540-6229.12482
Somerville, C. T. (1999). The industrial organization of housing supply: Market activity, land supply and the size of homebuilder firms. Real Estate Economics, 27(4), 669–694.
Sparber, S. (2023, November). 100% of buyers now shop homes online, NAR says.
Stacy, C., Davis, C., Freemark, Y. S., Lo, L., MacDonald, G., Zheng, V., & Pendall, R. (2023).
Land-use reforms and housing costs: Does allowing for increased density lead to greater affordability? [Publisher: SAGE Publications Ltd]. Urban Studies, 60(14), 2919–2940. https:
//doi.org/10.1177/00420980231159500
State of California. (2021, September). Governor Newsom Signs Historic Legislation to Boost
California’s Housing Supply and Fight the Housing Crisis.
Stawarz, N., Sander, N., & Sulak, H. (2021). Internal migration and housing costs—a panel analysis for germany. Population, Space and Place, 27(4), e2412.
Su, S., He, S., Sun, C., Zhang, H., Hu, L., & Kang, M. (2021). Do landscape amenities impact
private housing rental prices? A hierarchical hedonic modeling approach based on semantic
and sentimental analysis of online housing advertisements across five Chinese megacities.
Urban Forestry & Urban Greening, 58, 126968. https : / / doi . org / 10 . 1016 / j . ufug . 2020 .
126968
Sung, H., Lee, S., & Cheon, S. (2015). Operationalizing Jane Jacobs’s Urban Design Theory:
Empirical Verification from the Great City of Seoul, Korea [Publisher: SAGE Publications
Inc]. Journal of Planning Education and Research, 35(2), 117–130. https : / / doi . org / 10 .
1177/0739456X14568021
Thomas, J., & Reyes, R. (2021). National Walkability Index. Methodology and User Guide. United
States Environmental Protection Agency (EPA). https://www. epa. gov/sites/default/files/2021-
06/documents/national walkability index methodology and user guide june2021. pdf.
169

Tribby, C. P., Miller, H. J., Brown, B. B., Werner, C. M., & Smith, K. R. (2017). Analyzing walking route choice through built environments using random forests and discrete choice techniques. Environment and Planning B, 44(6), 1145–1167.
Van Dam, A. (2024). The school bus is disappearing. Welcome to the era of the school pickup line.
[Section: Business]. Washington Post.
Venter, C. J. (2020). Measuring the quality of the first/last mile connection to public transport.
Research in Transportation Economics, 83, 100949. https://doi.org/10.1016/j.retrec.2020.
100949
Vigdor, J. (2008). The economic aftermath of hurricane katrina. Journal of Economic Perspectives,
22(4), 135–154.
Wachs, M., & Kumagai, T. G. (1973). Physical accessibility as a social indicator. Socio-Economic
Planning Sciences, 7(5), 437–456. https://doi.org/10.1016/0038-0121(73)90041-4
Walker, K., Herman, M., & Eberwein, K. (2023, August). Tidycensus: Load US Census Boundary
and Attribute Data as ’tidyverse’ and ’sf’-Ready Data Frames.
Walker, K., & Rudis, B. (2023, May). Tigris: Load Census TIGER/Line Shapefiles.
WalkScore.com. (2020). Walk Score Methodology.
Wang, C., & Yin, L. (2023). Defining Urban Big Data in Urban Planning: Literature Review [Publisher: American Society of Civil Engineers]. Journal of Urban Planning and Development,
149(1), 04022044. https://doi.org/10.1061/(ASCE)UP.1943-5444.0000896
Wang, J., & Cao, X. (2017). Exploring built environment correlates of walking distance of transit
egress in the Twin Cities. Journal of Transport Geography, 64, 132–138. https://doi.org/
10.1016/j.jtrangeo.2017.08.013
Wang, L., He, S., Su, S., Li, Y., Hu, L., & Li, G. (2022). Urban neighborhood socioeconomic status
(SES) inference: A machine learning approach based on semantic and sentimental analysis
of online housing advertisements. Habitat International, 124, 102572. https://doi.org/10.
1016/j.habitatint.2022.102572
170

Wang, X., Li, K., & Wu, J. (2020). House price index based on online listing information: The case
of China. Journal of Housing Economics, 50, 101715. https://doi.org/10.1016/j.jhe.2020.
101715
Webb Jamme, H.-T., Bahl, D., & Banerjee, T. (2018). Between “broken windows” and the “eyes
on the street:” Walking to school in inner city San Diego. Journal of Environmental Psychology, 55, 121–138. https://doi.org/10.1016/j.jenvp.2018.01.004
Wegmann, J., Baqai, A. N., & Conrad, J. (2023). Here come the tall skinny houses. Cityscape,
25(2), 171–202.
Weinstein Agrawal, A., Schlossberg, M., & Irvin, K. (2008). How Far, by Which Route and Why?
A Spatial Analysis of Pedestrian Preference. Journal of Urban Design, 13(1), 81–98. https:
//doi.org/10.1080/13574800701804074
Welch, T. F., & Mishra, S. (2013). A measure of equity for public transit connectivity. Journal of
Transport Geography, 33, 29–41. https://doi.org/10.1016/j.jtrangeo.2013.09.007
Werner, C. M., Brown, B. B., & Gallimore, J. (2010). Light rail use is more likely on “walkable”
blocks: Further support for using micro-level environmental audit measures. Journal of
Environmental Psychology, 30(2), 206–214. https://doi.org/10.1016/j.jenvp.2009.11.003
Wissoker, P. (2016). Putting the supplier in housing supply: An overview of the growth and concentration of large homebuilders in the united states (1990–2007). Housing Policy Debate,
26(3), 536–562.
Xiong, C., Cheung, K. S., Levy, D. S., & Allen, M. (2024). The effect of virtual reality on the
marketing of residential property. Housing Studies, 39(3), 671–694. https : / / doi . org / 10 .
1080/02673037.2022.2074971
Yang, L., Chu, X., Gou, Z., Yang, H., Lu, Y., & Huang, W. (2020). Accessibility and proximity
effects of bus rapid transit on housing prices: Heterogeneity across price quantiles and
space. Journal of Transport Geography, 88, 102850. https://doi.org/10.1016/j.jtrangeo.
2020.102850
Yin, L., Cheng, Q., Wang, Z., & Shao, Z. (2015). ‘Big data’ for pedestrian volume: Exploring the
use of Google Street View images for pedestrian counts. Applied Geography, 63, 337–345.
https://doi.org/10.1016/j.apgeog.2015.07.010
171

Yin, L., & Wang, Z. (2016). Measuring visual enclosure for street walkability: Using machine
learning algorithms and Google Street View imagery. Applied Geography, 76, 147–153.
https://doi.org/10.1016/j.apgeog.2016.09.024
Yu, W., Ma, Z., Pant, G., & Hu, J. (2021). The Effect of Virtual Tours on House Price and Time on
Market [Publisher: Routledge eprint: https://doi.org/10.1080/09277544.2021.1876433]. Journal of Real Estate Literature, 28(2), 133–149. https://doi.org/10.1080/09277544.2021.
1876433
Zellner, M., Massey, D., Shiftan, Y., Levine, J., & Arquero, M. J. (2016). Overcoming the LastMile Problem with Transportation and Land-Use Improvements: An Agent-Based Approach. International Journal of Transportation, 4(1).
Zhang, F., Zhou, B., Liu, L., Liu, Y., Fung, H. H., Lin, H., & Ratti, C. (2018). Measuring human
perceptions of a large-scale urban region using machine learning. Landscape and Urban
Planning, 180, 148–160. https://doi.org/10.1016/j.landurbplan.2018.08.020
Zhang, Q., Moeckel, R., & Clifton, K. (2022). Assessing pedestrian impacts of future land use and
transportation scenarios. Journal of Transport and Land Use, 15(1), 547–566.
Zhou, Q., Wang, S., & Liu, Y. (2022). Exploring the accuracy and completeness patterns of global
land-cover/land-use data in OpenStreetMap. Applied Geography, 145, 102742. https://doi.
org/10.1016/j.apgeog.2022.102742
Zhu, L., Burinskiy, E., De la Roca, J., Green, R. K., & Boarnet, M. G. (2021). Los Angeles’
Housing Crisis and Local Planning Responses: An Evaluation of Inclusionary Zoning and
the Transit-Oriented Communities Plan as Policy Solutions in Los Angeles [Publisher: US
Department of Housing and Urban Development]. Cityscape, 23(1), 133–160.
172

Abstract (if available)

Abstract This dissertation discusses issues related to the urban experience, as it relates to transportation, housing searches, and where new housing units are being developed. All four papers are at the intersection of data analytics and urban policy analysis, broadly falling under the heading of Regional Science. They explore how novel data sources – including volunteer-generated geographical information such as OpenStreetMap and scraped online housing listings – can inform our understanding of local-scale urban experiences. Specifically, I explore how to measure the quality of pedestrian experiences, and what role consumption amenities and their accessibility play in housing development and the marketing of residential units, exploring such issues at broader geographic scales than were common previously.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on the economics of cities

PDF

Essays on congestion, agglomeration, and urban spatial structure

PDF

The long-term impact of COVID-19 on commute, employment, housing, and environment in the post-pandemic era

PDF

Spatial and temporal expenditure-pricing equity of rail transit fare policies

PDF

Urban air pollution and environmental justice: three essays

PDF

Who learns where: understanding the equity implications of charter school reform in the District of Columbia

PDF

Productive frictions and urbanism in transition: planning lessons from traffic flows and urban street life in Ho Chi Minh City, Vietnam

PDF

Unraveling decentralization of warehousing and distribution centers: three essays

PDF

Household mobility and neighborhood impacts

PDF

The demand for reliable travel: evidence from Los Angeles, and implications for public transit policy

PDF

Urbanization beyond the metropolis: three papers on urbanization patterns and their planning implications in the contemporary global south

PDF

Property and labor formalization in the age of the sharing economy: Airbnb, housing affordability, and entrepreneurship in Havana

PDF

Location of warehouses and environmental justice: Three essays

PDF

Population and employment distribution and urban spatial structure: an empirical analysis of metropolitan Beijing, China in the post-reform era

PDF

The impact of demographic shifts on automobile travel in the United States: three empirical essays

PDF

Environmental justice in real estate, public services, and policy

PDF

Urban spatial transformation and job accessibility: spatial mismatch hypothesis revisited

PDF

Community development agreements: addressing inequality through urban development projects

PDF

Urban conservation in the Middle Eastern historic cities: globalization and lack of identity

PDF

The Human Performance Intention experiment. A quantum field framework

Asset Metadata

Creator Pilgram, Clemens Andre (author)

Core Title Urban consumer amenities and their accessibility

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Urban Planning and Development

Degree Conferral Date 2024-08

Publication Date 08/02/2024

Defense Date 07/15/2024

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

Tag Housing development,OAI-PMH Harvest,pedestrianism,urban economics,Urban transportation

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Boarnet, Marlon (committee chair), Boeing, Geoff (committee member), Redfearn, Christian (committee member)

Creator Email clemens.pilgram@googlemail.com,pilgram@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113998R9F

Unique identifier UC113998R9F

Identifier etd-PilgramCle-13325.pdf (filename)

Legacy Identifier etd-PilgramCle-13325

Document Type Dissertation

Format theses (aat)

Rights Pilgram, Clemens Andre

Internet Media Type application/pdf

Type texts

Source 20240802-usctheses-batch-1191 (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.

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