University of Southern California Dissertations and Theses

Household mobility and neighborhood impacts

(USC Thesis Other)

Household mobility and neighborhood impacts

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Household Mobility and Neighborhood Impacts

By

Seva Rodnyansky

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA

In Partial Fulfillment of the Requirement for the Degree

DOCTOR OF PHILOSOPHY
URBAN PLANNING AND DEVELOPMENT

August 2018

Dissertation Committee:
Marlon G. Boarnet (Chair)
Raphael W. Bostic
Lisa Schweitzer
Jennifer Ailshire

Copyright 2018: Seva Rodnyansky

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
2

Copyright 2018
By
Seva Rodnyansky
All rights reserved.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
3

Disclaimer

Any opinions expressed in this dissertation are those of the author(s), not official positions of the
California Franchise Tax Board.
The views expressed herein are those of the authors and do not necessarily reflect those of the
Board of Governors of the Federal Reserve System or other System officials.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
4

Acknowledgements
I would like to profoundly thank my advisor and dissertation committee chair, Dr.
Marlon G. Boarnet, for guidance and support throughout my graduate education, dissertation
writing, and job market process. His steadfast belief in my abilities and future propelled me
toward higher quality output, refined ideas, and development from a student into a scholar. I am
grateful to Dr. Boarnet and his wife, Barbara, for being a home away from home in Southern
California.
I would like to sincerely thank Dr. Raphael W. Bostic for being a mentor, dissertation
committee member, and colleague. Dr. Bostic’s patience and care helped develop my scholarship
and ideation, introduced me to conference planning, improved my public speaking, and instilled
confidence in my work and ideas. He has also provided invaluable opportunities to network at
the highest level with policymakers, advisers, and real estate professionals.
I am grateful for the input and advice for dissertation committee members Drs. Lisa
Schweitzer and Jennifer Ailshire. Dr. Schweitzer is a superb teacher and educator from whom I
have learned much about pedagogy and classroom management. Her insightful editing and
wordsmithing has helped my dissertation progress and become more palatable and focused. Dr.
Schweitzer’s cross-disciplinary research viewpoint and pertinent questions have helped me
sharpen my ideas and become open to broader perspectives throughout my time in graduate
school. Dr. Ailshire has taken me in as a coauthor and has inspired an interest in demography
and planning over the life course, and for that I thank her.
I am indebted to Dr. Allen Prohofsky, Chief Economist at the California Franchise Tax
Board for providing me access to crucial tax filing data for the completion of this dissertation.
Dr. Prohofsky and his team, including Chad Angaretis, were instrumental in providing data,
answering questions, and ensuring work product respects confidentiality guidelines.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
5

I would also like to thank the California Community Foundation, specifically Ann Sewill
and Christopher Goett, for financial support for our research team projects on neighborhood
mobility, gentrification, and displacement around Los Angeles County’s transit system. Both
have become advisors and mentors for me in the process. I am also grateful to USC METRANS
and the USC Price Lusk Center for Real Estate for financial support for me and the research
team throughout my dissertation period.
I have benefited greatly from having a supportive professorial and administrative corps
while pursuing my studies at USC Price. While there are too many names to name, I would like
to specifically call out Dr. Dowell Myers, Dr. Gary Painter, Dr. Emma Aguila, Dr. Richard
Green, Dr. Christian Redfearn. Dr Matthew Kahn who have helped me advance as a scholar and
educator. I would like to wholeheartedly thank Julie Kim and Chris Wilson without whom USC
Price would not run.
My fellow graduate students have been a source of support, strength and collaboration,
during the thesis process. I would specifically like to thank Anthony W. Orland, Brian An,
Nathan Hutson, Raùl Santiago-Bartolomei, Johanna Thunell, Soledad de Gregorio, Vincent
Reina, Arthur Acolin, Danielle Williams, Hûe-Tâm Webb Jamme, Julia Harten, Andy Eisenlohr,
Sean Angst, and Gene Burinskiy.
My graduate studies would not have been possible without the loving support of my
whole family: my parents, my grandparents, my sister and my brother-in-law. They are my
foundation.
I would not have been successful in completing my graduate studies nor my thesis
without constant reinforcement and steadfast commitment from my loving wife, Rachel Podell.
She is my rock and keeps me focused and energized even through late nights of writing.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
6

Overview

This dissertation uses tax filing data for Los Angeles County to assess whether patterns of
residential mobility have changed after the introduction of a new rail transit system in the region.
This research provides new ways to measure how often and where households move and to
compare against prevailing neighborhood-level and county-level patterns. This research extends
the empirical literature on the effects of new rail transit systems in gentrifying nearby
neighborhoods and displacing prior residents, including at-risk sub-population such as lower-
income households, families with children, young households, and elderly households. Using a
series of empirical and descriptive methods, this dissertation explores whether new rail station
openings affect neighborhood-level mobility averages (Chapter 1), individual propensity to move
(Chapter 2), and move destination (Chapter 3). Findings suggest that effects of new rail transit
systems on moving are nuanced and heterogeneous with respect to time and geography. The
evidence also suggests that future research should focus not only on moving but on other
methods of coping with rising housing costs including overcrowding, consuming less, or paying
more. This dissertation also sheds light on the need to incorporate mobility statistics into
policymaking and urban planning and provides a potential data source and methodology.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
7

Table of Contents
Household Mobility and Neighborhood Impacts ........................................................................... 1
Acknowledgements ...................................................................................................................................... 4
Overview ....................................................................................................................................................... 6
List of Figures and Tables ............................................................................................................................ 10
Chapter 1: New Ways to Measure Residential Mobility: Assessing the Relationship of Rail Transit on
Neighborhood Mobility Rates in Los Angeles County ............................................................................... 13
Abstract ................................................................................................................................................... 14
Introduction ............................................................................................................................................ 15
Background & Literature Review ............................................................................................................ 17
Data, Study Area & Measurement .......................................................................................................... 21
Measurement, Geocoding and Sample Restriction ............................................................................ 22
Context of Study Area ......................................................................................................................... 27
Calculating Neighborhood Mobility Rates near L.A. Metro Rail Stations ........................................... 29
Selecting Control Neighborhoods ....................................................................................................... 32
Descriptive Analysis of Los Angeles Metro Rail Corridors ...................................................................... 32
Baseline Mobility Rates ....................................................................................................................... 32
Neighborhood Composition ................................................................................................................ 34
Methods and Analysis ............................................................................................................................. 40
Modeling Approach............................................................................................................................. 40
Modeling Extensions ........................................................................................................................... 43
Weighting ............................................................................................................................................ 45
Results ..................................................................................................................................................... 50
Results by Income Category ................................................................................................................ 52
Results from Model Extensions .......................................................................................................... 57
Discussion and Conclusion ...................................................................................................................... 60
Planning and Policy Directions ............................................................................................................ 62
Data Innovation and Limitations ......................................................................................................... 63
Appendix ................................................................................................................................................. 65
1-A. Los Angeles County: Annual Number of Observations with and without Geocodes ................. 65
1-B. Baseline Populations and Mobility Rates by Year ....................................................................... 66
1-C. Area Median Income by Year for the Los Angeles – Long Beach Metropolitan Statistical Area . 70
1-D. Control Neighborhood Location .................................................................................................. 71
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
8

1-E. Parallel Trends Assumption Graphs ............................................................................................. 75
1-F. Regression Results by Income by Transit Line ............................................................................. 79
1-G. Regression Results for Model Extensions ................................................................................... 86
Chapter 2: Do Rail Transit Station Openings Displace Low-Income Households? ................................... 91
Abstract ................................................................................................................................................... 92
Introduction ............................................................................................................................................ 93
Literature Review .................................................................................................................................... 95
What’s the Problem with Moving? ..................................................................................................... 95
Why Do Households Move? ................................................................................................................ 97
Rail Stations and Displacement ........................................................................................................... 99
Literature Gaps & Research Questions ............................................................................................. 101
Data and Study Area ............................................................................................................................. 101
Data ................................................................................................................................................... 101
Study Area ......................................................................................................................................... 103
Treatment and Control Household Selection and Descriptive Statistics .......................................... 104
Methods ................................................................................................................................................ 110
Model Setup ...................................................................................................................................... 110
Sub-population Models ..................................................................................................................... 111
Linear Probability Model and Bootstrapping .................................................................................... 113
Results: .................................................................................................................................................. 114
Rail Station Opening Effects .............................................................................................................. 114
Rail Station Effects on Key Sub-Populations ..................................................................................... 118
Longer-term and Anticipation Effects ............................................................................................... 127
Discussion and Conclusion .................................................................................................................... 135
Summary of Results .......................................................................................................................... 135
Discussion of Results ......................................................................................................................... 136
Limitations and Next Steps ............................................................................................................... 139
Appendix ............................................................................................................................................... 141
2-A. Geocoding .................................................................................................................................. 141
2-B. Non-filing of Taxes ..................................................................................................................... 141
2-C. Sample Restriction Effects ......................................................................................................... 143
2-D. Station-Level Comparisons between Treatment and Station Neighborhoods ......................... 144
2-E. Robustness Tests ........................................................................................................................ 153
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
9

Chapter 3: Does Living near Transit Beget Future Living Near Transit? ................................................. 156
Abstract ................................................................................................................................................. 157
Introduction and Motivation ................................................................................................................ 158
Geographic Move Patterns and New Rail Lines .................................................................................... 159
Statistical Analysis of Transit Exposure ................................................................................................. 166
Mobility Choice Model Setup ............................................................................................................ 167
Location Choice Model Setup ........................................................................................................... 168
Measuring Transit Proximity ............................................................................................................. 168
Descriptive Statistics for Mobility and Location Choices .................................................................. 170
Mobility Choice Results ..................................................................................................................... 172
Location Choice Results .................................................................................................................... 173
Mover Types and Transit Neighborhoods ............................................................................................. 179
Mover Types Over Time .................................................................................................................... 180
Mover Type Characteristics .............................................................................................................. 183
Discussion and Conclusion .................................................................................................................... 187
Appendix ............................................................................................................................................... 189
3-A. Los Angeles Metro Rail Map ...................................................................................................... 189
3-B. Additional Geographic Move Pattern Maps .............................................................................. 190
3-C. Weighted MNL Results for Alternative Transit Definitions ....................................................... 197
References ................................................................................................................................................ 200

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
10

List of Figures and Tables
Figure 1. Thematic Framework Relating Rail Station Investment, Gentrification, and Displacement ....... 17
Figure 2. Unadjusted Annual Mobility Rate by Distance Moved for Los Angeles County 1993-2013 ........ 23
Figure 3. Unadjusted versus Adjusted Annual Mobility Rate for Los Angeles County ............................... 26
Figure 4. Annual Mobility Rate by Housing Tenure .................................................................................... 27
Figure 5. Los Angeles County Metropolitan Transit Authority (L.A. Metro) Rail Lines, 2017 ..................... 28
Figure 6. Annual Average Mobility Rate by Income Category for Los Angeles County .............................. 33
Figure 7. Annual Average Baseline Out-Movement Rates by L.A. Metro Rail Line .................................... 33
Figure 8. Adjusted Baseline Household Population of Los Angeles County by Income Category .............. 34
Figure 9. A-J: Adjusted Baseline Household Population by Income Strata for Treatment and Control
Neighborhoods ........................................................................................................................................... 35
Figure 10. Difference in Difference Regression: Rail Station Impact of Opening, Announcement, and Five
Years after Opening on Neighborhood Baseline Out Mobility Rates, Systemwide and by Line ................ 51
Figure 11. Fixed Effects Regression: Rail Station Impact of Opening, Announcement, and Five Years after
Opening on Neighborhood Baseline Out Mobility Rates, Systemwide and by Line ................................... 52
Figure 12. Rail Effect Impact of Downtown and Overlapping Station Restrictions on Neighborhood Out-
Mobility Rate ............................................................................................................................................... 57
Figure 13. Subway versus Light Rail Station Differences Impact of Rail Effect on Baseline Neighborhood
Out-Mobility ................................................................................................................................................ 58
Figure 14. Highway Median Station Location versus non-Highway Median Station Location Impact of Rail
Effect on Baseline Neighborhood Out-Mobility Stations ............................................................................ 59
Figure 15. Map of Los Angeles City Neighborhoods and Red/Purple Subway Line Stations and Control
Intersections ............................................................................................................................................... 71
Figure 16. Map of Los Angeles City Neighborhoods and Gold Light Rail Line Stations and Control
Intersections ............................................................................................................................................... 72
Figure 17. Map of Expo Phase I, Green, and Blue Line Stations and Control Intersections ....................... 73
Figure 18. A-F: Parallel Trends Graphs: Neighborhood Out-Mobility Rate of Treatment versus Control
Neighborhood by Years Before and After Rail Station Opening for each Rail Corridor ............................. 75
Figure 19. A-C: Predicted Mobility Rates by Income Category for Red/Purple Line................................. 128
Figure 20. A-C: Predicted Mobility Rates by Income Category for Gold Line ........................................... 129
Figure 21. A-D: Predicted Mobility Rates by Number of Dependents by Income Category for Red/Purple
Line ............................................................................................................................................................ 131
Figure 22. A-D: Predicted Mobility Rates by Number of Dependents by Income Category for Gold Line133
Figure 23. A-C: Top Move Destinations (by 5 digit zip code) for Households Moving from L.A. Metro Gold
Line – Pasadena Branch neighborhoods, Before and After Transit service opened (2003) ..................... 162
Figure 24. Two-Stage Discrete Choice Model of Mobility and Location Decisions................................... 166
Figure 25. Proportion of Mover Type by Year for Open L.A. Metro Rail Station Neighborhoods ............ 180
Figure 26. Proportion of Mover Type by Year for L.A. Metro Rail Station Neighborhoods regardless of
whether station is open ............................................................................................................................ 181
Figure 27. Proportion of Mover Type by Year for TPA Neighborhoods .................................................... 182
Figure 28. Proportion of Mover Type by Year for HQTA Neighborhoods ................................................. 182
Figure 29. Map of Los Angeles Metro Rail Lines open in 2013 ................................................................. 189
Figure 30. A-C: Top Move Destinations (by 5 digit zip code) for Households Moving from L.A. Metro Gold
Line – Boyle Heights Branch neighborhoods, Before and After Transit service opened (2003). ............. 190
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
11

Figure 31. A-C: Top Move Destinations (by 5 digit zip code) for Households Moving from L.A. Metro Red
and Purple Line neighborhoods, Before and After Transit service opened (2003) .................................. 193

Table 1. Data and Measurement Issues in Prior Displacement Research .................................................. 20
Table 2. Consecutive Filing Years to Combat Data Dropout Issue for Los Angeles County ........................ 24
Table 3. Number of 9-digit Zip-Codes by Available Vintage and Differences in Distance between Vintages
.................................................................................................................................................................... 24
Table 4. Rail Station Neighborhood Adjusted Baseline Population and Count of Out-Movers by Year:
Breakout of Geocoding Type ...................................................................................................................... 31
Table 5. Household Population Growth Rates by Transit Line by Income Strata, Treatment and Control
Neighborhoods ........................................................................................................................................... 39
Table 6. Model and Weights Comparison: Neighborhood Out Mobility Rates and Rail Station Openings
for All Households. ...................................................................................................................................... 47
Table 7. Model and Weights Comparison: Neighborhood Out Mobility Rates and Rail Station
Announcements for All Households. .......................................................................................................... 47
Table 8. Model and Weights Comparison: Neighborhood Out Mobility Rates and Rail Station Five Years
After for All Households. ............................................................................................................................. 48
Table 9. Rail Station Effects on Neighborhood Out Mobility Rates by Timing for All Incomes .................. 50
Table 10. Income Category Model of Rail Station Effects on Neighborhood Out Mobility Rates .............. 52
Table 11. Rail Effect Magnitude Impact by Rail Corridor by Income .......................................................... 55
Table 12. Summary of Estimates and Impacts of Transit Network Variables on Neighborhood Out
Mobility Rates ............................................................................................................................................. 60
Table 13. Treatment and Control Station Descriptive Statistics: Distance, Year Opened, Sample Size, Out-
Mobility Rate, System and Station Characteristics ..................................................................................... 66
Table 14. DID Model Rail Station Effects on Neighborhood Out Mobility Rates by Timing and Rail Corridor
.................................................................................................................................................................... 79
Table 15. FE Model Rail Station Effects on Neighborhood Out Mobility Rates by Timing and Rail Corridor
.................................................................................................................................................................... 80
Table 16. Transit Corridor Income Category Model of Rail Station Opening on Neighborhood Out
Mobility Rates ............................................................................................................................................. 81
Table 17. Transit Corridor Income Category Model of Rail Station Announcement on Neighborhood Out
Mobility Rates ............................................................................................................................................. 82
Table 18. Transit Corridor Income Category Model of Rail Stations Five Years after Opening on
Neighborhood Out Mobility Rates .............................................................................................................. 84
Table 19. Downtown Exclusion: Rail Effect on Neighborhood Out Mobility .............................................. 86
Table 20. Overlapping Station & Downtown Exclusion: Rail Effect on Neighborhood Out Mobility ......... 86
Table 21. Subway versus Light Rail Comparison: Rail Effect on Neighborhood Out Mobility .................... 87
Table 22. Highway Median versus In-the-Neighborhood: Rail Effect on Neighborhood Out Mobility ...... 88
Table 23. Transit Network Variables: Rail Effect on Neighborhood Out Mobility ...................................... 90
Table 24. Neighborhood Description for Study Area ................................................................................ 104
Table 25. Descriptive Statistics of Household-Year Data for Treatment and Control Households by Transit
Line ............................................................................................................................................................ 108
Table 26. Regression Results for Household Mobility Determinants for Gold and Red/Purple Lines. ..... 116
Table 27. Income Category Regression Results for Gold and Red/Purple Lines ....................................... 119
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
12

Table 28. Income Category Regression Results Separated for Two Branches of the Gold Line ............... 120
Table 29. Dependents Category Regression Results and Dependents with Income Interaction Regression
Results for Gold and Red/Purple Lines ..................................................................................................... 122
Table 30. Age Category Regression Results for Gold and Red/Purple Lines ............................................. 126
Table 31. Effects of Sample Restriction by Transit Line by Year for Households of All Incomes .............. 143
Table 32. Treatment and Control Neighborhood Descriptive Statistics: Distance, Year Opened, Sample
Size, Mobility Rate .................................................................................................................................... 144
Table 33. Treatment and Control Neighborhood Descriptive Statistics: Age Category ........................... 145
Table 34. Treatment and Control Neighborhood Descriptive Statistics: Number of Dependents ........... 147
Table 35. Treatment and Control Neighborhood Descriptive Statistics: Marital Status .......................... 148
Table 36. Treatment and Control Neighborhood Descriptive Statistics: Income Categories ................... 150
Table 37. Treatment and Control Neighborhood Descriptive Statistics: Median Income ........................ 151
Table 38. Regression Results for Household Mobility Determinants for Gold and Red/Purple Lines ...... 153
Table 39. Regression Results for All Incomes for Red/Purple Line excluding stations opened in 1993 (Civic
Center, Pershing Square, Union Station, Westlake / MacArthur Park) .................................................... 154
Table 40. Regression Results by Income Group for Red/Purple Line excluding stations opened in 1993
(Civic Center, Pershing Square, Union Station, Westlake / MacArthur Park) ........................................... 155
Table 41. Descriptive Statistics for Mobility Choice and Location Choice Models ................................... 170
Table 42. Mobility Choice Logit Regression Results .................................................................................. 173
Table 43. Location Choice Weighted Logit Regression Results, displayed as Odds Ratios ....................... 174
Table 44. Alternative Transit Neighborhood Definitions Location Choice Weighted Logit Regression
Results, displayed as Odds Ratios ............................................................................................................. 177
Table 45. Mover Types by Transit Definition ............................................................................................ 179
Table 46. Mover Type Weighted MNL Model Results displayed as Relative Risk Ratios for Open L.A.
Metro Rail transit definition ..................................................................................................................... 185
Table 47. Cohort Age Effects of Mover Type Weighted MNL Model Results displayed as Relative Risk
Ratios for Open L.A. Metro Rail transit definition .................................................................................... 186
Table 48. Cohort Age Effects of Mover Type Weighted MNL Model Results displayed as Relative Risk
Ratios for L.A. Metro Rail with current and future stations transit definition ......................................... 197
Table 49. Mover Type Weighted MNL Model Results displayed as Relative Risk Ratios for TPA ............. 198
Table 50. Mover Type Weighted MNL Model Results displayed as Relative Risk Ratios for HQTA ......... 199

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
13

Chapter 1: New Ways to Measure Residential Mobility: Assessing
the Relationship of Rail Transit on Neighborhood Mobility Rates in
Los Angeles County

Author: Seva Rodnyansky
Co-authors: Dr. Marlon Boarnet, Dr. Raphael Bostic, Raúl Santiago-Bartolomei, Danielle
Williams, Dr. Allen Prohofsky

About the Authors:
Seva Rodnyansky is a PhD candidate in Urban Planning and Development at the Sol Price
School of Public Policy, University of Southern California.
Dr. Marlon G. Boarnet is a professor and chair of the Department of Urban Planning and Spatial
Analysis at the Sol Price School of Public Policy, University of Southern California.
Dr. Raphael W. Bostic the President and CEO of the Federal Reserve Bank of Atlanta and
Professor at the Sol Price School of Public Policy, University of Southern California.
Raúl Santiago-Bartolomei is a PhD candidate in Urban Planning and Development at the Sol
Price School of Public Policy, University of Southern California.
Danielle Williams is a PhD candidate in Public Policy and Management at the Sol Price School
of Public Policy, University of Southern California.
Dr. Allen Prohofsky is the Chief Economist of the California Franchise Tax Board.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
14

Abstract
Households move in and out of neighborhoods every year and this mobility is a natural part of
U.S. urban life. Neighborhoods have higher or lower mobility levels, but these levels do not
change much over time. However, sudden and / or large magnitude perturbations in mobility
rates may increase the number of households moving away from the neighborhoods. When this
outflow is too frequent, unexpected, or involuntary it may detrimentally affect the moving
households, especially for low-income households, a phenomenon known as displacement.
Prior studies have looked at what causes displacement, including gentrification and public
transportation investments (Zuk, Bierbaum, Chapple, Gorska, & Loukaitou-Sideris, 2017; Zuk et
al., 2015). Gentrification is a dual process of changes in neighborhood composition toward
higher-income and more educated households and increases in neighborhood housing prices.
New public transit investment such as rail transit could increase the desirability of a
neighborhood, set off a process of gentrification, and absent an increase in housing supply, cause
displacement of prior residents.
We study the effect of public transportation investments because rail stations represent a large
and relatively long term change to a neighborhood’s built environment, increase transportation
access and lower transportation cost for area residents, and involve a large public financial
allocation to a particular set of neighborhoods. Evidence of a rail – gentrification link already
exists (Zuk et al., 2017), but a rail – displacement link has eluded previous studies due in part to
measurement and data issues.
Using a unique dataset of tax filers, we explore this link in Los Angeles County, which has built
a completely new rapid rail transit system in the past three decades, including opening 93
stations. The data from the California Franchise Tax Board track all Los Angeles County tax
filers from 1993-2013 and provide data on income levels and household locations. This
longitudinal data enables us to measure year-to-year filing location differences of one half-mile
or greater. We assign filers to neighborhoods near rail stations and statically compare mobility
rates to similar counterfactual neighborhoods without rail stations. We also descriptively analyze
transit corridor-level changes in mobility rate and neighborhood income composition over time.
Our study yields three clear conclusions. First, there is little evidence for large scale or
systemwide rail-induced displacement of lower-income households, though some pockets of
displacement exist along some transit corridors. Second, amid high population growth,
neighborhoods along four of five transit corridors have seen a compositional shift toward Middle
and Upper Income households, suggesting a process of long-term gentrification. Third, 21% of
households in Los Angeles County move annually. These results highlight the need to take
neighborhood population dynamics into account for policymakers, urban planners, and
researchers.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
15

Introduction

Households move. Household mobility is a central feature of U.S. urban life. Moving from one
location to another is not inherently good or bad. For an individual household, the event of
moving could improve its lifestyle in terms of housing, neighborhood or employment, or could
lead to decreased health or educational outcomes (Morris, Manley, & Sabel, 2018). The specifics
of each households’ case determine the outcome. However, too frequent, unexpected, or
involuntary moves have been found to be detrimental to health and well-being after the move
and later in life (Jelleyman & Spencer, 2008; Morris et al., 2018; Goldsmith, Britton, Reese, &
Velez, 2017; Cox, Henwood, Rodnyansky, Wenzel, & Rice, 2017). For at-risk sub-populations
such as low-income, minority, or elderly households, forced, frequent, or unexpected moves are
even more likely to be detrimental (Jelleyman & Spencer, 2008; Morris et al., 2018; Goldsmith
et al., 2017; Cox et al. 2017).

Households live in neighborhoods. Just like households, neighborhoods differ in mobility levels,
due to differences in housing stock and tenure, the demographic and socioeconomic composition
of the neighborhoods. Neighborhoods with both high and low mobility rates can be stable or
unstable (Rossi, 1955). Most neighborhoods tend to be stable over time across various
characteristics (e.g., population density, income, education, house prices) relative to other
neighborhoods (Malone & Redfearn, 2018). Similarly, both high mobility and low mobility
neighborhoods display stability over time (Rossi, 1955); this is the case for both the U.S. and Los
Angeles County (Figures 4, 6, and 7 below). Sudden and / or large magnitude perturbations in
mobility rates may increase the number of households moving away from the neighborhoods.
When this outflow is too frequent, unexpected, or involuntary it may detrimentally affect the
moving households, especially for low-income households, who may be constrained in where
else they can move. We name this outflow displacement when the mobility rate is larger than the
average for that neighborhood.

We purposefully use the term displacement to connect to the broader conversation on whether
and how neighborhoods change and who is affected. Scholars have associated the concept of
displacement as the result of other urban processes such as neighborhood gentrification (Grier &
Grier, 1979) and neighborhood abandonment (Marcuse, 1985). Here, gentrification has been
defined as the change in neighborhood composition toward a greater proportion of young, white,
higher-income, and higher educational attainment households, and away from low-income,
working-class, minority, and elderly households, often in older and disinvested neighborhoods
within an urban area (Marcuse, 1985). Gentrification creates a competition for often scarce levels
of housing. Existing residents who lose this competition for housing may be displaced out of the
neighborhood.

Scholars and advocates have documented many causes of gentrification (Zuk, Bierbaum,
Chapple, Gorska, & Loukaitou-Sideris, 2017; Zuk et al., 2015). Scholars have also attempted to
empirically connect gentrification to displacement, but have faced mixed results (Zuk et al.,
2017; Zuk et al., 2015). We focus on one reported cause of gentrification – public investment in
rail transit – and test whether it affects displacement directly.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
16

We choose public transportation investments for several reasons. First, rail stations represent a
large and relatively long term change to a neighborhood’s built environment. Second, they
increase transportation access and lower transportation cost for area residents. Third, they
represent a large public financial allocation to a particular set of neighborhoods. Fourth, there is
evidence that public transit attracts higher-income, young, and higher-educated households (Zuk
et al., 2017; Freeman, 2005; Kahn, 2007). Fifth, rail transit tends to increase nearby land values
and housing prices (e.g., Higgins & Kanaroglou, 2016; Bartholomew & Ewing, 2011). This in
turn can lead to the development of new housing stock near stations priced much higher than
neighborhood averages (Boarnet et al., 2015). Taken together, public investment in rail
transportation changes the neighborhood and may be useful or desirous to the existing
population, but attracts other households to the neighborhood and increases local housing prices.
This setup indicates the potential association between rail transit and displacement.

Most previous studies of displacement had not considered rail at all (e.g., Freeman, 2005; Vigdor
et al., 2002; Freeman & Braconi, 2004; Newman & Wyly, 2006; Ellen & O’Regan, 2011). Those
that had considered rail either had not looked for displacement (e.g., Kahn, 2007; Lin, 2002;
Grube-Cavers & Patterson, 2015; Dawkins & Moeckel, 2016), had found mixed results on the
relationship between rail and displacement (Ong, Zuk, Pech, and Chapple, 2017), or had not
considered neighborhood-level displacement (Delmelle & Nilsson, 2018). Data and
measurement limitations have precluded the above studies from adequately assessing the
relationship between and impact of rail transit on displacement as embodied by changes in
neighborhood mobility (Rayle, 2015; Zuk et al., 2017; Zuk et al., 2015).

To address these data and measurement limitations, our study uses a rich longitudinal database of
more than 100 million tax filer records over a 21 year time period to assess the relationship
between rail transit and neighborhood mobility rates to determine whether rail transit is
associated with displacement. This dataset from the California Franchise Tax Board (FTB)
enables us to track annual neighborhood mobility rates for Los Angeles County, California for
the years 1993 to 2013. We choose Los Angeles County because of its immense investment in
rail transit over the past three decades – a new system with 93 stations opened since 1989 – and
its high proportion of renters (Abodo.com, 2017) who are more mobile than homeowners in
general (Clark & Huang, 2003), and more sensitive to increases in housing costs (Ioannides &
Kan, 1996; Rossi, 1955; Varady, 1983), and have lower moving costs than owners (van
Ommeren, Rietveld, & Nijkamp, 1999).

We use the FTB data to answer our main research question: Does the introduction of a rail
station to a neighborhood change the neighborhood mobility rate, and if so, does this indicate
displacement? We also explore several related questions: a) does the rail station effect occur at
the station’s announcement, opening, or five years later? b) are lower-income households more
affected? c) how does transit technology, station location within a neighborhood or transit
system, and system size moderate the rail station effect?

Due to the methodological issues and lack of adequate data, one of the most pressing policy
issues related to new rail transit is being debated with careful, but ultimately indirect evidence.
Our research is the first study to assess the relationship between rail transit investment and
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
17

displacement in transit-proximate neighborhoods using two decades of longitudinal household
data and a very large sample size.

The rest of the paper reviews previous attempts in more detail, describes the data and
measurements, introduces the study area, provides a descriptive analysis of the transit corridors
in question, lays out the analytical approach, and concludes with a discussion of the results.

Background & Literature Review
We hypothesize that public investment in rail transit leads to neighborhood-level displacement of
existing, lower-income residents. This occurs through a mechanism of neighborhood-level
gentrification as laid out in the thematic framework in Figure 1. We examine each
interconnection in Figure 1 to situate our analysis in the prior literature. To do so, we rely on two
extensive literatures by Zuk et al. (2015, 2017) that document the evidence on the
interconnections between rail investment, gentrification, and displacement.

Figure 1. Thematic Framework Relating Rail Station Investment, Gentrification, and Displacement

New rail investment may induce a change in neighborhood household composition, because
higher-income, more educated, younger, or non-minority households prefer rail transit as a fast
and reliable alternative to driving and because it signals physical reinvestment in a previously
disinvested neighborhood (Zuk et al., 2015; Pollack et al., 2010). Numerous studies provided
evidence that in-movers to rail station neighborhoods change the neighborhood composition
across one or more dimensions (Zuk et al., 2017; Freeman, 2005; Baker & Lee, 2017;
McKinnish, Walsh & White, 2008; Kahn, 2007; Ellen & O’Regan, 2010; Grube-Cavers &
Patterson, 2015).
At the same time, both the increased transportation access and the reinvestment signal have been
shown to increase land and housing prices in rail station neighborhoods (Zuk et al., 2015). This
has occurred because previous rents were low due to disinvestment and then increased as the
train line was built (Smith, 1979) or because the new transportation investment brought along
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
18

new residential or commercial construction, which raised area housing prices (Knaap, Ding &
Hopkins, 2001). Moreover, the changing neighborhood composition and increase in prices
reinforce each other. Scholars have generally confirmed that residential real estate appreciates to
varying degrees after rail stations open in numerous U.S. cities large and small: Phoenix
(Atkinson-Palombo, 2010; Golub, Guhathakurta, & Sollapuram, 2012), Buffalo (Hess and
Almeida, 2007), Atlanta (Immergluck, 2009), San Diego (Duncan, 2011), and in multi-city
analyses (Higgins & Kanaroglou, 2016; Bartholomew & Ewing, 2011). However, measurements
of land and housing price appreciation due to transit were found to be sensitive to station location
in the metropolitan area (Dong, 2017), number of years after opening (Pilgram & West, 2017),
and specification (Redfearn, 2009). Through these two interlaced mechanisms, most studies have
found that rail transit proximity and new rail transit openings are positively associated with
gentrification.
Scholars have further theorized that gentrification (neighborhood composition change and/or
increased housing prices) leads to neighborhood-wide displacement (Marcuse, 1985).
Gentrifying neighborhoods do not have enough housing units to meet increased demand and so
existing lower-income, minority, and working class households are priced out (Marcuse, 1985).
Even when additional housing units are added, they are often priced to reflect the new increased
rents and remain unaffordable for prior or existing residents (Zuk et al. 2015). Displacement can
have more forceful forms as existing residents are coerced to leave, evicted, foreclosed, or have
their rental units converted to condominiums (Grier & Grier, 1979; Zuk et al., 2017). Marcuse
(1985) defined additional forms of displacement: exclusionary displacement where a household
is unable to move in to the neighborhood and displacement pressure where extant neighborhood
residents have a broad feeling that the neighborhood is changing, spurring some of them to
consider moving. Renter households in particular are more sensitive to rising housing costs
(Ioannides & Kan, 1996; Rossi, 1955; Varady, 1983). We hypothesize that renters feel the brunt
of displacement more than homeowners.
When households are priced out they face several options: paying more, doubling-up with others,
moving to smaller and thus cheaper units if available, moving out of the neighborhood, or
becoming temporarily or permanently homeless. While each of these options deserves careful
examination, much of the literature has focused on the moving option, likely due to data scarcity
on the other options. Our analysis follows suit and also focuses on moving and not the other
options.
In addition, our study area (Los Angeles County) has a high proportion of renters (Abodo.com,
2017) and neighborhoods near rail station areas have even higher proportions of renters (Boarnet
et al., 2015). Since renters are more mobile (Clark & Huang, 2003) and have lower moving costs
than homeowners (van Ommeren, Rietveld, & Nijkamp, 1999), we hypothesize that renters will
choose the move option, making our focus on moves even more salient.
Zuk et al. (2017) documented ten studies which have tried to relate gentrification and
displacement in a variety of contexts and at multiple spatial scales (municipality, metropolitan
area, and national) (Zuk et al., 2017 Table 2). Of these studies, five have found that gentrification
was related to or predicted displacement (NIAS 1981; Schill, Nathan, & Persaud, 1983;
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
19

Atkinson, 2000; Freeman, 2005; McKinnish et al., 2010) and five have found that it did not
(Vigdor et al., 2002; Freeman & Braconi, 2004; Ellen & O’Regan, 2011; Ding, Hwang, &
Divringi, 2016; Sims, 2016). Additionally, Newman & Wyly (2006) and Wyly, Neman,
Schafran, and Lee (2010) had found a positive relationship between gentrification and
displacement, but Freeman, Cassola, and Cai (2016) had not. Thus, evidence on the link between
displacement and gentrification continues to be mixed in quantitative work.
Only two studies focus on the link between rail transit and displacement, via a gentrification
mechanism. Ong, Zuk, Pech, and Chapple (2017) developed a simulation model to test the effect
of proximity to rail stations, the proportion of old housing, and employment density for census
tract neighborhoods on the likelihood of gentrification and on the loss of low-income households
(equated with displacement) in the San Francisco Bay Area. Their simulation model correctly
predicted actual gentrified tracts for 73 of 85 tracts (86%) which gentrified from 2000-2013, but
incorrectly predicted gentrification in 383 of 512 tracts (75%) that did not gentrify during the
same time period. On displacement, the model correctly predicted 470 of 537 tracts (88%)
which experienced displacement of low-income households, but incorrectly predicted 769 of
1009 tracts (76%) which did not experience displacement. A different study used the geocoded
version of Panel Study of Income Dynamics (PSID) to measure the effect of new rail station
openings on displacement of individual households (Delmelle & Nilsson, 2018). They tracked
about 1000 households across 55 metropolitan areas that have built new rail stations since 1970,
measuring move likelihoods using a logit model. They found that new rail station openings were
not associated with a higher propensity to move, controlling for household and neighborhood
characteristics (Delmelle & Nilsson, 2018).

Overall, the evidence for a gentrification to displacement link is inconclusive at best and the
evidence for a rail-displacement is nascent. Ong et al.’s (2017) simulation model overpredicted
rail-induced displacement as often as it predicted it correctly and focused on legacy rail
transportation access in the San Francisco Bay Area during 2000-2013, a time period where few
new rail stations were opened. Their study also did not provide a baseline measure for average
mobility even at a neighborhood level. Delmelle and Nilsson (2018) provided the closest
evidence to date on the question, looking at a longitudinal dataset over across many cities.
However, their small sample size stretched across four decades and fifty-five cities and was very
sensitive to the specific households in the survey. While they did provide a national baseline
mobility, mobility rates differ by neighborhood and by metropolitan area, possibly making their
impact assessment too general.

To sum up this literature, Rayle (2015) pointed to four hypotheses for why scholars had not
found a rail-displacement link: 1) data and measurement issues, 2) a narrow definition of
displacement focused only on physical moves out of the neighborhood and not on exclusionary
displacement (where low income households are excluded from moving into the vacated unit
because of newly high costs) or displacement pressure (current household devoting a higher
proportion of income toward housing costs) (Marcuse, 1985), 3) existing households coping with
increased cost of housing through lower transportation cost, and 4) either community opposition
to rail transit projects and plans or advocacy group effectiveness in the political process deters or
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
20

defers displacement pressure. While, each of these hypotheses merit further explanation, our
paper tries to improve on the data and measurement issues.

Zuk et al. (2015) posited four methodological shortcomings of these studies which preclude a
more coherent assessment of the rail transit – displacement relationship: 1) inconsistent
displacement and gentrification definitions, 2) differences in types of control, 3) analysis time
scales that are insufficient to measure the full extent of neighborhood change, 4) lack of
consistency in measuring impact. Concurring with Zuk et al. (2015), we briefly explore the
methodological weaknesses of the above papers.

Table 1 summarizes the measurement issues that befall the quantitative displacement studies
listed above. A few studies use spatial scales that are too large to draw neighborhood inferences
such as PUMAs (Vigdor et al., 2012) and New York City sub-boroughs (Freeman & Braconi,
2004; Newman & Wyly, 2006; Wyly et al., 2010). Other studies only consider displacement
within a specific municipality and not the whole metropolitan area, neglecting that mobility
patterns often transcend jurisdictional boundaries. Certain studies do not measure displacement
estimates relative to a control group or even a baseline move rate, making it difficult to identify
any changes as actual displacement. One study uses a predictive simulation model with relatively
poor performance when compared to actual events. Studies which use census data exclusively
may miss displacement altogether, because large time differences smooth over effects or because
of an inability to measure year to year changes. Many of these studies use relatively sparsely
sampled data. While this data may be representative at the national or metropolitan area level, it
is potentially too small to draw neighborhood-level inferences. Moreover, all but two of these
studies do not consider the relationship between rail and displacement at all.

Table 1. Data and Measurement Issues in Prior Displacement Research
Data and Measurement Issue Papers Affected
Spatial scale too large Vigdor et al. (2002), Freeman & Braconi (2004),
Newman & Wyly (2006), Wyly et al. (2010)
Municipal boundary defines
geography
Freeman & Braconi (2004), Newman & Wyly
(2006), Wyly et al. (2010), Ding et al. (2016), NIAS,
Schill et al. (1983), Sims (2016)
No control group or baseline NIAS (1981), Schill et al. (1983), Sims (2006), Ong
et al. (2017)
Poor model performance relative to
baseline
Ong et al. (2017)
Long time period between
observations
Freeman et al. (2016), McKinnish et al. (2010),
Atkinson (2000)
Sample size too small Ding et al. (2016), Freeman (2005), Delmelle &
Nilsson (2018), NIAS (1981), Schill et al. (1983),
Ellen & O’Regan (2011)
Lack of neighborhood level analysis Ellen & O’Regan (2011), Delmelle & Nilsson (2018)

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
21

In light of the above issues, it is our view that research that can establish whether, and to what
extent, rail transit causes displacement of low income households must have four characteristics:
the research should track individual households by income, instead of focusing on housing units
as some previous research has done; the research should analyze household migration into and
out of rail-proximate neighborhoods before and after rail stations are built; the research should
compare transit neighborhood migration with a reasonable “no rail transit” counterfactual,
ideally using the same dataset; and the research should have frequent, preferably annual, data, to
allow fidelity in the ability to assess patterns over time.
No previous study in the literature has combined those four characteristics. As a result, one of the
most pressing policy issues related to new rail transit is being debated with careful, but
ultimately indirect evidence. This study is the first to combine these four innovations to test the
causal link between rail transit investment and displacement in transit-proximate neighborhoods.

Data, Study Area & Measurement
Measuring residential mobility at a neighborhood level is a challenging task in the U.S. context,
given the lack of sufficient data at a fine temporal and spatial scale. Most data sources used in
previous studies have had at least one of the following issues: long timeframe between responses
(decennial census), non-representativeness at the neighborhood level (American Community
Survey, U.S. Current Population Survey, Panel Study of Income Dynamics, American Housing
Survey, New York City Housing and Vacancy Survey, etc.), or lack of comparison group within
same metropolitan area (Making Connections Survey).
We rectify these issues by using 21 consecutive years of tax filing data at the household level
obtained from the California Franchise Tax Board (FTB). The data universe contains all
individuals who have ever filed California taxes from 1993 to 2013 in Los Angeles County,
creating a dataset of over 100 million records, or about 4.8 million records per year. For
households who moved in or out of the county, records from any years in which they lived
outside of Los Angeles County and filed California tax returns, even if from outside of
California, are also in the dataset. Household-specific identifiers enable year-to-year tracking.
1

The longitudinal dataset includes information available on the California tax return including the
households’ income, state taxes paid, approximate location, and other characteristics in every
year the household is in the data. From this dataset, we construct a station-area longitudinal panel
of household mobility patterns as a measure of neighborhood-level displacement.
Previous longitudinal research using income tax data has proven its reliability and precision over
cross-section or aggregate data in topics that range from historical income inequality (Piketty &
Saez, 2003; Atkinson, Piketty, & Saez, 2011; Saez & Zucman, 2016) to intergenerational
mobility (Solon, 1992; Corak & Heisz, 1999; Chetty et al., 2014). Another advantage of using
longitudinal income tax data is that transitory fluctuations of earnings can be avoided (Österberg,

1
Certain households file in earlier or later years. We detrend these observations to their nominal filing year to obtain
a more balanced panel. We also remove any duplicate entries.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
22

2000). We thus feel comfortable using tax data and extending it to the study of residential
mobility.

Measurement, Geocoding and Sample Restriction
To detect whether a household has moved, we rely on changes in tax filing location provided in
the FTB data. Due to confidentiality constraints, the FTB suppresses filers’ filing addresses.
Instead, the FTB provided location information at the 9-digit zip code level for all 9-digit zip
codes with at least 10 filers,
2
and 5-digit zip codes for the remaining filers. 9-digit zip codes are
unique U.S. Postal Service (U.S.P.S.) designations for a set of addresses within one city block,
block-face, set of buildings, or individual building, and correspond roughly to a specific portion
of a U.S.P.S. delivery route. These 9-digit zip codes identify a tax filer’s filing location to within
one city block.
To geocode 9-digit zip code locations, we match the 9-digit zip codes with latitude and longitude
coordinates using conversion files from Geolytics, Inc., a private provider of location data. 9-
digit zip code locations may change over time, based on U.S.P.S. needs. To correct for this, we
use Geolytics data for 9-digit zip codes in as many years as available (which were 2000, 2004,
2007, 2009, 2012, 2013, and 2014) and match coordinates based on the closest available data
year and on latitude and longitude availability. For those 9-digit zip codes not in Geolytics or for
households without a 9-digit zip code, we used the latitude and longitude of the centroid of the 5-
digit zip code.
3
Of the 102 million observations (4.8 million per year), we dropped 1.4 million
total (64,000 per year) which had neither 5- nor 9-digit zip codes in any year (See Appendix 1-
A).
After geocoding, we compute a mobility rate for year t using equation 1:
𝐸𝑞𝑢𝑎𝑡𝑖𝑜𝑛 1: 𝑈𝑛𝑎𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑀𝑜𝑏𝑖𝑙𝑖𝑡𝑦 𝑟𝑎𝑡𝑒 𝑡 =
(𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑖𝑙𝑒𝑟𝑠 𝑐 ℎ𝑎𝑛𝑔𝑖𝑛𝑔 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑓𝑟𝑜𝑚 𝑦𝑒𝑎𝑟 𝑡 𝑡𝑜 𝑡 + 1)
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑖𝑙𝑒𝑟𝑠 𝑖𝑛 𝑦𝑒𝑎𝑟 𝑡
However, because not every household files California taxes in every year, the unadjusted
mobility rate is sensitive to households dropping out of the data, as dropped out households do
not show a location for the dropped out year. Reasons for dropping out of the dataset include
death, moving outside of the U.S. or out of California and not maintaining any tax relationship in
the state, income dropping below the filing threshold, becoming a dependent of another filer,
non-compliance or other circumstances.
Estimates of tax non-compliance (non-filing), and thus not being in the data, range from 10 –
16% in the mid-2000s at the federal level (IRS, 2010; IRS, 2012; IRS, 2017) and about 11%
during the same time period in California (FTB, 2006; FTB, 2017). Households are only required
to file taxes federally if their annual incomes are above a certain level. In 2013, this threshold
was $20,000 for families and $10,000 for individuals federally (IRS, 2013), and $25,125 for

2
This is done to protect the identity of the tax filers.
3
We obtained 5-digit zip code latitude and longitude coordinates for the U.S. and for California from Boutell.com’s
”Free Zip Code Latitude and Longitude Database” and from Zip-code.com’s “California ZIP Code database”.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
23

families and $12,562 for individuals in California (FTB, 2013). Income-based filing thresholds
are indexed to inflation and thus vary slightly from year to year. A study of federal tax non-filers
using 2005 data estimated that 77% households who did not file had annual incomes below
$20,000, the federal income mandate (Lawrence, Udell & Young, 2011). This shows that lack of
mandate to file, not blatant tax evasion, is the primary reason not to file taxes. We are not
concerned, however, that our data is missing a significant portion of the lowest-income
households, because many households with incomes below the mandatory filing threshold still
file taxes to take advantage of tax credits. For example, 75% of eligible California households
claim the Earned Income Tax Credit (EITC), a tax credit for low and lower-middle income
working households often with children, which requires them to file a tax return (IRS, 2014).
Thus, many of the lowest-income households still file taxes, even if they fall below the
mandatory filing threshold.
Throughout the geocoded dataset, about 15% of households each year drop out: they appear in
year t but not in year t+1. They may, however, reappear in subsequent years. Reappearance is
especially common for lower-income households teetering above and below the filing income
threshold. The dropout proportion varies little from year to year (see Figure 2), though the
dropout rate has trended downward very slightly from 1993 to 2013.

Figure 2. Unadjusted Annual Mobility Rate by Distance Moved for Los Angeles County 1993-2013
Source: author calculations of FTB data

We combat the dropout issue by adjusting our measurement of mobility to only count
consecutive filers in both the numerator and denominator. Consecutive filers are those whose
filing record appears in the dataset for multiple consecutive years, even if it is outside of Los
Angeles County or California. Increasing the number of consecutive years in the data decreases
number of available observations (Table 2). To use the largest amount of households in this
study, while enabling a clean measurement not subject to dropouts, we use only filers that appear
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Dropped Out of Data Moved Fewer than 100 Meters Moved 100 Meters - 0.5 miles
Moved Further than 0.5 Miles Did Not Move
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
24

in at least two consecutive years in the data, from here on out. This reduces the overall number of
observations by 15% to 81.3 million across the time period or 4.1 million annually.
Table 2. Consecutive Filing Years to Combat Data Dropout Issue for Los Angeles County

Number of Consecutive Filing Years Los Angeles County
Number of
Consecutive
Years in Data
T
(e.g.
1993)
T+1
(e.g.
1994)
T+2
(e.g.
1995)
T+3
(e.g.
1996)
T+4
(e.g.
1997)
% of Total
Geocoded
Observations
Average
Annual Number
of Observations
1

100% 4.8 million
2

85% 4.1 million
3 74% 3.6 million
4

65% 3.2 million
5

58% 2.9 million
Source: author calculations of FTB data

The U.S.P.S. periodically introduces, removes, or changes the location of 9-digit zip-codes based
on business needs. The geographic coordinates of 9-digit zip-codes collected by Geolytics, Inc.
reflect these changes across the 2000-2014 time frame. The number of 9-digit zip-codes in
California more than doubled from 2000 to 2014 (Table 3). Changes in location of the same-
numbered 9-digit zip-code rarely exceed a distance of 1 kilometer (in a year to year comparison
to the latest available 2014 data); most often, changes reflect at most a 100 meter or less
difference in distance (Table 2). Since not every 9-digit zip-code is present in every vintage of
the Geolytics geographic coordinates files, we match each tax filer to the closest year possible.
For example, observations from 2001 FTB tax data were first matched with 2000 Geolytics
coordinates; if there was no match to 2000 coordinates, 2004 Geolytics coordinates were used; if
2004 was not a match, then 2007 Geolytics coordinates were used, etc., until a match was found.
Table 3. Number of 9-digit Zip-Codes by Available Vintage and Differences in Distance between
Vintages
4

2000 2004 2007 2009 2012 2013 2014
Unique Coordinates 1,493,268 1,184,266 2,647,236 2,719,413 2,857,990 2,910,488 3,218,139
Chained Match 3,101,577 3,110,990 3,209,568 3,213,871 3,217,684 3,218,139 3,218,139
Difference in Distance
to 2014

No difference 0.6% 0.5% 14.3% 47.0% 52.9% 94.7% 0%

4
We compute differences in distance directly using the spherical law of cosines formula, instead of using a GIS-
based buffer tool. The spherical law of cosines computes distance calculations between two points on a coordinate
system while taking into account the curvature of the earth (“Spherical Law of Cosines” (n.d.). The formula for the
distance between points (x 1,y 1) and (x 2,y 2) is: acos(sin(x 1)*sin(x 2)+cos(x1)*cos(x2)*cos(y2-y1))*earth’s radius. The
earth’s radius in Los Angeles is 6371.461 km (“Earth Radius Latitude Calculator” (n.d.)). Calculations yield similar
results to the well-established Haversine formula (Shumaker and Sinnott, 1984) for distances above 1 meter, which
are the distances we are concerned about in our data. See: https://webtrough.wordpress.com/2011/09/19/latlon-
distance-formula-in-excel-haversine-and-spherical-law-of-cosines/,
http://gis.stackexchange.com/questions/4906/why-is-law-of-cosines-more-preferable-than-haversine-when-
calculating-distance-b.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
25

0-10 meters 14.4% 12.7% 29.7% 18.9% 20.0% 1.6% 0%
10-100 meters 59.4% 58.7% 45.8% 24.9% 19.4% 2.3% 0%
100-1000 meters 24.7% 27.3% 9.4% 8.6% 7.3% 1.2% 0%
> 1 km 1.0% 0.8% 0.8% 0.7% 0.3% 0.2% 0%
Total 100% 100% 100% 100% 100% 100% 0%
Source: Author calculations of Geolytics, Inc. data

Changes in 9-digit zip-code location due to U.S.P.S. business needs cause some disturbances in
our geocoding method at short distances. Specifically, the number of geocoding-related 9-digit
zip-code coordinate changes of 0-100 meters spiked in 2003-2004, 2006-2007, and 2011-2012,
and of 100-800 meters miles spiked in 2006-2007 (see Figure 2). As a result, moves under 800
meters (0.5 miles), likely constitute noise from geocoding and likely do not reflect actual
household behavior. Since we are unable to adequately distinguish coordinate changes from
actual household moves for distances of less than 0.5 miles, we define true household moves as
any moves of at least 0.5 miles for the remainder of our analyses. This restriction reduces the
annual number of location changes by about 15 percent of all Los Angeles County observations.
Putting together the two sample restrictions, we revise our definition of the household mobility
rate in equation 2:
𝐸𝑞𝑢𝑎𝑡𝑖𝑜𝑛 2: 𝐴𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑀𝑜𝑏𝑖𝑙𝑖𝑡𝑦 𝑟𝑎𝑡𝑒 𝑡 =
(
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑖𝑙𝑒 𝑟𝑠 𝑐 ℎ𝑎𝑛𝑔𝑖𝑛𝑔 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑓𝑟𝑜𝑚 𝑦𝑒𝑎𝑟 𝑡 𝑡𝑜 𝑡 + 1 𝑏𝑦 𝑚𝑜𝑟𝑒 𝑡 ℎ𝑎𝑛 0.5 𝑚𝑖𝑙𝑒𝑠 | 𝑒𝑥𝑖𝑠𝑡𝑖𝑛𝑔 𝑖𝑛 𝑑𝑎𝑡𝑎 𝑖𝑛 𝑦𝑒𝑎𝑟𝑠 𝑡 𝑎𝑛𝑑 𝑡 + 1
)
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑖𝑙𝑒𝑟𝑠 𝑖𝑛 𝑦𝑒𝑎𝑟 𝑡 𝑡 ℎ𝑎𝑡 𝑒𝑥𝑖𝑠𝑡 𝑖𝑛 𝑡 ℎ𝑒 𝑑𝑎𝑡𝑎 𝑖𝑛 𝑦𝑒𝑎𝑟𝑠 𝑡 𝑎𝑛𝑑 𝑡 + 1

Thus, the adjusted mobility rate counts all households who move at least 0.5 miles for
households in the data for at least two consecutive years t and t+1 and divides this by the number
of possible movers: the households who exist in the data in years t and t+1.
We compare the adjusted and unadjusted annual mobility rate for Los Angeles County in Figure
2. On average the countywide unadjusted annual mobility rate is 49%, nearly double the adjusted
annual mobility rate is 21%. We prefer the adjusted rate because it reduces errors from data
dropouts and potential geocoding noise. Nevertheless the adjusted rate provides a lower-bound
average, since some data dropouts may in fact signify moves which can not be adequately
tracked by the FTB data, though this is likely a small proportion of all data dropouts.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
26

Figure 3. Unadjusted versus Adjusted Annual Mobility Rate for Los Angeles County
Source: Author calculations of FTB data
To externally validate our mobility rate estimates and our adjusted methodology, we compare
our FTB data-derived rates with those from other existing estimates. The U.S. Census Bureau’s
March edition Current Population Survey (CPS) annually asks a representative sample of
100,000 U.S. residents whether they moved in the past year. Based on the CPS, the annual
national mobility rate averaged about 14% from 1988-2016, declining from 18% in 1988 to 11%
in 2016 (Figure 4) (U.S. Census Bureau, 2016). The 2015 American Community Survey from
the U.S. Census Bureau suggested a 12% national mobility rate (U.S. Census Bureau, 2015). In
comparison, our 21% annual mobility rate average for Los Angeles County is well above these
national averages. The declining pattern over time is similar: in Los Angeles County, the
mobility rate declined from 30% in 1993 to 18% in 2013.
We conjecture that average mobility rate differences between the U.S. and Los Angeles County
are due in large part to differences in housing tenure. Figure 4 shows that move rates for U.S.
renters are 15-20 percentage points higher than for U.S. homeowners, ranging from 35% in 1988
to 18% in 2016 (Figure 4) (U.S. Census Bureau, 2016). Los Angeles County (and city) is one of
the largest renter markets in the United States with most households renting rather than owning
their residence (Abodo.com, 2017), which may explain the relatively higher household mobility
rate.
Additionally, our 21% average annual mobility rate for Los Angeles County squares well with
other survey studies. For example, Clark and Ledwith (2006) found an 18% annual mobility rate
for the Los Angeles using a sample of 2,644 households in sixty-five neighborhoods in Los
Angeles County over 2002-2006. Our findings are also close to Coulton, Theodos, and Turner
(2012), who found a 19% annual mobility rate (57% over three years) in their survey of 10 low-
income and changing neighborhoods in metropolitan areas across the U.S. Taken together, we
are encouraged that our county-level adjusted mobility rates logically relate to prior findings and
national level statistics. This shows the potential for this administrative dataset to yield
reasonable results in line with other sources.
0%
10%
20%
30%
40%
50%
60%
70%
80%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Unadjusted Mobility Rate (no restrictions)
Adjusted Mobility Rate (adjusted for geocoding noise and data drop outs)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
27

Figure 4. Annual Mobility Rate by Housing Tenure
Source: U.S. National (U.S. Census Bureau, Current Population Survey, 1988-2016); Los Angeles
County (Author calculations of FTB data)

Context of Study Area
The Los Angeles metropolitan area presents an ideal study area for analyzing residential
mobility, rail transit, and displacement. From rail transit perspective, prior to 1990, Los Angeles
County had not had any intra-urban rail transit service for decades. Since then the Los Angeles
Metropolitan Transit Authority (L.A. Metro) has opened 80 new subway and light-rail rapid
transit stations along six rail lines from 1990-2013, 13 more from 2013-2017 (see Figure 5 for
map), and an additional 17 stations are currently under construction (Boarnet et al., 2016). This
buildout amounted to about half of the U.S. spending on new rail transit (L.A. Metro 2009, p.
23). 21 percent of L.A. Metro’s budget between 2005-2040 has gone/will go toward rail transit
capital and operations expenditures (L.A. Metro 2009, p. 23). These rail corridors connect
regional and local employment hubs including Downtown Los Angeles, Downtown Long Beach,
Santa Monica, Pasadena, Hollywood, and the Eastern San Fernando Valley among others. The
neighborhood surrounding these stations have median household incomes below the county
average, and higher than county average proportions of minority, foreign-born, and renter
households (Boarnet et al., 2015). These neighborhoods also have some of the higher population
densities in Los Angeles County (Boarnet et al., 2015).
0
5
10
15
20
25
30
35
40
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
U.S. All Households U.S. Homeowners U.S. Renters L.A. County All Households
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
28

Figure 5. Los Angeles County Metropolitan Transit Authority (L.A. Metro) Rail Lines, 2017
Source: Author ArcGIS maps using SCAG shapefiles (SCAG, 2015)
Besides massive investments in rail transit, Los Angeles County has faced severe housing
undersupply since 1980, despite robust housing demand (Taylor, 2015). Los Angeles County had
underbuilt 30,000 new housing units annually from 1980-2010 compared to estimated demand
(Taylor, 2015) and the city of Los Angeles (accounting for nearly 40% of county population) had
permitted an average of only 7,500 units annually since 2000 (U.S. HUD, n.d.), producing a
housing affordability crisis in both city and county. As a result, in 2013, Los Angeles County
was one of the most unaffordable counties in the nation, with its residents an average of 30
percent of their income on housing, a higher proportion than households in other California
metropolitan areas and about 7 percentage points higher than major metropolitan areas in other
states (Taylor, 2015). Home prices and incomes have diverged widely: a median income
household in 2012 Los Angeles was able to afford $190,000 home, yet home prices averaged
$400,000 (LADCP 2013). Renters have not been spared: rents had increased by more than 20
percent in real terms between 1990-2010, despite slightly decreasing incomes (Collinson 2011).

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
29

Calculating Neighborhood Mobility Rates near L.A. Metro Rail Stations
To assess the relationship between neighborhood mobility and rail transit, we assign households
to neighborhoods based on the proximity to L.A. Metro rail stations. We define a rail-station area
neighborhood as a circle with a 0.5 mile radius from the rail station in any direction. The half-
mile distance neighborhood is frequently used in transportation planning as a catchment area for
rail transit and is cited in many transit-oriented development policies (Guerra, Cervero, &
Tischler, 2011). We geocode rail station locations for all 80 stations opened in or before 2013
using shapefiles provided by the Southern California Association of Governments (SCAG) and
draw half-mile boundary neighborhoods around them (SCAG, 2015).
We consider households whose tax filing location falls within the half-mile circle as part of that
station’s neighborhood and include them in mobility rate calculations. Systemwide, an average
of 337,000 households fall into the 80 station-area neighborhood boundaries annually for whom
data are available for at least two consecutive years (to avoid data dropouts).. Adjusted baseline
populations and out-mobility rates for each station area are displayed in Appendix 1-B. The
largest station neighborhood in 2012 was Wilshire/Western on the Purple Line with 18,523
households and the smallest was Douglas on the Green line with 248 households.
5
It takes fewer
household moves to influence the neighborhood move rates for neighborhoods with lower
populations such as Douglas. To control for this, we weight our analyses by the number of
households in the neighborhood in each year (see discussion on Weighting below).
Because of FTB data confidentiality restrictions, we are forced to intermix 9-digit and 5-digit zip
code geocoding as we assign households to rail station neighborhoods based on distance to
station. Since 9-digit zip codes represent at most one street block, we are confident that we can
accurately assign observations with 9-digit zip codes to a rail-station neighborhood. However,
many observations have 5-digit zip codes in some years: we assign these to rail-station
neighborhoods as well. Of all households assigned to rail-station neighborhoods who are in the
data for at least two consecutive years, about 34% have 9-digit zip codes in both years t and t+1,
an additional 7% have a 9-digit zip code in year t only (but a 5-digit zip code in year t+1), and
11% have a 9-digit zip code in year t+1 only (but a 5-digit zip code in year t) (see Table 2). The
remaining 48% of observations have 5-digit zip codes in both year t and t+1. We follow the
same intermixed geocode procedure to determine the number of households who moved from
each rail-station neighborhood in each year. Decomposing the out-mover proportion, we
calculate that 13% of out-movers had 9-digit zip codes in both years t and t+1, an additional 22%
have a 9-digit zip code in year t only (but a 5-digit zip code in year t+1), and 38% have a 9-digit
zip code in year t+1 only (but a 5-digit zip code in year t) (see Table 4). The remaining 26% of
observations have 5-digit zip codes in both year t and t+1.
We believe that our strategy is the best course of action. Dropping all households who do not
have a 9-digit zip code in both years t and t+1 would significantly undercount both households

5
We exclude 6 rail station neighborhoods where nearly 100% of the neighborhood is devoted to industrial and/or
commercial uses and which have very few households living in them, including El Segundo, Mariposa, and
Redondo Beach on the Green Line; Irwindale on the Gold Line Foothills Extension Line; and Artesia and Del Amo
on the Blue Line.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
30

who are eligible to move and those who do move (Table 4). Nevertheless, this is a potential
limitation of our neighborhood assignment strategy. As a result, we weight our analyses also by
the proportion of households in the neighborhood who have 9-digit zip codes in each year (see
discussion on Weighting below).
We use the mobility rates and adjusted baseline populations for each neighborhood in each year
to analyze the connection between residential mobility and rail transit. To understand effects
across the income distribution, we also compute adjusted baseline populations and out-mobility
rates for four income groups within each neighborhood, with group cutoffs established relative to
the area median income (AMI) for the Los Angeles – Long Beach Metropolitan Statistical Area
(see Appendix 1-C for specific values by year). These income groups are Lowest Income (less
than 30% of AMI, or less than $15,000 in 2013), Low Income (30-50% of AMI, or $15,000-
$25,000 in 2013), Lower Middle Income (50-80% of AMI), or $25,000-$40,000 in 2013, and
Middle and Upper Income (above 80% of AMI, or more than $40,000 in 2013). We use FTB
data on annual household income to assign each household into these income categories.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
31

Table 4. Rail Station Neighborhood Adjusted Baseline Population and Count of Out-Movers by Year: Breakout of Geocoding Type
Proportion of Baseline Population with… Proportion of Out-Movers with…
Year
Adjusted
Baseline
Neighborhood
Population
Adjusted
Population as
Proportion of
Unadjusted
9-digit zip
codes in
years t
and t+1
9-digit zip
codes in year
t, but 5-digit
in t+1
9-digit zip
codes in
year t+1, but
5-digit in t
5-digit
zip codes
in both
years
Number
of Out-
movers
9-digit
zip codes
in years t
and t+1
9-digit zip
codes in
year t, but
5-digit in t+1
9-digit zip
codes in
year t+1, but
5-digit in t
5-digit zip
codes in
both
years
1993 308,094 77% 11% 3% 21% 65% 95,508 5% 8% 56% 31%
1994 293,306 79% 23% 6% 15% 57% 80,372 10% 16% 46% 28%
1995 279,202 77% 27% 7% 12% 54% 70,090 11% 21% 40% 27%
1996 287,262 80% 29% 7% 12% 51% 72,555 12% 20% 40% 28%
1997 293,527 81% 31% 7% 12% 50% 72,042 12% 22% 40% 26%
1998 306,678 83% 33% 7% 11% 49% 72,948 12% 21% 40% 27%
1999 316,187 83% 34% 7% 11% 48% 73,338 13% 22% 39% 26%
2000 323,025 83% 35% 7% 11% 47% 72,190 13% 23% 39% 26%
2001 331,802 84% 36% 7% 10% 48% 74,693 13% 23% 36% 28%
2002 323,845 82% 36% 7% 10% 47% 73,281 13% 25% 35% 27%
2003 328,210 85% 36% 7% 10% 47% 75,160 15% 22% 36% 27%
2004 347,994 85% 36% 7% 10% 47% 80,571 13% 24% 35% 28%
2005 356,652 85% 37% 7% 10% 46% 81,579 14% 23% 37% 27%
2006 367,198 86% 38% 7% 10% 45% 86,841 19% 22% 34% 24%
2007 377,213 84% 39% 8% 9% 44% 83,390 14% 26% 34% 25%
2008 378,954 85% 38% 7% 10% 44% 91,217 17% 24% 34% 25%
2009 380,668 86% 39% 8% 10% 44% 88,088 15% 26% 34% 25%
2010 386,047 85% 39% 7% 10% 43% 88,540 15% 24% 37% 24%
2011 383,683 84% 40% 8% 11% 42% 92,616 17% 25% 35% 23%
2012 378,569 83% 40% 8% 9% 43% 84,771 15% 26% 34% 24%
Average 337,406 83% 34% 7% 11% 48% 80,490 13% 22% 38% 26%
Source: Author calculation of FTB and Geolytics data
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
32

Selecting Control Neighborhoods
We use the neighborhood-level mobility rates assigned above to assess the relationship between
residential mobility and rail transit, before and after stations open. However, what if the new
transit system affects other more distant parts of the city in the same way? Or, what if mobility
rates fluctuate generally, unrelated to the new transit system? Such questions may confound the
relationship between residential mobility and rail transit. To better isolate the relationship of rail
transit on rail-proximate neighborhoods, we test the relationship against a set of control
neighborhoods. These control neighborhoods are similar to rail-station (treatment)
neighborhoods demographically, but are not located walking distance from a rail station.
We select a set of potential control neighborhoods by aggregating controls households that are
not walking distance a rail station, but in a similar part of the County, usually within 0.5 – 3
miles of the rail station. Following Schuetz, Giuliano, and Shin (2018), our control
neighborhoods are centered around a busy intersection, since rail stations are often built near
busy intersections. To choose between all potential control intersections, we examine the
demographic characteristics of the neighborhoods within a half-mile radius of the intersection,
including income, race, ethnicity, housing tenure and educational status using the U.S. Census
and American Community Survey. We then pick the intersection with demographics most
similar to the treatment neighborhood. To avoid overlap issues, we ensure that the central point
of the control neighborhood is at least 1 miles away from all other non-paired stations, other
control neighborhoods, and other L.A. Metro rail stations not in the analysis.
6

We compute the adjusted population baseline, out-mobility rate for each control neighborhood,
for all income categories, in each year using the same intermixed geocoding strategy as for the
treatment neighborhoods described above. See Appendix 1-B for station-level details.

Descriptive Analysis of Los Angeles Metro Rail Corridors
Before we address the effects of rail station development on neighborhood mobility rates and
estimate potential displacement, we document the baseline mobility rates for the County and for
each rail corridor of interest. We also document the changing size and income composition of
rail station corridors versus control neighborhoods and summarize growth by income. These
descriptive analyses will ground our statistical models in the evolution of the neighborhoods in
the study area from 1993-2013.
Baseline Mobility Rates
On average, 21% of Los Angeles County households move every year. This does not vary much
by income: 23% of households with incomes below 80% of AMI move annually and 18% of
households with incomes above 80% of AMI. Moreover, decreasing trends by income over time

6
In a few cases, finding a control neighborhood that is of the correct distance parameters creates a challenges
especially in the transit-dense areas of Downtown L.A. and Central L.A. In these cases, I lower the centroid to
centroid distance requirement of 1 mile to 0.9 miles. For instances where this occurs, see Appendix 1-B, Table 13.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
33

follow the county average (Figure 6) and the national average (Figure 4). Thus, we are studying
mobility in the context of gently falling mobility nationwide and countywide. Any hypothesis
involving displacement would run against this long-term trend.

Figure 6. Annual Average Mobility Rate by Income Category for Los Angeles County
Source: Author Calculations of FTB data
L.A. Metro rail stations are located in relatively denser neighborhoods than the county average
and these neighborhoods have a higher proportion of low-income, renter, foreign-born, and
minority households (Boarnet et al., 2015). We thus expect higher baseline mobility rates along
L.A. Metro rail corridors than for the County overall, even without accounting for additional
effects from the rail station. Figure 7 shows the annual average out-mobility rate by transit
corridor versus the county average. Indeed, several of the lines have baseline mobility rates 5 or
more percentage points higher than the county (Red/Purple and Blue lines). However, the Expo
Phase I, Gold and Green Lines have baseline rates very similar to the county average. These
corridor differences reflect the diversity and size of the L.A. Metro rail system and the county in
general.

Figure 7. Annual Average Baseline Out-Movement Rates by L.A. Metro Rail Line
0%
5%
10%
15%
20%
25%
30%
35%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
All Incomes Above 80% AMI 50-80% AMI 30-50% AMI Below 30% AMI
10%
15%
20%
25%
30%
35%
40%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Blue Expo Gold Green Red/Purple Los Angeles County
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
34

Source: Author Calculations on FTB data

Neighborhood Composition
The adjusted baseline household population of Los Angeles County has grown by 40% during
the study time period from 3.2 million to 4.5 million and averages 4 million households per year
(Figure 8). About 27% of households have incomes below 30% of AMI, 35% of households
have incomes between 30 and 80% of AMI, and 38% of households have incomes above 80% of
AMI. Each income group has grown in absolute terms as the number of households has grown.
The absolute population of Middle and Upper Income households (AMI >80%) has grown by
over 50%, or 2.2% per year, slightly faster than the other groups.

Figure 8. Adjusted Baseline Household Population of Los Angeles County by Income Category
Note: Adjusted Baseline Household Population are all the households who appear in the data for at least
two consecutive years
Source: Author Calculations on FTB data
At the transit corridor level (Figures 9A-J), the Blue and Red/Purple Lines neighborhoods have
the most households, averaging about 90,000 and 80,000 households respectively per year. The
Gold Line is next in size with about 65,000 households per year. The Expo Phase I and Green
Lines are the smallest with about 27,000 and 21,000 households annually. The transit corridors
also differ by their income composition. On average, 35% of Gold, Blue, Green, and Expo Phase
I Line households have incomes below 30% of AMI, but the figure is 41% for Red/Purple Line
households. The Gold Line has the most households with incomes above 80% of AMI while the
Red/Purple Line has the fewest.

28%
30% 29% 30% 29% 29% 28% 27% 27% 27%
24%
26% 25% 25% 24% 25%
28% 28% 28% 27%
18%
18% 18%
18% 18% 18% 18% 18%
18% 18%
17%
17% 17% 17% 17%
17%
17% 18% 17%
17%
18%
18% 18%
18% 18%
18%
18% 18%
18% 18%
18%
18% 18%
18% 18%
18%
17% 17% 17%
17%
36%
34% 35%
35%
35%
35%
36%
37%
36%
37% 41%
39%
40%
41% 42% 40% 38% 38% 38%
39%
0
1,000,000
2,000,000
3,000,000
4,000,000
5,000,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Los Angeles County
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
35

Figure 9. A-J: Adjusted Baseline Household Population by Income Strata for Treatment and Control
Neighborhoods

Note: Adjusted Baseline Household Population are all the households who appear in the data for at least
two consecutive years
Source: Author Calculations on FTB data

45%
48%
46%
48% 46% 45% 44% 42%
43% 42%
37%
38% 37% 36% 33% 34%
39% 39% 39% 38%
22%
23%
23%
23% 24%
24%
25% 25%
25%
25%
25%
24% 24% 24%
23%
23%
23%
23% 23%
23%
16%
15%
15%
15%
16%
16%
17% 17%
17%
18%
19%
19% 19%
19% 20%
20%
18%
18% 18%
18%
16% 14%
15%
14%
14%
14%
15%
16%
15%
16%
19%
18%
20%
21%
23% 23%
20%
20% 21%
21%
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
80,000
90,000
100,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Red/Purple Line
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
37% 40% 39% 40% 39% 39% 38% 36% 37% 37%
32%
34% 32% 32% 30% 32%
35% 35% 35% 35%
21%
21%
21%
22% 22%
22% 23% 23%
23%
23%
22%
22%
22% 21% 21%
21%
21% 22% 21%
21%
17%
17%
17%
17% 17%
17%
18% 18%
18%
18%
19%
19%
19% 19% 20% 19%
18% 18% 17%
18%
25% 22%
23%
22%
22%
22%
22%
23%
22%
22%
26%
25% 27%
27% 29% 28%
25%
25% 26%
26%
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
80,000
90,000
100,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Red/Purple Line Control Neighborhoods
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
36

36% 39%
37% 39% 38% 37% 37% 35% 36% 35%
31%
32% 31% 31% 29% 30%
34% 34% 34% 34%
22% 22%
22%
22%
22%
22% 23% 23%
23% 23%
23%
23% 22% 22% 22%
22%
22% 22% 22%
21%
18% 18%
18%
18%
18%
19% 18% 19%
19%
19%
20%
20% 19%
19% 19%
19%
18% 18% 17%
18%
23%
21%
23%
22%
23%
22% 22% 23%
22%
23% 27%
25% 27%
28%
30% 29% 26% 26% 27%
27%
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
80,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Gold Line
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
35% 38%
37% 38% 37% 36% 36% 35% 35% 34%
29%
31% 30% 30% 29% 30%
34% 34% 34% 34%
23% 23%
22%
22% 23% 24% 24% 23% 24%
24%
23%
23% 23% 22%
22% 22%
22% 22% 22% 22%
20% 18%
19%
18%
18%
19%
18% 19%
19%
20%
21%
20% 20% 20%
20% 20%
18% 18% 18%
18%
22%
20%
22%
22%
22%
21%
22%
23%
22%
23%
27%
26% 27% 28% 29% 28%
26% 26% 26%
26%
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
80,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Gold Line Control Neighborhoods
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
37% 39%
38%
39%
41% 36% 36% 36% 36% 36%
30%
32% 31% 30% 29% 29%
34% 34% 34% 34%
23% 24%
23%
24%
24% 25%
25%
25%
25%
25%
25%
25% 25% 24% 24%
24%
23%
24% 23% 23%
19% 18%
19%
19%
19% 19%
20%
20%
20%
21%
22%
21% 21%
21% 21%
21%
19%
18%
18%
18%
22%
19%
20%
18%
18%
18%
18%
19%
18%
18% 23%
22%
23%
24%
27%
26%
24%
23%
24%
25%
0
20,000
40,000
60,000
80,000
100,000
120,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Blue Line
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
37

35% 38% 37% 38% 38% 37% 37% 36% 36% 35%
30%
33% 32% 32% 30% 31%
36% 37% 37% 37%
23%
23%
23%
23% 24%
24%
25% 25%
26%
25%
25%
25% 25%
25%
25% 26%
25%
25% 25%
25%
19% 18%
19%
18%
18%
19%
19% 19%
19% 20%
21%
20% 20%
20%
20% 20%
19%
18% 18%
18% 23% 20%
22%
20%
20%
19%
19%
20%
19% 20% 24%
22% 23%
23%
24% 23% 20%
20% 20%
21%
0
20,000
40,000
60,000
80,000
100,000
120,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Blue Line Control Neighborhoods
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
35% 38%
36%
38% 37% 34%
36% 33% 34% 33%
28%
31% 31% 31% 30% 30%
35% 35% 36% 36%
26% 25%
24%
24% 25% 26%
26% 26%
27%
26%
26%
25% 25% 25% 25% 26%
26% 26% 25% 25%
21% 20%
20%
20%
20% 21%
20% 21%
20%
21%
23%
22% 23%
22% 22%
22%
20% 20% 20%
19%
18% 17%
19%
18%
19%
18%
18%
19%
18%
19% 23%
21%
22%
22%
22% 21%
19% 19% 19%
19%
0
5,000
10,000
15,000
20,000
25,000
30,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Green Line
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
26% 28%
27% 29% 29% 29% 29% 28% 28% 28%
24%
26% 26% 26% 25% 25%
29% 30% 30% 31%
20%
21%
21%
22% 22%
22% 23% 23%
23%
23%
21%
22% 22% 22% 21%
22%
22%
23% 23% 22%
22%
22%
22%
22% 22%
22%
22% 23%
23%
23%
23%
23% 23%
23%
23%
23%
22%
21% 21%
21%
32%
29%
30%
28%
28%
27%
26%
27%
26%
27%
31%
29%
30%
30% 31% 30% 27%
26% 26%
26%
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Green Line Control Neighborhoods
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
38

The household population of neighborhoods surrounding each L.A. Metro rail corridor has
increased over the study time period, but in different patterns. We summarize these growth
patterns in Table 5.A-B. The Red/Purple Line corridor has grown the most over this time period,
increasing its household population by 46% from 1993 to 2013: it is growing faster than the
County. This was followed closely by the Blue Line and the Green Line, both of which have
grown by over 30% since 1993, which is close to County growth rates. In contrast, the Gold and
Expo Line neighborhoods have only grown by 12% and 15% respectively, a rate nearly twice as
slow as the County average.
The household population has also increased in Blue and Red/Purple Line control
neighborhoods, but only by 30%, about 10-15% less than treatment neighborhoods. Gold Line
control neighborhoods barely grew, with growth rates even lower than in Gold Line treatment
neighborhoods. Expo Line control neighborhood growth was nearly double that of treatment
35% 38% 37% 38% 39% 36% 37% 37% 37% 36%
32%
34% 33% 34% 32% 32%
36% 36% 36% 36%
23% 23%
23%
23%
23% 24%
24% 24%
24%
24%
23%
24% 23%
23%
22% 24%
23% 24%
23%
23%
20% 19%
20%
19%
20% 20%
20% 20%
20%
20%
21%
20% 21%
21%
21% 20%
19% 19%
19%
19%
22%
19%
20%
19%
20% 19%
19%
20%
19%
20% 24%
22% 23%
23% 25% 24% 21% 21%
22%
22%
0
5,000
10,000
15,000
20,000
25,000
30,000
35,000
40,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Expo Phase I Line
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
37% 40% 39% 40% 39% 38% 38% 38% 38%
37%
32%
35% 35% 34% 33% 33%
38% 38% 38% 38%
25% 24%
23%
23% 24%
24%
25% 24%
25%
25%
25%
25% 25% 24%
24% 25%
24%
24% 24%
24%
21% 20%
20%
19%
19%
19%
19%
19%
18%
19%
21%
20% 20%
20%
20% 20%
18%
18% 18%
18% 17%
17%
19%
18%
18%
18%
18%
19%
18%
18% 22%
20% 21%
21%
23% 22% 20%
20% 20%
20%
0
5,000
10,000
15,000
20,000
25,000
30,000
35,000
40,000
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Expo Phase I Line Control Neighborhoods
<30% AMI 30-50% AMI 50-80% AMI >80% AMI
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
39

neighborhoods. Green Line control neighborhoods actually lost population over this time period,
despite growth in the treatment neighborhoods.
Each income group has grown in each transit corridor, accompanying the topline growth in the
number of households. For control neighborhoods, though, several income strata on the Gold and
Green Lines saw population decline. While growth has been uneven by income, the number of
Middle and Upper Income households has grown faster in rail-adjacent areas in both high-
growth corridors (Red/Purple, Blue, Green) and in low-growth corridors (Gold). Red/Purple,
Expo, and Gold Line controls followed this phenomenon, but not Green or Blue Line controls.
Despite the Gold Line’s low growth overall, the annual growth rate of Middle and Upper Income
households has been more than triple that of other income categories with a similar but even
more pronounced patter for Gold Line controls. The Red/Purple Line has the highest proportion
of Lowest Income households among the transit lines, but its number of Middle and Upper
Income households nearly doubled over the study time period. Middle and Upper Income growth
in Red/Purple Line controls was high overall, but one-third that of Red/Purple Line treatment
areas. In the Blue and Green Lines, the Middle and Upper Income household category grew by
around 50%. Only in the low-growth Expo Line neighborhoods has the growth been distributed
relatively evenly by income group.
The compositional change toward a higher income household population indicates possible
gentrification for Blue, Green, Gold, and Red/Purple Line neighborhoods. For Red/Purple, Blue,
and Green Line neighborhoods growth in the higher-income stratum outpaced that of their
control neighborhoods by more than three times. This suggests that rail-proximate
neighborhoods have gentrified over this time period, while their control neighborhood pairs have
not. Gentrification appears to be a very localized process in these areas, because control
neighborhood centers are on average only 1.5 miles away from stations.
It is possible that these longer-term neighborhood population changes are accompanied by
parallel increases in these neighborhoods’ housing stock, but this is unlikely given the housing
permitting and construction patterns in Los Angeles County during this time period (Taylor,
2015; U.S. HUD, n.d.). In fact, these patterns correlate well with the Urban Displacement
Project’s findings that at least 15% of census tracts in Los Angeles County at-risk of
gentrification had gentrified between 1990-2013. These patterns highlight the potential for a link
between rail, gentrification, and displacement in our study area.
Table 5. Household Population Growth Rates by Transit Line by Income Strata, Treatment and Control
Neighborhoods
Treatment
Neighborhoods
Income
Category
Los Angeles
County
Red/Purple
Line
Gold
Line
Blue
Line
Green
Line
Expo
Line
Absolute
Growth Rate
<30% AMI 35% 23% 4% 29% 34% 17%
30-50% AMI 36% 47% 6% 38% 32% 19%
50-80% AMI 30% 61% 9% 32% 22% 9%
>80% AMI 52% 92% 31% 58% 47% 14%
Total 40% 46% 12% 38% 33% 15%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
40

Compound
Annual Growth
Rate (CAGR)
<30% AMI 1.6% 1.1% 0.2% 1.3% 1.6% 0.8%
30-50% AMI 1.6% 2.1% 0.3% 1.7% 1.5% 0.9%
50-80% AMI 1.4% 2.6% 0.4% 1.5% 1.1% 0.5%
>80% AMI 2.2% 3.5% 1.4% 2.5% 2.0% 0.7%
Total 1.8% 2.0% 0.6% 1.7% 1.5% 0.7%

Control
Neighborhoods
Income
Category
Los Angeles
County
Red/Purple
Line
Gold
Line
Blue
Line
Green
Line
Expo
Line
Absolute
Growth Rate
<30% AMI 35% 20% 2% 33% 11% 31%
30-50% AMI 36% 25% -1% 39% 3% 23%
50-80% AMI 30% 28% -2% 21% -12% 7%
>80% AMI 52% 35% 31% 13% -25% 52%
Total 40% 26% 7% 28% -7% 28%
Compound
Annual Growth
Rate (CAGR)
<30% AMI 1.6% 0.9% 0.1% 1.5% 0.5% 1.4%
30-50% AMI 1.6% 1.2% 0.0% 1.8% 0.2% 1.1%
50-80% AMI 1.4% 1.3% -0.1% 1.0% -0.7% 0.4%
>80% AMI 2.2% 1.6% 1.4% 0.6% -1.5% 2.2%
Total 1.8% 1.2% 0.4% 1.3% -0.4% 1.3%
Note: Absolute Growth Rate = (Adjusted Household Baseline in 2012/ Adjusted Household Baseline in
1993)-1. Compound Annual Growth Rate (CAGR) = (Adjusted Household Baseline in 2012/ Adjusted
Household Baseline in 1993)^(1/(2012-1993))-1.
Source: Author Calculations on FTB data

Methods and Analysis
We use the FTB data to answer our main research question: Does the introduction of a rail
station to a neighborhood change the neighborhood mobility rate, and if so, does this indicate
displacement? We also explore several related questions: a) does the rail station effect occur at
the station’s announcement, opening, or five years later? b) are lower-income households more
affected? c) how does transit technology, station location within a neighborhood or transit
system, and system size moderate the rail station effect? To answer these questions, we test the
hypothesis that a neighborhood with rail transit will experience higher mobility rates than a
counterfactual control neighborhood.

Modeling Approach
To measure the effects of a new transit system on neighborhood mobility rates and potential
displacement, we set up statistical models to compare between treatment and control
neighborhoods and to control for spatial and temporal variation. We measure the effects of rail
opening, but also of rail announcement and five years after completion to assess effect timing.
We categorize differences by income group and by transit line. We also take into account train
technology, station location within the system, within a neighborhood and within the
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
41

metropolitan area, changes in system size, and neighborhood type. Having estimated the effects
on mobility rates, we compare them to average outflow rates in each neighborhood. This
difference in move rates between rail and control neighborhoods after rail opens reflects the
displacement impact of rail.
Two statistical models – the difference-in-difference (DID) model and the panel fixed effects
(FE) model – help us identify a rail effect on mobility rates in treatment neighborhoods. The DID
model is an Ordinary Least Squares (OLS) regression model that measures a causal Average
Treatment Effect on the Treated, by comparing a treatment versus a control group before and
after the treatment as the identification strategy (Cameron and Trivedi, 2005). In our case, the
main “treatment” is the opening of a rail station, the “treatment group” is the 0.5 mile
neighborhood surrounding the station, and the control group is the similar neighborhood that did
not receive a rail station.
We implement the DID model using the regression in Equation 3. Y represents the mobility rate
for neighborhood i in year t. Treatment is a binary variable which equals 1 for treatment
neighborhoods and 0 for control neighborhoods. Post is a binary variable which equals 1 if the
treatment has already occurred (i.e., rail station has opened) and 0 if it has not yet occurred.
Treatment*Post is the interaction between Treatment and Post and represents the variable of
interest: the treatment effect on the treated group. The constant term α represents the baseline
mobility rate. Since each treatment neighborhood is paired with an assigned control
neighborhood based on similar characteristics, we cluster standard errors by the neighborhood
pair, which are also robust to heteroskedasticity.
Equation 3: Difference − in − Difference (DID) Model
𝑌 𝑖𝑡
= 𝛼 + 𝛽 ∗ 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝑖 + 𝛾 ∗ 𝑃𝑜𝑠𝑡 𝑡 + 𝛿 ∗ 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑃𝑜𝑠𝑡 𝑖𝑡
+ ∈
𝑝
To estimate a causal effect using the DID model, four assumptions must be met:
1) the treatment is unrelated to outcome at baseline (allocation of treatment was not
determined by outcome),
2) the composition of treatment and control groups is stable for repeated cross-sectional
design,
3) there are no spillover effects, and
4) treatment and control groups have parallel trends in outcomes (Cameron and Trivedi,
2005).
Our model meets the first assumption, because there is no evidence that rail stations siting takes
population mobility into account. On the contrary, the availability of rail right-of-way and
political feasibility seem to inform location choice for L.A. Metro stations and rail corridors
(Schuetz, Giuliano, and Shin, 2017; Taylor, Kim, and Gahbauer 2009). Our model satisfies the
second assumption since the location of our treatment and control neighborhoods does not vary
over the study time period. There are cases where households may belong to multiple treatment
neighborhoods because stations are within less than one mile of each other. We run a “no-
overlap” specification with the overlapping treatment stations removed and find no difference
with or without overlap stations on neighborhood mobility rates. To satisfy the third assumption,
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
42

we design our control groups to be far enough from the treatment (at least one mile distance
between treatment and control neighborhood centroid) in order to avoid spillover effects. The
parallel trends assumption visually appears to hold for rail opening for the whole sample as well
as for each transit corridor individually (see Appendix 1-E for details). While the assumptions
for a causal DID seem satisfied, we also use a second and more stringent model that does not
rely on DID assumptions as a on check DID estimates.
The panel fixed effects (FE) model takes advantage of longitudinal data and controls for the
effects of unobserved time-invariant variables with time-invariant effects (Allison, 2009). We
employ a fixed effects model with a binary treatment, with a dummy variable for each year and
for each neighborhood. In our case, the binary treatment variable equals 1 if the rail station is
open in that year in that neighborhood and 0 otherwise. A pooled OLS regression consistently
estimates the binary treatment effect in the presence of the a full set of year and neighborhood
dummy variables (Cameron and Trivedi, 2005). We drop one dummy variable from each set to
avoid collinearity and to include a constant term in the regression. Since the errors in the FE
model are potentially serially correlated and/or heteroskedastic, we use cluster-robust standard
errors at the neighborhood pair level, which provides a solution to both issues (Cameron and
Trivedi, 2005).
In our application, the FE model nets out any unobserved effects which may affect all
neighborhoods in a particular year or may affect a particular neighborhood across the years. Our
identifying assumption is that once these dummy variables control for any unobservable effects
common to a specific year or neighborhood, what remains is δ, a consistent and unbiased
estimator of the rail station effect on Y it, the neighborhood mobility rate for neighborhood i in
year t for rail-station neighborhoods after the rail stations open (see Equation 4). The constant
term α represents the baseline mobility rate.
Equation 4: Panel Fixed Effect (FE) Model
𝑌 𝑖𝑡
= 𝛼 + 𝛿 ∗ 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑃𝑜𝑠𝑡 𝑖𝑡
+ ∑ 𝜑 𝑗 ∗ 𝑌𝑒𝑎𝑟 𝑡 𝑡 𝑡 =1
+ ∑ 𝛾 𝑗 ∗ 𝑁𝑒𝑖𝑔 ℎ𝑏𝑜𝑟 ℎ𝑜𝑜𝑑
𝑖 𝑖 𝑖 =1
+∈
𝑖𝑡

The FE assumes that that time-varying explanatory variables are not collinear, have non-zero
within variance, and do not have too many extreme values (Cameron and Trivedi, 2005). Our
model satisfies these assumptions.
We obtained estimates of the average rail effect on neighborhoods treated with rail stations. This
gives us average changes in neighborhood mobility rates, but not displacement. To calculate a
measure of displacement, we divide these point estimates by the baseline mobility rate derived
from the constant term in the model, adjusting the baseline mobility rate for group differences
where appropriate. This quotient reflects the magnitude or the rail station impact. This proportion
of the mobility rate due to rail stations is over and above the baseline rate. We infer that in cases
where effects are statistically significant, this represents the degree of rail-related displacement
in a neighborhood.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
43

Modeling Extensions
The DID and FE models provide a basic answer to whether rail station openings affect
neighborhood mobility. However, do other variables confound this effect? Due to the relative
lack of data either temporally or spatially corresponding to our zip-code-derived neighborhood or
our FTB data-derived tax filers, we are somewhat constrained in our ability to test for potential
confounding variables. Nevertheless, we provide several extensions to the basic DID and FE
models to test hypotheses about whether estimates differ by income group and transit line, by rail
transit technology, by station location within the system, neighborhood, and metropolitan area,
by changes in system size, and by neighborhood type. We also account for lag and anticipation
effects.
We hypothesize that in addition to rail station opening, rail station announcement will also
increase neighborhood mobility for treatment neighborhoods due to anticipation of the increased
access from the future rail station. Prior evidence exists that light-rail plans in Washington
County, Oregon and in Atlanta increase property values in station area neighborhoods even
before stations open (Knaap et al., 2001; Immergluck, 2009) Since changes to the built
environment may occur over more than the initial year of a station’s opening, we also
hypothesize the existence of a lag effect on neighborhood mobility. The average time between
the announcement and opening of an L.A. Metro rail station was 4.7 years in our sample.
Rounding this to five years, we test for an anticipation effect five years before opening by
changing the Post variable to represent post-announcement in equations 3 and 4. To test for lag
effects, we change the Post variable in equations 3 and 4 to an analogous five years after station
opening.
The system-level average rail station effect may obscure differences in effect of an individual
transit corridor. Since transit corridors may differ by topography, ethnoracial composition of its
inhabitants, economic and industrial profile, or numerous other variables which we can not
measure but may affect mobility rates, we estimate both models for each of five rail corridors
when possible: Red/Purple, Gold, Green, Expo Phase I, and Blue.
We hypothesize that lower-income households are more likely to be displaced than higher-
income households. We test whether the neighborhood mobility rate for a particular income
category changes more or less than other income categories due to rail station effects. For each
neighborhood, we calculate a mobility rate for four income categories introduced above [Lowest
income (filers with incomes below 30 percent of AMI), Low income (filers with incomes
between 30 and 50 percent AMI) Lower-Middle income (filers with incomes between 50 and 80
percent of AMI) and Middle and Upper income (filers with income above 80% of AMI)]. We
estimate a DID and an FE model with income category dummies w interacted with key variables
of interest (see Equations 5 and 6 below). The variable of interest in these specifications is the
interaction of Treatment, Post, and Income. We estimate specifications with income categories
for the whole system and for each transit line. We calculate the baseline mobility rates for each
income group by adding the constant term α to the non-interacted income terms.
Equation 5: Difference − in − Difference (DID) Model with income category dummies
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
44

𝑌 𝑖𝑡
= 𝛼 + 𝜔 ∗ 𝐼𝑛𝑐𝑜𝑚𝑒 𝑤 + 𝛽 ∗ 𝐼𝑛𝑐𝑜𝑚𝑒 𝑤 ∗ 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝑖 + 𝛾 ∗ 𝐼𝑛𝑐𝑜𝑚𝑒 𝑤 ∗ 𝑃𝑜𝑠𝑡 𝑡 + 𝛿 ∗
𝐼𝑛𝑐𝑜𝑚𝑒 𝑤 ∗ 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝑖 ∗ 𝑃𝑜𝑠𝑡 𝑡 + ∈
𝑝
Equation 6: Panel Fixed Effect (FE) Model with income category dummies
𝑌 𝑖𝑡
= 𝛼 + 𝜔 ∗ 𝐼𝑛𝑐𝑜𝑚𝑒 𝑤 + 𝛿 ∗ 𝐼𝑛𝑐𝑜𝑚𝑒 𝑤 ∗ 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝑖 ∗ 𝑃𝑜𝑠𝑡 𝑡 + ∑ 𝜑 𝑗 ∗ 𝑌𝑒𝑎𝑟 𝑡 𝑡 𝑡 =1
+
∑ 𝛾 𝑗 ∗ 𝑁𝑒𝑖𝑔 ℎ𝑏𝑜𝑟 ℎ𝑜𝑜 𝑑 𝑖 𝑖 𝑖 =1
+∈
𝑖𝑡

The L.A. Metro rail system uses both heavy and light rail technology. Heavy rail is a subway
train which travels underground and is used for the Red/Purple lines. This increases speed,
reduces accidents and at-grade stopping, and can often accommodate more frequent headways
than light-rail trains which often run at-grade. Due to their underground nature, subways are
more expensive and take longer to build; yet, they may be less intrusive on the visible built
environment, not requiring at-grade or elevated right-of-way. Previous research has found that
property value impacts are greater from subways than from light-rail systems (Zuk et al., 2017).
We hypothesize that heavy rail station effects may differ from light-rail station effects. We create
a dummy variable for each neighborhood subway which equals 1 if the nearby station is a
subway and 0 if it is light rail.
7
We then interact the subway dummy with the other rail-related
variables.
Certain stations in the L.A. Metro system are located in the freeway median. These stations were
built there to take advantage of the existing freeway right-of-way (Gold Line) or as part of
freeway construction (Green Line). Siting within a median may serve to substitute for highway
traffic. However, such stations may also face difficulties with access and egress if patrons have
to cross busy streets at potentially unsafe crossings. We hypothesize that highway median
stations may have lower effect on neighborhood mobility than other stations which are situated
in more of a neighborhood setting. To test this, we create a dummy variable highwaymedian
which equals 1 if the nearby station is in the highway median and 0 otherwise. We then interact
the highwaymedian dummy with the other rail-related variables.
Much of the L.A. Metro rail system is focused on connecting Downtown Los Angeles, the
historical central business district, with other portions of Los Angeles County. Additionally, the
Blue Line connects Downtown Los Angeles with Downtown Long Beach, another historical
high-density concentration. The high density and overlapping nature of stations serving these
two downtowns may confound estimation of rail effects on neighborhood mobility. We run
separate models with and without these stations to test for this confounding effect.
In addition, certain L.A. Metro stations were built very close to each other, with less than one
mile distance from station to station. This makes estimation of the effect of a particular station
difficult to separate, since our neighborhoods are defined as one half-mile circles around a
station. We run separate models with and without these stations to test for this confounding
effect.

7
In cases of transfer stations or when two trains run along the same route, we assign the value of whichever
technology was built first at that station.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
45

The location of a station within the system may affect its ridership and the surrounding
neighborhood’s real or perceived popularity, which could alter neighborhood mobility rates. For
example, transfer stations may receive more retail development than singleton stations and
stations with fewer stops to Downtown Los Angeles may be receive more residential
development catering to downtown employees. We run additional specifications testing the
effect the number of stops to Downtown Los Angeles or to the nearest transfer station. Also, as
the transit system grows, the number of other stations riders can access from each new stations
increases. Hence, later stations may generate more nearby development than earlier stations,
holding other factors constant. We run a specification taking account of system size at the time
of station opening.

Weighting
The neighborhoods making up our sample have different populations (see Appendix 1-B), yet we
are interested in understanding the rail station effect on for an “average” station in the system or
in a particular transit corridor. Particularly large or small stations may unduly influence the
estimates and/or will not behave like a representative station. In addition, our sample restrictions
and geocoding solutions introduce additional differences into neighborhood sample sizes. We
use weights in our regression models to attempt to control for these issues.
We consider four different weights to address the above issues. First, we try weighting by the
adjusted baseline population in each neighborhood (see Appendix 1-C for specific values), which
controls for different neighborhood population size in each year and levels the estimates between
growing, shrinking, and unchanging neighborhoods. This weight isolates mobility from
underlying population growth. Second, we try weighting by the adjusted neighborhood
population which has 9-digit zip code geocodes in years t and t+1. This weight preferences
neighborhoods with a higher degree of the more precisely located 9-digit geocodes and also
controls for population differences. Third, we try weighting by the proportion of a
neighborhood’s mobility rate derived from observations with 9-digit zip code geocodes in years t
and t+1. This weight preferences neighborhoods with a larger number of the more precisely
located 9-digit moves and controls for differences in mobility measures. Fourth, we try
weighting by the quotient of the first and second weights: the proportion of adjusted baseline
population that is attributed to 9-digit zip codes geocodes in years t and t+1 by the overall
adjusted baseline population. This creates a probability weight based on the 9-digit zip code
coverage in a neighborhood.
Table 6 compares an unweighted DID and FE model to the four weighted models with rail
station opening as the treatment variable. Across all weights, the models weighted by the
adjusted baseline population (weight 1) explain the largest amount of variation (the highest
adjusted-R
2
value), produce the highest estimate of the baseline mobility (constant term), and are
the only models with a statistically significant rail treatment effect. The unweighted model and
the 9-digit zip mobility rate proportionally weighted model (weight 4) do not provide jointly
significant results based on the F-test in the DID case. Models weighted by the proportion of 9-
digit zip codes in consecutive years throughout the population (weight 2) have the lowest
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
46

baseline mobility rate in the DID specification and a lower baseline mobility rate than weights 1
and 3 in the FE specification. Both weights employing proportions of 9-digit zip codes (weights
2 and 3) show a lower magnitude and not statistically significant coefficient on post*treatment,
compared to the weight 1 model. The AIC and BIC scores highlight that of weights 1, 2, and 3,
weight 1 is slightly less efficient, though differences are slight. We also compare unweighted and
weighted models for the rail announcement and the rail lag treatment variables (Tables 7 and 8).
The results on adjusted-R
2
, on AIC, and on BIC are largely the same as on the rail station
opening model.
Based on the comparison in Table 6, we reject the unweighted and the 9-digit proportional
mobility weighted (weight 4) models based on poor explanatory value and a lack of jointly
significant outcomes. The performance of the remaining weights is largely similar. We choose
the adjusted baseline population (weight 1) as it produces the most statistically significant results
in the largest amount of cases, is the easier weight to interpret, and avoids reliance on a
geocoding-derived weight. The remainder of the estimations in this paper use the adjusted
baseline population weight.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
47

Table 6. Model and Weights Comparison: Neighborhood Out Mobility Rates and Rail Station Openings for All Households.
Model Difference-in-Difference Fixed Effects
Weight Type Unweighted (1) (2) (3) (4) Unweighted (1) (2) (3) (4)
Treatment 0.0230** 0.00912 0.00717 0.0151* 0.0197*
(0.00902) (0.00961) (0.00816) (0.00884) (0.0110)
Post 0.00225 -0.00926 0.00626 0.00358 0.00601
(0.00710) (0.00629) (0.00601) (0.00632) (0.00736)
Treatment *
Post -0.00731 0.0187* 0.0143 0.0126 -0.0132 0.00434 0.00198 0.00188 0.00420 0.00457
(0.00981) (0.00949) (0.00889) (0.00914) (0.0115) (0.00337) (0.00368) (0.00249) (0.00325) (0.00369)
Constant
(baseline
mobility rate) 0.201*** 0.215*** 0.187*** 0.195*** 0.190*** 0.278*** 0.299*** 0.281*** 0.291*** 0.264***
(0.00626) (0.00712) (0.00557) (0.00576) (0.00654) (0.00420) (0.00387) (0.00478) (0.00546) (0.00566)
Year fixed
effect
No No No No No Yes Yes Yes Yes Yes
Neighborhood
fixed effect
No No No No No Yes Yes Yes Yes Yes
Adjusted R
2
0.022 0.054 0.041 0.052 0.009 0.820 0.872 0.874 0.869 0.814
AIC -8313.9 -9431.0 -9165.4 -8744.8 -8127.9 -13489.0 -15521.4 -15356.1 -14781.2 -13256.8
BIC -8289.9 -9407.0 -9141.4 -8720.8 -8103.9 -13369.0 -15401.5 -15236.1 -14661.2 -13136.8
F-test 3.09 5.11 4.89 6.61 1.4 63.05 36.53 50.47 49.73 55.59
Prob > F 0.0318 0.0028 0.0036 0.005 0.2477 0 0 0 0 0
Number of
Observations 2980 2980 2980 2979 2980 2980 2980 2980 2979 2980
Weight types: (1) Adjusted Baseline Population in Neighborhood, (2) Adjusted Baseline Population in Neighborhood, with 9-digit zip codes in
years t and t+1, (3) Proportion of Out Mobility Rate with 9-digit zip codes in years t and t+1, and (4) weight (2) divided by weight (1)
Notes: Standard errors clustered by treatment-control pair.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Table 7. Model and Weights Comparison: Neighborhood Out Mobility Rates and Rail Station Announcements for All Households.
Model Difference-in-Difference Fixed Effects
Weight Type Unweighted (1) (2) (3) (4) Unweighted (1) (2) (3) (4)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
48

Treatment 0.025** 0.014 0.005 0.011 0.021*
(0.01) (0.011) (0.008) (0.009) (0.012)
Post 0.009 0.001 0.009 0.008 0.011
(0.007) (0.008) (0.006) (0.006) (0.007)
Treatment *
Post -0.009 0.01 0.015* 0.015 -0.012 0.004 -0.001 -0.002 -0.001 0.004
(-0.011) (0.012) (0.009) (0.009) (-0.012) (0.004) (-0.004) (-0.003) (-0.003) (0.005)
Constant
(baseline
mobility rate) 0.195*** 0.208*** 0.184*** 0.191*** 0.185*** 0.278*** 0.299*** 0.282*** 0.292*** 0.263***
(0.006) (0.006) (0.005) (0.005) (0.006) (0.004) (0.004) (0.005) (0.005) (0.006)
Year fixed
effect
No No No No No Yes Yes Yes Yes Yes
Neighborhood
fixed effect
No No No No No Yes Yes Yes Yes Yes
Adjusted R
2
0.023 0.05 0.039 0.054 0.01 0.82 0.872 0.874 0.869 0.814
AIC -8315.49 -9418.56 -9159.46 -8750.76 -8128.4 -13488 -15520.1 -15355.7 -14776.5 -13255.5
BIC -8291.49 -9394.56 -9135.46 -8726.76 -8104.4 -13368 -15400.1 -15235.7 -14656.5 -13135.5
F-test 4.01 4.84 4.48 6.13 2.09 63.07 37.51 47.93 45.87 60.17
Prob > F 0.0104 0.0038 0.0059 0.008 0.1077 0 0 0 0 0
Number of
Observations 2980 2980 2980 2979 2980 2980 2980 2980 2979 2980
Weight types: (1) Adjusted Baseline Population in Neighborhood, (2) Adjusted Baseline Population in Neighborhood, with 9-digit zip codes in
years t and t+1, (3) Proportion of Out Mobility Rate with 9-digit zip codes in years t and t+1, and (4) weight (2) divided by weight (1)
Notes: Standard errors clustered by treatment-control pair.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Table 8. Model and Weights Comparison: Neighborhood Out Mobility Rates and Rail Station Five Years After for All Households.
Model Difference-in-Difference Fixed Effects
Weight Type Unweighted (1) (2) (3) (4) Unweighted (1) (2) (3) (4)
Treatment 0.019** 0.011 0.009 0.017** 0.015
(0.008) (0.008) (0.007) (0.008) (0.01)
Post -0.008 -0.015*** -0.002 -0.007 -0.004
(-0.007) (-0.006) (-0.006) (-0.007) (-0.007)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
49

Treatment *
Post -0.002 0.022*** 0.016* 0.013 -0.01 0.006* 0.010*** 0.006** 0.007* 0.003
(-0.007) (0.008) (0.008) (0.008) (-0.009) (0.003) (0.003) (0.003) (0.003) (0.004)
Constant
(baseline
mobility rate) 0.206*** 0.216*** 0.193*** 0.201*** 0.196*** 0.279*** 0.299*** 0.282*** 0.292*** 0.264***
(0.006) (0.006) (0.006) (0.006) (0.007) (0.004) (0.004) (0.005) (0.005) (0.006)
Year fixed
effect
No No No No No Yes Yes Yes Yes Yes
Neighborhood
fixed effect
No No No No No Yes Yes Yes Yes Yes
Adjusted R
2
0.027 0.061 0.033 0.047 0.014 0.82 0.874 0.875 0.87 0.814
AIC -8328.9 -9450.8 -9141.5 -8729.3 -8140.8 -13493.3 -15569.2 -15370.1 -14793.2 -13254.4
BIC -8304.7 -9426.8 -9117.5 -9705.3 -8116.8 -13373.3 -15449.2 -15250.1 -14673.2 -13134.4
F-test 2.63 8.58 3.83 5.49 1 59.16 41.02 48.98 48.44 55.42
Prob > F 0.0556 0.0001 0.0129 0.0018 0.399 0 0 0 0 0
Number of
Observations 2980 2980 2980 2979 2980 2980 2980 2980 2979 2980
Weight types: (1) Adjusted Baseline Population in Neighborhood, (2) Adjusted Baseline Population in Neighborhood, with 9-digit zip codes in
years t and t+1, (3) Proportion of Out Mobility Rate with 9-digit zip codes in years t and t+1, and (4) weight (2) divided by weight (1)
Notes: Standard errors clustered by treatment-control pair.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
50

Results
On a systemwide level, we find that the opening and continued presence of an L.A. Metro rail
station is associated with increased neighborhood mobility rates. In the opening year,
neighborhood out-move rates for all incomes increase by an average of 1.9 percentage points
(p.p.), which increases to 2.2 p.p. annually five years after opening (Table 9). Thus, systemwide
mobility rates are 8.7% and 10.2% higher owing to rail station opening and 5 year presence in
the neighborhood (Figure 10). Rail announcement does not appear to have a statistically
significant effect on neighborhood mobility systemwide (Figure 10). These results are based on
the DID models. In contrast, the FE model shows a lower magnitude (1.0 p.p.) increase in
mobility five years after opening and no statistically significant effect from rail station opening
(Table 9). Differences between the FE and DID models indicate that the systemwide rail opening
effect may be attributable to year-specific or neighborhood-specific idiosyncrasies. However, the
magnitude of the effect five years after opening is high enough to be statistically significant even
in the FE model.
Results differ by transit corridor by timing and statistical model. In the opening year, DID results
indicate a 1.3 p.p. increase in Gold Line neighborhood mobility (6.6% impact over baseline,
Figure 10), while the remaining corridors are not statistically significant (Figure 10, Table 14) In
the FE model, this Gold Line effect is non-existent, but Red/Purple Line station openings
increase out-mobility rates by 1.1 p.p. (Appendix 1-F Table 15). In both FE and DID models, the
Gold and Blue lines have increased mobility rates five years after rail station opening: 2.3 and
3.3 p.p. and 12% and 11.7% impact over mobility (Appendix 1-F Tables 14, 15; Figures 10, 11).
The FE model results also show a statistically significant 1.0 p.p. mobility increase on the
Red/Purple Line and a 1.7 p.p. decrease on the Green Line (Appendix 1-F Table 15). Upon
announcement, neighborhood mobility rates increased by 1.0 p.p. along the Expo Phase I Line in
both FE and DID models (Appendix 1-F Table 14, 15), a 5% impact over baseline. We save our
interpretation of corridor-level differences for the results income category below and for the
Discussion section.

Table 9. Rail Station Effects on Neighborhood Out Mobility Rates by Timing for All Incomes
Treatment
Variable
Rail Station
Announcement (~4.7
years before Opening)
Rail Station Opening 5 Years after Rail Station
Opened
Model Difference-
in-
Difference
Fixed
Effects
Difference-
in-
Difference
Fixed
Effects
Difference-
in-
Difference
Fixed
Effects
Treatment 0.014 0.00912 0.011
(0.011) (0.00961) (0.008)
Post 0.001 -0.00926 -0.015**
(0.008) (0.00629) (0.006)
Treatment * Post
0.010 -0.001 0.0187* 0.00198 0.022** 0.010**
(0.012) (0.004) (0.00949) (0.00368) (0.008) (0.003)
Constant
(baseline mobility
rate) 0.208*** 0.299*** 0.215*** 0.299*** 0.216** 0.299**
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
51

(0.006) (0.004) (0.00712) (0.00387) (0.006) (0.004)
Year fixed effect No Yes No Yes No Yes
Neighborhood
fixed effect
No Yes No Yes No Yes
Adjusted R
2
0.050 0.872 0.054 0.872 0.061 0.874
AIC 418.561 -15520.138 -9431.0 -15521.4 -9450.78 -15569.174
BIC 394.563 -15400.144 -9407.0 -15401.5 -9426.78 -15449.180
F-test 4.84 37.51 5.11 36.53 8.58 8.43
Prob > F 0.0038 0 0.0028 0 0.0001 0.0048
Number of
Observations 2980 2980 2980 2980 2980 2980
Note. Weighted by Adjusted Baseline Population in Neighborhood; Standard errors clustered by
treatment-control pair
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Figure 10. Difference in Difference Regression: Rail Station Impact of Opening, Announcement, and Five
Years after Opening on Neighborhood Baseline Out Mobility Rates, Systemwide and by Line
Model types: Difference-in-Difference; Weighted by Adjusted Baseline Population in Neighborhood;
Standard errors clustered by treatment-control pair; Data values displayed only for statistically significant
results
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
52

Figure 11. Fixed Effects Regression: Rail Station Impact of Opening, Announcement, and Five Years
after Opening on Neighborhood Baseline Out Mobility Rates, Systemwide and by Line
Model types: Fixed Effects; Weighted by Adjusted Baseline Population in Neighborhood; Standard errors
clustered by treatment-control pair; Data values displayed only for statistically significant results
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Results by Income Category
The effect and impact of rail stations is not uniform across the income spectrum. In both FE and
DID models, we find statistically significant neighborhood mobility increases for Lower-Middle
and Middle and Upper Income categories upon station opening and five years after opening
(Table 10). The FE model on announcement also shows increased out-mobility rates for the
Middle and Upper Income households and decreased out-mobility rates for Lowest Income and
Low Income groups (Table 10).

Table 10. Income Category Model of Rail Station Effects on Neighborhood Out Mobility Rates
Treatment
Variable
Rail Station
Announcement (~4.7
years pre-opening)
Rail Station Opening Five Years After Opening
Model Difference-
in-
Difference
Fixed
Effects
Difference-
in-Difference
Fixed
Effects
Difference-
in-
Difference
Fixed
Effects
Income = <30%
AMI 0.046*** 0.045*** 0.042*** 0.044*** 0.043** 0.042**
(0.007) (0.002) (0.007) (0.002) (0.006) (0.003)
Income = 30-
50% AMI 0.040*** 0.040*** 0.036*** 0.039*** 0.038** 0.037**
(0.007) (0.002) (0.007) (0.002) (0.006) (0.002)
Income = 50-
80% AMI 0.026*** 0.028*** 0.023*** 0.027*** 0.024** 0.026**
(0.007) (0.002) (0.006) (0.002) (0.005) (0.002)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
53

Treatment *
Income = <30%
AMI 0.013 0.010 0.008
(0.013) (0.012) (0.009)
Treatment *
Income = 30-
50% AMI 0.007 0.003 0.004
(0.011) (0.010) (0.009)
Treatment *
Income = 50-
80% AMI 0.011 0.005 0.007
(0.010) (0.009) (0.008)
Treatment *
Income = >80%
AMI 0.027** 0.014 0.020*
(0.010) (0.010) (0.009)
Post * Income =
<30% AMI 0.000 -0.010 -0.019**
(0.008) (0.007) (0.006)
Post * Income =
30-50% AMI 0.001 -0.009 -0.017*
(0.008) (0.007) (0.007)
Post * Income =
50-80% AMI 0.005 -0.004 -0.010
(0.008) (0.007) (0.006)
Post * Income =
>80% AMI 0.007 -0.007 -0.011*
(0.010) (0.008) (0.007)
Treatment * Post
* Income =
<30% AMI 0.004 -0.008* 0.009 -0.005 0.015* 0.003
(0.014) (0.004) (0.012) (0.004) (0.009) (0.004)
Treatment * Post
* Income = 30-
50% AMI 0.009 -0.007* 0.016 -0.004 0.020* 0.003
(0.013) (0.004) (0.010) (0.003) (0.008) (0.004)
Treatment * Post
* Income = 50-
80% AMI 0.012 0.001 0.023** 0.005 0.026** 0.014**
(0.012) (0.004) (0.009) (0.004) (0.008) (0.004)
Treatment * Post
* Income =
>80% AMI 0.013 0.015*** 0.031*** 0.018*** 0.031** 0.027**
(0.012) (0.005) (0.009) (0.005) (0.007) (0.004)
Constant
(baseline
mobility rate) 0.176*** 0.269*** 0.187*** 0.269*** 0.188** 0.271**
(0.009) (0.004) (0.010) (0.004) (0.008) (0.004)
Year fixed effect No Yes No Yes No Yes
Neighborhood
fixed effect
No Yes No Yes No Yes
Adjusted R
2
0.090 0.796 0.095 0.796 0.102 0.798
AIC -36130.878 -54096.601 -36198.044 -54092.795 -36279.656 -54177.742
BIC -36012.704 -53904.568 -36079.870 -53900.762 -36161.481 -53985.709
F-test 13.69 69.61 17.65 74.17 18.94 55.75
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
54

Prob > F 0 0 0 0 0 0
Number of
Observations 11919 11919 11919 11919 11919 11919
Model types: Difference-in-Difference and Fixed Effects; Weighted by Adjusted Baseline Population in
Neighborhood; Standard errors clustered by treatment-control pair
Reference Category: Control neighborhoods with income above 80% in years where rail stations were not
open (or announced or open for 5 years) in paired treatment neighborhoods
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

We next report results at the transit corridor level for each income group. Regression results are
in Appendix 1-F Tables 16-18 and rail station impacts are shown in Table 11 below.
On a transit corridor level, the Red/Purple and Gold Lines follow the systemwide pattern almost
exactly: rail opening is associated with a 9-16% mobility rate increase over baseline for middle
and upper income groups (>80% AMI) across both DID and FE specifications (Table 11). In a
few cases, this extends to the lower-middle income group (50-80% AMI) for the Red/Purple,
Gold, and whole system models (Table 11), but these are not significant for both FE and DID
models, reflecting less confidence in these estimates. There are no statistically significant effects
for the Green and Expo Phase I Lines, except for an ~5% increase in mobility rates over base line
for Lowest Income households (<30% AMI) in the FE specification (but not in the DID).
Likewise, there no statistically significant effects across both models for Gold or Red/Purple
Line households below 50% AMI.
Rail station announcement effects are most pronounced for the Expo Phase I Line (Table 11).
There, mobility rates increase 4-6% over baseline for the Lowest Income (<30% AMI) group and
3-5% for the All Incomes group, an effect statistically significant in both FE and DID models.
For Gold Line neighborhoods, the Middle and Upper Income group (>80% AMI) increases
mobility rates by 7-9% upon rail station announcement. There are no announcement effects for
Red/Purple Line neighborhoods.
Effects appear to be more pronounced five years after rail stations open (Table 11). Both for the
whole system and for the Blue Line, we find statistically significant associations increases in
mobility rates for both models in nearly every income group, with magnitudes of 3-17% over
baseline for the whole system and 8-21% for the Blue Line. These lagged effects also show up
for Middle and Higher Income (>80% AMI) households in Gold and Red/Purple Line
neighborhoods, but much less consistently for low income groups along those corridors.
The coefficient and impact estimates by income group provide a window into how the transit
system affects neighborhood mobility before, after and during opening, across the income
distribution. Variation by line, timing, and income group highlight the heterogeneity of impact
by neighborhood.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
55

Table 11. Rail Effect Magnitude Impact by Rail Corridor by Income
Rail Station Announcement
Income Group
Whole System
(N=11,920)
Red/Purple
(N=2,400)
Gold
(N=3,120)
Green
Expo Phase 1
(N=1,440)
Blue
Model FE DID FE DID FE DID

FE DID

All Incomes

3.5%* 5.0%* x
Lowest (<30% AMI)
-2.5%*

-3.8%*

4.5%* 5.9%* x
Low (30-50% AMI)
-2.3%*

x
Lower-Middle (50-80% AMI)

2.8%* x
Middle and Upper (>80% AMI)
5.6%***

7.5%** 8.7%*

x

Rail Station Opening
Income Group
Whole System
(N=11,920)
Red/Purple
(N=2,400)
Gold
(N=3,120)
Green
(N=1,840)
Expo Phase 1
(N=1,440)
Blue
Model FE DID FE DID FE DID FE DID FE DID

All Incomes
8.7%** 3.5%*

Lowest (<30% AMI)
-3.1%*

4.2%*
Low (30-50% AMI)

Lower-Middle (50-80% AMI)
11%** 4.5%* 5.7%*

Middle and Upper (>80% AMI)
6.7%*** 16.9%*** 13.3%** 9.9%* 5.9%* 11.6%*

Five Years After Opening
Income Group
Whole System
(N=11,920)
Red/Purple
(N=2,400)
Gold
(N=3,120)
Green
(N=1,840)
Expo Phase 1
Blue
(N=3,120)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
56

Model FE DID FE DID FE DID FE DID FE DID
All Incomes
3.3%** 10.2%** 3.2%*** 3.9%* 12%** -5.6%*

11.6%** 11.7%**
Lowest (<30% AMI)
6.5%*

8.1%* 12.9%**
Low (30-50% AMI)
8.8%* 10.4%**

8.6%* 8.8%*
Lower-Middle (50-80% AMI)
4.7%** 12.3%** 5.1%*** 11.2%** -6.3%*

13.5%** 11.7%**
Middle and Upper (>80% AMI)
10%** 16.5%** 13.1%*** 9.2%*** 9.7%** 15.9%**

21%*** 15.5%***

Note. Effect of Rail Station Effect variable divided by Baseline Mobility for each Income Group for Each Rail Line (corresponding to the
coefficient from regression results on Appendix 1-F Tables 16-18 on Treatment*Post*Income divided by the Constant). Impacts shown only when
effect is statistically significant. Impact range reflects differences between Fixed Effects and Difference-in-Difference Models. Regression models
weighted by Adjusted Baseline Population in Neighborhood. Standard errors clustered by treatment-control pair.
Restrictions: FTB data is unavailable to measure Green and Blue Lines announcement, Blue line opening, or 5 years after Expo Phase I opening.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
57

Results from Model Extensions
We extend the above models to test whether adding other variables will further clarify the
relationship between rail stations and neighborhood mobility.
Compared to the full 80-station model, rail station effect estimates excluding downtown and
excluding overlap stations are lower in magnitude (Appendix 1-G, Table 19). In the full system
regression model, rail station openings increase out-mobility by 8.7% relative to baseline and by
10.2% five years after opening (Figure 12). In contrast, these effects are of lower magnitude and
are no longer statistically significant when the 10 Downtown stations are excluded (Figure 12).
Moreover, when both the 10 Downtown and the 13 Overlap stations are removed, the magnitude
of the rail station effect is lower still and the results are not statistically significant (Figure 12;
Appendix 1-G, Table 20).
These results may indicate that some of the differences in neighborhood mobility may be
occurring in Downtown area stations or in neighborhoods that are densely served by the L.A.
Metro rail system. It would be difficult in the scope of this paper to analyze each station area
specifically to understand which neighborhoods alter the effect most. We leave this to future
work.

Figure 12. Rail Effect Impact of Downtown and Overlapping Station Restrictions on Neighborhood Out-
Mobility Rate
Notes: Difference-in-Difference Weighted by Adjusted Baseline Population in Neighborhood; Standard
errors clustered by treatment-control pair; Data values displayed only for statistically significant results
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data
We test whether differences in train technology alter the rail effect on neighborhood mobility.
Separating stations by subway versus light rail, we find no statistically significant differences
between the two types of trains on station opening or announcement. We do find that light rail
stations increase the out-mobility rate relative to neighborhood baseline move rates by 2.5
8.7%*
10.2%**
-4%
-2%
0%
2%
4%
6%
8%
10%
12%
Whole System Excluding Downtown Excluding Overlap Station and
Downtown Stations
Impact of Rail-Related Effect on Baseline
Mobility Rate
Rail Station Announcement Effect Rail Station Opening Effect Five Years after Station Opens
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
58

percentage points five years after rail stations open (Appendix 1-G, Table 21). This increases
out-move rates by 12% relative to baseline move rates in those neighborhoods (Figure 13). These
findings indicate that the out-mobility rate is not too sensitive to the relatively faster and more
frequent service on a subway and its underground location, relative to light rail.

Figure 13. Subway versus Light Rail Station Differences Impact of Rail Effect on Baseline Neighborhood
Out-Mobility
Model types: Difference-in-Difference Weighted by Adjusted Baseline Population in Neighborhood;
Standard errors clustered by treatment-control pair
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data
We also test whether the location of a station within the highway median alters the rail effect on
neighborhood mobility. Regressions incorporating a separation by highway median reveal
increased out-mobility at station opening and five years later for stations not in highway medians
and decreased out-mobility for those located in highway medians (Appendix 1-G, Table 22).
Figure 14 highlights the large impact of this variable relative to baseline mobility for each group.
Highway median stations decrease neighborhood mobility rates by as much as 15-16%, while
non-highway median stations see an increase of 10-11%. The decrease due to highway median is
larger than for any other variable in any other model we test.
The results on highway median stations possibly indicate the lack of development activity and/or
price pressure around these stations. Their location in the center of a freeway may not be
attractive to developers or to in-moving households. Therefore, existing residents are less likely
to move out, driving the neighborhood out-move rate down. Additionally, highway median
location may indicate lower ridership or potentially a less walkable neighborhood, each of which
could lead to decreased out-mobility rates.
12.0%**
-10.0%
-5.0%
0.0%
5.0%
10.0%
15.0%
Light Rail Subway Impact of Rail-Related Effect on
Baseline Mobility Rate
Announcement Opening 5 Years After Opening
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
59

Figure 14. Highway Median Station Location versus non-Highway Median Station Location Impact of
Rail Effect on Baseline Neighborhood Out-Mobility Stations
Model types: Difference-in-Difference Weighted by Adjusted Baseline Population in Neighborhood;
Standard errors clustered by treatment-control pair; Data only displayed for statistically significant effects
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data
We test whether system size, proximity to Downtown Los Angeles, and the existence of a
transfer point between lines alter the rail effect. We find that these variables reduce the
magnitude of rail effects negligibly: 0.1-0.2 percentage points or 0.2-0.3% of baseline, relative to
the model without these variables (Table 10). Aside from altering the rail effect, system size
appears to decrease the neighborhood out-mobility rate by 1.4-1.7 p.p. for a 6-7% impact (Table
10). Yet, in order to achieve this stabilizing effect, the system would need to add 100 stations or
effectively double its current size. Based on these calculations, adding the 18 stops currently
under construction (Crenshaw Line, Regional Connector, and Purple Line Extension) would on
net decrease the neighborhood out-mobility rate by 1-1.3%. Larger systems are attractive to
riders. If we assume that current residents have some preference for living near train stations,
then they will benefit from the increased access provided by a larger system and may be willing
to pay more or make more effort to stay in the neighborhood to utilize the transit service.
The more stops away a station is from Downtown Los Angeles, the more its neighborhood out-
mobility increases (Appendix 1-G, Table 23). Compared to a downtown station, a station 10
stops away from Downtown has a 1.2-1.4 p.p. higher out-mobility rate reflecting an impact of 5-
6% percent on the baseline (Table 10). Many L.A. Metro rail riders value access to Downtown
Los Angeles for employment, shopping, entertainment, transportation access or other reasons.
They thus value having a shorter ride (fewer stops) to Downtown they and may be willing to pay
more or make more effort to stay in a neighborhood relative to households living in station area
neighborhood with more stops to Downtown.
The L.A. Metro rail system is focused on Downtown LA and there are currently relatively few
transfer points between lines otherwise. As a result, these transfer neighborhoods may not be as
10%*
-15%*
11%**
-16%**
-30%
-25%
-20%
-15%
-10%
-5%
0%
5%
10%
15%
Not in Highway Median Highway Median
Impact of Rail-Related Effect on
Baseline Mobility Rate
Announcement Opening 5 Years After Opening
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
60

valuable to residents as they are in more distributed transit systems. As such we find no mobility
effect of being closer to a transfer station (Table 10).

Table 12. Summary of Estimates and Impacts of Transit Network Variables on Neighborhood Out
Mobility Rates
Transit System-
Level Options
Coefficient Estimate
Magnitude Impact on
Neighborhood Out-Mobility
Baseline Rate
Rail
Station
Announce
ment
Model
Rail
Station
Opening
Model
Five
Years
after
Station
Model
Rail
Station
Announce
ment
Model
Rail
Station
Opening
Model
Five
Years
after
Station
Model
Add 100 Stations
(double the system)
-1.4%** -1.7%** -1.4%*** -6.9%** -5.9%** -6.5%***
Move Station 10
Stops Closer to
Transfer Station
1.0% 1.0% 0.0% 5.0% 4.0% 0.0%
Move Station 10
Stops Further from
Downtown Los
Angeles
1.2%* 1.3%* 1.4%** 6.2%* 6.2%* 5.5%**
Rail Effect 1.0% 1.8%* 2.0%* 4.9% 8.9%* 9.9%*
Model types: Difference-in-Difference; Weighted by Adjusted Baseline Population in Neighborhood;
Standard errors clustered by treatment-control pair
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Discussion and Conclusion
We explore the rail transit – displacement link and arrive at four main conclusions. First, the
opening and continued presence of rail transit stations increases neighborhood outflow rates by
up to 10% annually above baseline. Second, we find little systemwide evidence that the opening
of a rail station disproportionately increases mobility rates for Lowest and Low Income
households (below 50% AMI), and announcement effects only seem salient for the Expo Phase I
Line. However, five years after stations open, we estimate 0-9% additional outflow due to rail
stations for the Lowest and Low Income groups systemwide and along the Blue Line. Third,
evidence suggests that rail effects increase mobility rates for Middle and Upper Income groups
(>80% AMI) most often. We also find significant heterogeneity by transit corridor by income.
For example, the longer-term presence of rail along the Blue Line, appears to increase
neighborhood outflow across incomes by 8 to 21% -- the largest impact in our study, but no
impact for the Green Line and very little for the Red/Purple Line. Our results signal that rail
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
61

station effects on neighborhood out-mobility rates may be most important years after stations
open. This suggests the need for further evaluation at perhaps even further time periods including
10 or 15 years after opening.
Our descriptive analyses shed additional light on population shifts over time in these transit
corridors. We observe a growing population of households in the Los Angeles County and along
each transit line, with the County and the Red/Purple, Blue, and Green Line neighborhoods
gaining 40-50% more households from 1993-2013. Neighborhood composition has changed
drastically toward increased proportions of Middle and Upper Income households in all corridors
except Expo Phase I. The relative growth of Middle and Upper Income households has been
much larger in rail station neighborhoods than their control counterparts. We believe these
compositional changes indicate that a process of gentrification has unfolded on a corridor-wide
level along the Gold, Blue, Red/Purple, and Green Lines over the two decades under study.
8
We
also observe no net loss of Lowest, Low, or Lower-Middle Income populations in any case, just
a slower gain in most cases. This may have some of the hallmarks of exclusionary displacement
on the neighborhood level, where existing lower-income households are not directly displaced,
but new ones may not be able to afford to enter (Marcuse, 1985).
These descriptive findings are in line with other speculations about neighborhood gentrification
in Los Angeles County in the wake of the housing affordability crisis, the Great recession and its
recovery, spatial changes in employment, among other factors. The Urban Displacement Project
in Los Angeles County had conducted an analysis of gentrification at the census tract level and
their results indicated that 1,040 of 2,347 tracts (44 percent) were at-risk of gentrifying in the
periods 1990-2000 and 2000-2013. Of these census tracts, 15 percent actually gentrified in either
(or both) decades (author calculations of data from Zuk & Chapple, 2015). The Urban
Displacement Project considered all gentrifying tracts at risk of displacement, but did not give
specific information on movers into and out of these neighborhoods. Our findings confirm these
gentrification trends and provide additional nuance to the out-movement patterns from these
neighborhoods.
We also utilize the FTB dataset to describe how often households move on average, regardless of
any rail intervention. Our data indicate a 21% average annual move rate for Los Angeles County
as a whole, in line with a weighted average of national level move statistics (U.S. Census
Bureau, 2016), and survey based work in Los Angeles (Clark & Ledwith, 2006) and in low-
income neighborhoods nationally (Coulton et al., 2012). Middle and Upper Income households
have an average mobility rate of 18% countywide, 5 percentage points below other income
groups, because they are more likely to be homeowners, who tend to move less often than
renters. Blue and Red/Purple Line neighborhoods have mobility rates that are 5-8 percentage
points higher than both the County average and other transit corridors.

8
Because we use relative measures of income (percentage of AMI) we can make a credible claim about
gentrification and not just income growth for existing area residents.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
62

Combining the above conclusions, we find islands of displacement in seas of gentrification in
oceans of mobility.
9
Moving is the norm in an the overarching urban system where one fifth of
households moves annually. In L.A. Metro rail corridors, household growth overall and
compositional change toward higher incomes are also the norm. In comparison, rail-induced
displacement of lower-income household is not the norm. In fact, outflow rates above baseline
mobility are highest for Middle and Upper Income. Nevertheless, in certain cases, the opening
and continued presence of new rail stations does increase outflow rates by a non-trivial 5-10%
over baseline.
Our study contributes to the displacement literature in a few ways. It provides a new
methodology and data type to address the question. Our findings that impacts on neighborhood
out-mobility vary meaningfully by rail corridor and by time deepen the understanding of transit
system externalities. We can not fully reject the hypothesis that new rail stations are associated
with a larger outflow of lower-income households. Yet, at the same time, our evidence suggests
that mobility increases are affecting a different portions of the income spectrum in different
times on different lines. These findings are in line with the mixed results documented by recent
literature reviews (Zuk et al., 2017; Zuk et al., 2015; Rayle, 2015).
Our study also points to the need for more longitudinal research on gentrification, displacement,
and public transit systems. Our five year time frame before and after stations open may perhaps
miss some of the mobility-related changes in the neighborhood that occur even later after
opening. In addition, future research should consider other coping mechanisms through which
lower-income households may cope with high baseline mobility and gentrification in their
neighborhoods, including seeking housing assistance or affordable housing, decreasing housing
search costs, paying more to stay in a neighborhood, doubling up, downsizing housing units,
changing consumption, selling assets, or working more (see Chapter 2).

Planning and Policy Directions
The confluence of mobility and gentrification present challenges and opportunities for urban
planners and policymakers. Our research raises questions about current practice: Do public
officials take into account that 21% of households in Los Angeles County move every year? Do
fund allocations, policy decisions, and ordinance changes adapt to new households? Have
planners and policymakers adjusted to the drawn out onset of gentrification, or perhaps have they
over-adjusted? These questions are out of scope for this paper, but are important in bringing
practice in line with the facts on the ground. Further research is needed on these topics.
For transportation planners and rail transit operators, our findings yield additional outcomes. The
nuanced link between rail transit investments and displacement of lower-income households
from our study does not take the pressure off of these stakeholders. To stem displacement in the

9
We purposefully use the marine geography metaphor to connect to an older debate on gentrification versus urban
decay. Wyly & Hammel (1999) article “Islands of Decay in Seas of Renewal: Housing Policy and the Resurgence of
Gentrification” found resurgent gentrification across the U.S. in the 1990s in the recovery period of the early 1990s
economic recession. This article was a response to Berry’s (1985) “Islands of Renewal in Seas of Decay” who had
claimed that gentrification was a passing fad that would dissipate after the next recession.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
63

places where it is already happening, operators and planners should work with local and regional
stakeholders to alleviate pressure on the property market, by increasing both unit supply and set-
asides for households at a variety of income levels below 80% AMI. These increases should
reflect at least some understanding of who is moving in as well as who is moving out, given the
21% average annual move rate in Los Angeles County. In planning new rail lines and extensions,
care should be taken to address displacement concerns before stations open, through
participatory station area planning, community benefits agreements, and acknowledging the need
to increase unit density and affordable set-asides. The growth in households along rail corridors
also presents a real opportunity for rail transit operators to engage new households to increase
ridership. Possible strategies include transit demand management, better first and last mile
connections, attractive design of egress and ingress, coordination with bus, paratransit, bike, and
transportation network company connections, and partnerships with local employers and
businesses.

Data Innovation and Limitations
The FTB dataset and new mobility and displacement measurements provide an improvement
over past attempts to quantify mobility and the rail – displacement connection. The data provide
a longitudinal way to track households across time by income and to group households into
neighborhoods. The data and methods address many of the methodological criticisms of past
studies, including having comparable baselines, tracking the same households over time, having
a large enough sample to draw neighborhood-level conclusions, robust model performance,
appropriately scaled neighborhoods, and a control group drawn from the same dataset (Zuk et al.,
2015; Rayle, 2015). Our study highlights the high potential of administrative data, and tax return
data in particular, to study population level mobility processes at the neighborhood level,
including gentrification and displacement.
As with any new data source, the FTB data come with their own particular limitations.
Confidentiality requirements complicate the geocoding strategy. We take pains to alleviate the
concerns of mis-assignment to a particular neighborhood by layering in both 9-digit and 5-digit
zip codes, while preferencing 9-digit zip codes where available, and we consider weighting by 9-
digit zip code proportions. The impact of mis-assignment on estimates is unclear. Mis-
assignment may slightly over-assign households to particular neighborhoods and under-assign to
others. Likely, the impact balances out between under- and over-estimation.
The slight geographic instability of 9-digit zip codes over time presents additional problems. We
rectify these by using multiple years of 9-digit zip code coordinates and by limiting move rate
estimates to moves of at least one half-mile. This strategy likely undercounts actual mobility,
since some households certainly make very short-distance moves. Therefore, we provide a
lower-bound estimate on the effects of rail station openings on neighborhood mobility and
displacement.
Data dropouts present an additional challenge. The data is limited to households that file taxes in
years that they file. We can not add in additional tax filers from other sources and we do not
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
64

generate synthetic tax filers. We confront this issue by only considering households who file
taxes in consecutive years. This reduces the sample size by about 15%. To make sure this
restriction does not disproportionately affect certain neighborhoods, we weighting our analyses
by the population in each neighborhood in each year adjusted for consecutive filers.
The censoring presents a further issue because the distribution of non-filers skews toward the
bottom of the income distribution. In California, about 11% of households who should be filing,
do not file taxes (FTB, 2006; FTB, 2017). But, households with incomes below $25,125 for
families and $12,562 for individuals in 2013 dollars are not required to file California income
taxes (FTB, 2013). At the same time, evidence from EITC compliance suggests that at least 75%
of low-income Californians file taxes, even for those who are not mandated (IRS, 2013). Still,
there is a potential gap of up to 14% (89% - 75%) of California tax filing across the distribution.
There are several circumstances that possibly mitigate this limitation. The data does not show a
spatial skewing of income-based non-filing. Therefore, we are no more likely to have fewer low-
income filers in one station area than another or in a station area versus a control area.
Additionally, over half of the sample represent neighborhoods with over 50% of households who
earned fewer than 50% of AMI ($25,000 in 2013). Though this population may be filing at a
slightly lower rate, they are overrepresented in our study area and sample. While the censoring
remains a concern, our overall identification strategy, inclusion of year and neighborhood fixed
effects, and weighting by annual household population help allay some of the concern.
Our direct estimation of the rail – displacement link strengthens model identification, but does
not directly address the gentrification mechanism. In future work, we will work toward a two-
stage estimation strategy, measuring the relationship of rail transit to gentrification and
gentrification to displacement using the same dataset. It is also likely that the processes of
gentrification and displacement reinforce and influence one another. To address this, we will
consider a system of simultaneous equations in future work. Even with our directly-modeled
setup, the study provides a clear identification strategy to assess the relationship between rail and
displacement.
This paper builds on a trajectory of displacement research and provides a rigorous approach to
understanding the link between displacement and rail transit development. The new dataset and
measurements will inform future studies. The descriptive and statistical findings we hope will
help policymakers and planners consider neighborhood mobility processes into account more
often in decision-making and allocations of public resources.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
65

Appendix
1-A. Los Angeles County: Annual Number of Observations with and without Geocodes
Year
Observations
with No
Geocodes
Observations
with
Geocodes
Total
Observations
1993 135,081 3,955,863 4,090,944
1994 138,174 4,018,980 4,157,154
1995 125,930 4,101,161 4,227,091
1996 72,535 4,206,264 4,278,799
1997 81,618 4,307,095 4,388,713
1998 86,781 4,427,481 4,514,262
1999 68,771 4,569,994 4,638,765
2000 61,584 4,710,441 4,772,025
2001 57,033 4,779,360 4,836,393
2002 37,561 4,820,743 4,858,304
2003 36,207 4,747,424 4,783,631
2004 42,727 4,952,670 4,995,397
2005 48,143 5,062,591 5,110,734
2006 48,951 5,176,470 5,225,421
2007 49,419 5,331,896 5,381,315
2008 47,129 5,279,720 5,326,849
2009 46,348 5,211,823 5,258,171
2010 46,152 5,259,581 5,305,733
2011 44,964 5,283,107 5,328,071
2012 43,361 5,253,149 5,296,510
2013 40,785 5,186,696 5,227,481
Total 1,359,254 100,642,509 102,001,763
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
66

1-B. Baseline Populations and Mobility Rates by Year

Table 13. Treatment and Control Station Descriptive Statistics: Distance, Year Opened, Sample Size, Out-Mobility Rate, System and Station
Characteristics
# Station Name Control Intersection Branch
Station
Opening
Year
Miles
between
Treatment
& Control
Centroid
Adjusted Baseline
Population Out-Mobility Rate
Downtown Overlap
Highway
Median
Stops
to
CBD Treatment Control Treatment Control
1
Civic Center /
Grand Park
1
st
/ 2
nd
/ Lucas / Beverly
/ Glendale
Red 1993 1.0 23,600 51,044 28% 24% Yes 0
2
Hollywood /
Highland
Fairfax / Santa Monica Red 2000 1.5 39,331 44,091 32% 29% 8
3
Hollywood /
Vine
Melrose / La Brea Red 1999 1.7 157,260 33,504 26% 22% Yes 7
4
Hollywood /
Western
Wilton / Santa Monica Red 1999 0.9 49,138 47,115 33% 26% 6
5 North Hollywood
Victory / Lankershim /
Colfax
Red 2000 1.4 165,777 170,814 26% 23% 10
6 Pershing Square San Pedro / 8
th
St Red 1993 0.9 57,706 23,476 25% 27% Yes 0
7
Universal City /
Studio City
Ventura / Laurel Canyon Red 2000 1.9 16,852 164,454 37% 24% 9
8 Union Station Main / Griffin Red 1993 1.5 63,739 40,555 25% 16% Yes 1
9
Vermont /
Beverly
Western / Beverly Red 1999 1.0 54,496 268,687 31% 22% 3
10
Vermont / Santa
Monica
Sunset / Silver Lake Red 1999 1.2 185,206 259,233 20% 23% Yes 4
11 Vermont / Sunset Rowena / Hyperion Red 1999 1.4 49,617 34,053 21% 25% 5
12
Westlake /
MacArthur Park
Venice / Hoover Red 1993 1.0 154,375 57,761 21% 20% 1
13
Wilshire /
Normandie
Pico / Western Purple 1996 1.1 199,727 50,265 25% 24% Yes 3
14
Wilshire /
Vermont
Beverly / Rampart Purple 1996 1.0 75,041 71,821 31% 22% 2
15
Wilshire /
Western
Wilshire / La Brea Purple 1996 1.0 322,613 47,540 24% 23% 4
16 Allen Washington / Allen Gold 2003 1.2 27647 34332 23% 14% Yes 12
17 Atlantic Garfield / Riggin Gold 2009 1.4 34299 31792 13% 12% 9
18 Chinatown Sunset / Echo Park Gold 2003 1.5 82086 52408 22% 19% 2
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
67

19 Del Mar California / Allen Gold 2003 2 19620 9930 33% 24% Yes 9
20
East Los Angeles
Civic Center
Beverly / Garfield Gold 2009 1.4 44668 37134 13% 16% Yes 8
21 Fillmore
Huntington / Garfield /
Atlantic / Los Robles
Gold 2003 2 20933 26937 31% 22% 8
22 Highland Park OR: York / Avenue 50; Gold 2003 1.1 199165 51885 23% 15% 6
23 Heritage Square Heritage / Soto Gold 2003 1.3 32242 13642 17% 21% Yes 4
24 Indiana Olympic / Ditman Gold 2009 1.1 65874 42652 14% 14% 6
25 Lake Lake / Washington Gold 2003 1.2 34518 132792 27% 22% Yes 11
26
Lincoln Heights /
Cypress Park
Cypress / Division Gold 2003 1.9 100325 29566 22% 15% 3
27
Little Tokyo /
Arts District
7
th
and Santa Fe? Gold 2009 1.2 26173 1277 26% 48% Yes 2
28 Memorial Park Fair Oaks / Washington Gold 2003 1.4 107426 103053 28% 21% 10
29 Mariachi Plaza Olympic / Lorena Gold 2009 2.1 121708 86657 21% 22% Yes 4
30 Maravilla Olympic / Atlantic Gold 2009 1.3 43748 37764 15% 14% 7
31 Pico / Aliso Soto / 8
th
Gold 2009 1.4 16382 20225 22% 17% 3
32
Sierra Madre
Villa
California / Rosemead Gold 2003 1 19244 17458 14% 21% Yes 13
33 Soto City Terrace / Pomeroy Gold 2009 1.3 145377 128210 19% 19% 5
34 South Pasadena Huntington / Main Gold 2003 1.4 128678 41880 21% 19% 7
35
Southwest
Museum
Eastern / Huntington Gold 2003 1.8 24660 143511 19% 20% 5
36 Avalon Avalon / 135
th
Green 1995 1.3 33,738 24,789 16% 17% Yes 11
37 Aviation / LAX Hawthorne / El Segundo Green 1995 1.7 8,535 326,071 16% 25% Yes 16
38 Crenshaw Crenshaw / Century Green 1995 1.4 15,392 16,402 15% 22% Yes 14
39 Douglas Rosencrans / Hawthorne Green 1995 1.8 7,144 50,582 11% 19% 19
40 El Segundo Main / Grand in El Seg Green 1995 1.7 339 107,375 33% 21% Yes 18
41 Harbor Freeway Vermont / Century Green 1995 1.3 74,230 48,717 22% 16% Yes Yes 12
42
Hawthorne /
Lennox
La Brea / Arbor Vitae Green 1995 1.3 46,696 172,881 17% 18% Yes 15
43
Lakewood
Boulevard
Paramount / Stewart and
Gray
Green 1995 1.5 38,618 37,984 14% 20% Yes 12
44
Long Beach
Boulevard
Atlantic / Imperial Green 1995 1.6 147,126 48,551 25% 16% Yes 11
45 Mariposa Aviation / Arbor Vitae Green 1995 2.1 455 8,693 30% 36% Yes 17
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
68

46 Norwalk
Pioneer / Rosecrans /
San Antonio
Green 1995 1.5 26,135 222,971 15% 28% Yes 14
47 Redondo Beach
Sepulveda / Manhattan
Beach Blvd
Green 1995 1.6 1,841 31,327 25% 15% 20
48
Vermont /
Athens
Vermont / 135
th
Green 1995 1.4 33,063 34,624 20% 17% Yes 13
49 Anaheim Street
Santa Fe / Pacific Coast
Highway
Blue 1990 1.6 170,293 15,479 24% 20% Yes 17
50 Artesia
Long Beach Blvd /
Greenleaf
Blue 1990 1.1 3,267 43,202 19% 18% 12
51 Compton Compton / Bullis Blue 1990 1.0 30,766 133,205 19% 25% 11
52 Del Amo Central / University Blue 1990 2.3 0 105,671 N/A 20% 13
53
Downtown Long
Beach
2
nd
/ Livingston Blue 1990 3.1 196,051 51,277 27% 19% Yes 19
54 5
th
Street Cherry / 7
th
Blue 1990 1.2 205,226 42,919 27% 30% 18
55 Florence Avalon / 79
th
Blue 1990 1.3 123,524 56,115 22% 15% 7
56 Firestone Firestone / State Blue 1990 1.8 49,267 63,606 16% 18% 8
57 1
st
Street Cherry / 7
th
Blue 1990 1.2 203,517 42,919 27% 30% Yes 20
58 Grand / LATTC Adams / Normandie Blue 1990 1.8 52,807 114,746 28% 19% 2
59 Pacific Avenue 2
nd
/ Livingston Blue 1990 3.1 209,384 51,277 27% 19% Yes 18
60
Pacific Coast
Highway
Cherry / Pacific Coast
Highway
Blue 1990 1.2 41,203 37,299 22% 23% 16
61 Pico Venice / Hoover Blue 1990 1.1 60,328 57,761 27% 23% Yes 1
62 San Pedro Street
Jefferson / Avalon / San
Pedro
Blue 1990 0.9 31,279 63,166 17% 16% Yes 3
63 Slauson Miles / Gage Blue 1990 1.3 37,001 29,857 18% 17% 6
64
7
th
Street / Metro
Center
San Pedro / 8
th
Blue 1990 0.9 95,237 23,476 22% 31% Yes 0
65 Vernon Avalon / Vernon Blue 1990 1.3 34,832 177,293 17% 23% 5
66 Washington Slauson / Atlantic Blue 1990 4.0 13,762 115,958 16% 16% 4
67 Wardlow Orange / Bixby Blue 1990 1.2 35,076 153,845 17% 20% 14
68
Willowbrook /
Rosa Parks
Wilmington / Stockwell Blue 1990 1.3 26,754 102,733 19% 18% Yes 10
69 Willow Willow / Cherry Blue 1990 1.3 132,471 29,397 24% 20% 15
70
103
rd
Street /
Watts Towers
Century / Avalon Blue 1990 2.1 29,435 56,115 19% 15% 9
71 Culver City Culver / Overland Expo I 2012 1.3 84,785 40,332 20% 14% 11
72 Expo / Crenshaw
Vernon / Crenshaw /
Leimert
Expo I 2012 1.3 34,452 25,444 15% 16% 7
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
69

73 Expo Park / USC Vermont / Vernon Expo I 2012 1.1 8,944 140,346 31% 21% Yes 4
74 Expo / Vermont Vermont / Vernon Expo I 2012 1.0 28,071 140,346 23% 21% 5
75 Expo / Western Western / Vernon Expo I 2012 1.0 60,536 95,317 15% 18% 6
76 Farmdale Adams / Arlington Expo I 2012 1.7 32,604 149,181 13% 21% Yes 8
77 Jefferson / USC Main / Vernon Expo I 2012 1.3 94,664 61,082 31% 17% 3
78
Expo / La Brea /
Ethel Bradley
La Brea / Washington Expo I 2012 1.2 169,032 45,465 20% 16% 9
79
La Cienega /
Jefferson
Pico / Fairfax Expo I 2012 1.8 20,746 48,864 13% 18% 10
80
LATTC / Ortho
Institute / 23
rd
St
Adams / Normandie Expo I 2012 1.5 10,562 114,746 28% 21% Yes 2

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
70

1-C. Area Median Income by Year for the Los Angeles – Long Beach Metropolitan
Statistical Area
Source: U.S. HUD
Year
Area Median
Income (AMI)
80% of AMI 50% of AMI 30% of AMI
1993 $33,840 $27,072 $16,920 $10,152
1994 $36,160 $28,928 $18,080 $10,848
1995 $36,160 $28,928 $18,080 $10,848
1996 $37,520 $30,016 $18,760 $11,256
1997 $38,240 $30,592 $19,120 $11,472
1998 $39,840 $31,872 $19,920 $11,952
1999 $41,040 $32,832 $20,520 $12,312
2000 $41,680 $33,344 $20,840 $12,504
2001 $43,600 $34,880 $21,800 $13,080
2002 $44,080 $35,264 $22,040 $13,224
2003 $40,240 $32,192 $20,120 $12,072
2004 $43,360 $34,688 $21,680 $13,008
2005 $43,560 $34,848 $21,780 $13,068
2006 $44,960 $35,968 $22,480 $13,488
2007 $45,200 $36,160 $22,600 $13,560
2008 $47,840 $38,272 $23,920 $14,352
2009 $49,680 $39,744 $24,840 $14,904
2010 $50,400 $40,320 $25,200 $15,120
2011 $51,200 $40,960 $25,600 $15,360
2012 $51,840 $41,472 $25,920 $15,552
2013 $49,520 $39,616 $24,760 $14,856

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
71

1-D. Control Neighborhood Location

Figure 15. Map of Los Angeles City Neighborhoods and Red/Purple Subway Line Stations and Control
Intersections
Note: Each red and blue dot represents a circular neighborhood with radius of one-half mile, centered
about a Red/Purple Line station or control intersection. Note dots are not to scale.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
72

Figure 16. Map of Los Angeles City Neighborhoods and Gold Light Rail Line Stations and Control
Intersections
Note: Each red and blue dot represents a circular neighborhood with radius of one-half mile, centered
about a Red/Purple Line station or control intersection. Note dots are not to scale.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
73

Figure 17. Map of Expo Phase I, Green, and Blue Line Stations and Control Intersections
Note: Each red and blue dot represents a circular neighborhood with radius of one-half mile, centered about a Red/Purple Line station or control
intersection. Note dots are not to scale.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
74

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
75

1-E. Parallel Trends Assumption Graphs
To demonstrate that the parallel trends assumption holds, we graph the dependent variable –
annual neighborhood out-mobility rate – for treatment versus control households over the study
period and highlight when treatments occurred. Since each rail corridor has its own opening, we
draw the graph relative to the number of years before and after rail opening, with year 0 on the x-
axis as the opening year, in the graphs below. The five-year anniversary year occurs when year
equals 5. The average time between station announcement and opening across the L.A. Metro
system is 4.7 years. We round this to 5 years; hence, station announcement occurs when year
equals -5 in the graphs below.
Figure 18. A-F: Parallel Trends Graphs: Neighborhood Out-Mobility Rate of Treatment versus Control
Neighborhood by Years Before and After Rail Station Opening for each Rail Corridor

Legend: Announcement Year
Opening Year
Five Year Opening Anniversary

0%
5%
10%
15%
20%
25%
30%
35%
-19
-18
-17
-16
-15
-14
-13
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Full L.A. Metro System
Control Treatment
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
76

0%
5%
10%
15%
20%
25%
30%
35%
-19
-18
-17
-16
-15
-14
-13
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Red/Purple Line
Control Treatment
0%
5%
10%
15%
20%
25%
30%
35%
-19
-18
-17
-16
-15
-14
-13
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Gold Line
Control Treatment
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
77

0%
5%
10%
15%
20%
25%
30%
35%
-19
-18
-17
-16
-15
-14
-13
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Green Line
Control Treatment
0%
5%
10%
15%
20%
25%
30%
35%
-19
-18
-17
-16
-15
-14
-13
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Expo Phase I Line
Control Treatment
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
78

0%
5%
10%
15%
20%
25%
30%
35%
-19
-18
-17
-16
-15
-14
-13
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Blue Line
Control Treatment
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
79

1-F. Regression Results by Income by Transit Line

Table 14. DID Model Rail Station Effects on Neighborhood Out Mobility Rates by Timing and Rail Corridor
Treatment
Variable
Rail Station Announcement Rail Station Opening 0 Years After Opening
Rail Corridor Red/
Purple
Gold Expo
Phase 1
Red/
Purple
Gold Green Expo
Phase 1
Red/
Purple
Gold Green Blue
Treatment 0.029** 0.016 0.013 0.018 0.015 0.004 0.017 0.018 0.016* -0.010 0.013
(0.003) (0.010) (0.025) (0.014) (0.010) (0.035) (0.025) (0.012) (0.009) (0.032) (0.017)
Post -0.096** -0.032** -0.028** -0.041** -0.020** -0.088** -0.030** -0.026*** -0.007 -0.050*** -0.089***
(0.007) (0.005) (0.003) (0.004) (0.006) (0.011) (0.004) (0.005) (0.008) (0.006) (0.012)
Treatment *
Post -0.007 0.007 0.010* 0.007 0.013* -0.025 0.005 0.008 0.023** -0.012 0.033**
(0.012) (0.006) (0.005) (0.009) (0.006) (0.019) (0.007) (0.008) (0.009) (0.012) (0.014)
Constant
(baseline
mobility rate) 0.322** 0.208** 0.200** 0.259** 0.197** 0.301** 0.191** 0.242*** 0.191*** 0.256*** 0.282***
(0.005) (0.009) (0.010) (0.004) (0.008) (0.020) (0.009) (0.004) (0.008) (0.017) (0.015)
Year fixed
effect
No No No No No No No No No No No
Neighborhood
fixed effect
No No No No No No No No No No No
Adjusted R
2
0.122 0.112 0.068 0.195 0.056 0.296 0.033 0.132 0.046 0.242 0.344
AIC -2199.72 -2555.30 -1121.49 -2252.11 -2507.95 -1534.01 -1108.51 -2206.51 -2499.55 -1500.37 -2773.54
BIC -2182.13 -2536.66 -1105.95 -2234.55 -2489.31 -1517.48 -1092.96 -2188.92 -2480.91 -1483.85 -2754.9
F-test 697.09 13.76 26.83 41.17 6 42.77 16.83 10.59 4.53 46.77 54.87
Prob > F 0 0.0001 0.0001 0 0.0047 0 0 0.0005 0.0147 0 0
Number of
Observations 600 780 360 600 780 460 360 600 780 460 780
Model types: Difference-in-Difference; Weighted by Adjusted Baseline Population in Neighborhood; Standard errors clustered by treatment-
control pair
Notes: We exclude rail corridors whose key years (opening, announcement, or 5 years after) do not fall between 1993-2013 where FTB data is
available. This includes Blue Line for announcement and opening, Green line for announcement, and Expo line for 5 years after opening.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
80

Table 15. FE Model Rail Station Effects on Neighborhood Out Mobility Rates by Timing and Rail Corridor

Rail Station Announcement Rail Station Opening 5 Years After Opening
Rail Corridor Red/
Purple
Gold Expo
Phase 1
Red/
Purple
Gold Green Expo
Phase 1
Red/
Purple
Gold Green Blue
Treatment *
Post
-0.010 0.001 0.010** 0.011* 0.001 -0.023 0.006 0.010*** 0.011* -0.017* 0.035**
(0.011) (0.004) (0.004) (0.005) (0.003) (0.016) (0.007) (0.003) (0.006) (-0.008) (0.014)
Constant
(baseline
mobility rate)
0.317*** 0.279*** 0.289*** 0.311*** 0.279*** 0.305*** 0.289*** 0.312*** 0.279*** 0.305*** 0.302***
(0.007) (0.008) (0.009) (0.006) (0.008) (0.007) (0.009) (0.006) (0.007) (0.008 (0.008)
Year fixed
effect
Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Neighborhood
fixed effect
Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adjusted R
2
0.702 0.730 0.730 0.705 0.730 0.769 0.728 0.706 0.732 0.770 0.699
AIC -10987.66 -13461 -5984 -11007.5 -13460.9 -7901.9 -5976.3 -11016.7 -13483.9 -7909.16 -12939.1
BIC -10900.91 -13346 -5937 -10920.7 -13346.1 -7835.7 -5928.8 -10929.9 -13369 -7842.95 -12818.2
F-test 0.82 0.09 6.31 3.99 0.07 2.00 0.79 8.76 3.59 3.84 5.80
Prob > F 0.379 0.7678 0.033 0.064 0.789 0.183 0.3981 0.010 0.074 0.074 0.025
Number of
Observations
2400 3120 1440 2400 3120 1840 1440 2400 3120 1840 3119
Model types: Fixed Effects; Weighted by Adjusted Baseline Population in Neighborhood; Standard errors clustered by treatment-control pair
Notes: We exclude rail corridors whose key years (opening, announcement, or 5 years after) do not fall between 1993-2013 where FTB data is
available. This includes Blue Line for announcement and opening, Green line for announcement, and Expo line for 5 years after opening.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
81

Table 16. Transit Corridor Income Category Model of Rail Station Opening on Neighborhood Out
Mobility Rates
Treatment
Variable
Rail Station Opening
Rail Corridor Red/Purple Gold Green Expo Phase I
Model Difference-
in-
Difference
Fixed
Effects
Difference-
in-
Difference
Fixed
Effects
Difference-
in-
Difference
Fixed
Effects
Difference-
in-
Difference
Fixed
Effects
Income =
<30% AMI 0.042** 0.033** 0.036** 0.033** 0.045** 0.046** 0.067** 0.059**
(0.006) (0.003) (0.005) (0.005) (0.008) (0.004) (0.008) (0.005)
Income = 30-
50% AMI 0.038** 0.033** 0.033** 0.031** 0.048** 0.041** 0.057** 0.046**
(0.008) (0.004) (0.005) (0.005) (0.007) (0.004) (0.010) (0.005)
Income = 50-
80% AMI 0.022* 0.025** 0.021** 0.023** 0.023* 0.026** 0.040** 0.029**
(0.008) (0.005) (0.004) (0.004) (0.012) (0.004) (0.010) (0.006)
Treatment *
Income =
<30% AMI 0.012 0.005 0.012 0.028
(0.015) (0.009) (0.035) (0.032)
Treatment *
Income = 30-
50% AMI 0.012 0.004 -0.005 0.014
(0.017) (0.010) (0.033) (0.024)
Treatment *
Income = 50-
80% AMI 0.011 0.018 0.000 0.007
(0.014) (0.011) (0.040) (0.019)
Treatment *
Income =
>80% AMI 0.027* 0.041** -0.031 0.016
(0.013) (0.010) (0.037) (0.021)
Post * Income
= <30% AMI -0.050** -0.015* -0.091** -0.028**
(0.005) (0.006) (0.010) (0.007)
Post * Income
= 30-50%
AMI -0.046** -0.016* -0.097** -0.034**
(0.005) (0.006) (0.009) (0.003)
Post * Income
= 50-80%
AMI -0.031** -0.014* -0.084** -0.035**
(0.005) (0.007) (0.014) (0.006)
Post * Income
= >80% AMI -0.032** -0.021** -0.085** -0.023**
(0.005) (0.006) (0.014) (0.007)

Treatment *
Post * Income
= <30% AMI 0.006 0.003 0.001 -0.009* -0.026 -0.017 0.008 0.013*
(0.010) (0.006) (0.006) (0.005) (0.019) (0.017) (0.009) (0.007)
Treatment *
Post * Income
= 30-50%
AMI 0.003 0.002 0.008 -0.005 -0.016 -0.023 0.000 -0.006
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
82

(0.011) (0.005) (0.005) (0.003) (0.020) (0.018) (0.008) (0.006)
Treatment *
Post * Income
= 50-80%
AMI 0.012 0.014* 0.011* 0.003 -0.032 -0.028 0.005 -0.003
(0.008) (0.005) (0.006) (0.004) (0.024) (0.017) (0.008) (0.008)
Treatment *
Post * Income
= >80% AMI 0.023* 0.038** 0.020* 0.015* -0.009 -0.024 0.006 0.016
(0.009) (0.007) (0.011) (0.007) (0.020) (0.017) (0.009) (0.011)
Constant
(baseline
mobility rate) 0.232** 0.286** 0.173** 0.256** 0.274** 0.278** 0.145** 0.251**
(0.003) (0.008) (0.009) (0.009) (0.022) (0.006) (0.012) (0.011)
Year fixed
effect
No Yes No Yes No Yes No Yes
Neighborhood
fixed effect
No Yes No Yes No Yes No Yes
Adjusted R
2
0.202 0.749 0.087 0.773 0.354 0.866 0.241 0.857
AIC -8588.6 -11407 -9623.1 -14012 -5982.3 -8904.9 -4474.2 -6908.7
BIC -8501.8 -11320 -9526.3 -13897 -5916.1 -8838.7 -4426.7 -6861.3
F-test 29.22 192.71 11.24 46.4 302.71 124.15
Prob > F 0 0 0 0 0 0
Number of
Observations 2400 2400 3120 3120 1840 1840 1440 1440
Potential
Degrees of
Freedom
Issue
Yes Yes Yes
Model types: Difference-in-Difference and Fixed Effects; Weighted by Adjusted Baseline Population in
Neighborhood; Standard errors clustered by treatment-control pair
Reference Category: Control Neighborhoods with Income above 80% in Years where Rail Stations Were
not Open in Paired Treatment Neighborhoods
Notes: Blue Line excluded from this analysis, since it opened in 1990, prior to the earliest year of the FTB
dataset available.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Table 17. Transit Corridor Income Category Model of Rail Station Announcement on Neighborhood Out
Mobility Rates
Treatment Variable Rail Station Announcement
Rail Corridor Red/Purple Gold Expo Phase I
Model Difference-
in-
Difference
Fixed
Effects
Difference-
in-
Difference
Fixed Effects Difference-
in-
Difference
Fixed
Effects
Income = <30%
AMI 0.030** 0.033** 0.035** 0.037** 0.068** 0.059**
(0.002) (0.003) (0.005) (0.005) (0.008) (0.005)
Income = 30-50%
AMI 0.031** 0.032** 0.033** 0.035** 0.057** 0.048**
(0.009) (0.005) (0.005) (0.005) (0.009) (0.006)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
83

Income = 50-80%
AMI 0.026** 0.027** 0.021** 0.026** 0.039** 0.030**
(0.003) (0.006) (0.004) (0.004) (0.010) (0.007)
Treatment * Income
= <30% AMI 0.029** 0.009 0.024
(0.003) (0.010) (0.031)
Treatment * Income
= 30-50% AMI 0.014 0.006 0.012
(0.015) (0.011) (0.025)
Treatment * Income
= 50-80% AMI 0.012* 0.018 0.003
(0.005) (0.011) (0.020)
Treatment * Income
= >80% AMI 0.045** 0.040** 0.013
(0.006) (0.010) (0.021)
Post * Income =
<30% AMI -0.098** -0.028** -0.029**
(0.005) (0.005) (0.003)
Post * Income = 30-
50% AMI -0.101** -0.028** -0.026**
(0.014) (0.006) (0.004)
Post * Income = 50-
80% AMI -0.100** -0.028** -0.026**
(0.007) (0.006) (0.005)
Post * Income =
>80% AMI -0.098** -0.032** -0.024**
(0.006) (0.006) (0.004)
Treatment * Post *
Income = <30%
AMI -0.015 -0.018 -0.005 -0.011* 0.013* 0.014*
(0.011) (0.011) (0.006) (0.005) (0.005) (0.006)
Treatment * Post *
Income = 30-50%
AMI -0.002 -0.019 0.003 -0.004 0.005 0.003
(0.020) (0.011) (0.005) (0.004) (0.006) (0.005)
Treatment * Post *
Income = 50-80%
AMI 0.008 -0.013 0.007 0.005 0.011 0.008*
(0.013) (0.011) (0.006) (0.004) (0.007) (0.004)
Treatment * Post *
Income = >80%
AMI 0.000 0.012 0.016* 0.019** 0.007 0.015
(0.013) (0.011) (0.009) (0.006) (0.008) (0.009)
Constant (baseline
mobility rate) 0.303** 0.293** 0.183** 0.254** 0.154** 0.251**
(0.003) (0.009) (0.009) (0.009) (0.012) (0.011)
Year fixed effect No Yes No Yes No Yes
Neighborhood fixed
effect
No Yes No Yes No Yes
Adjusted R
2
0.132 0.741 0.133 0.778 0.264 0.859
AIC -8397.4 -11325.7 -9785.2 -14080.2 -4519.3 -6931.2
BIC -8339.5 -11239.0 -9688.5 -13965.3 -4471.9 -6883.8
F-test 415.56 28.93 41.48 13.11 72.54 89.6
Prob > F 0 0 0 0 0 0
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
84

Number of
Observations 2400 2400 3120 3120 1440 1440
Potential Degrees of
Freedom Issue
Yes Yes
Model types: Difference-in-Difference and Fixed Effects; Weighted by Adjusted Baseline Population in
Neighborhood; Standard errors clustered by treatment-control pair
Reference Category: Control Neighborhoods with Income above 80% in Years where Rail Stations Were
not Announced in Paired Treatment Neighborhoods
Notes: Blue Line excluded from this analysis, since it opened in 1990, prior to the earliest year of the FTB
dataset available. We exclude Green Line from the announcement analysis, since the Green Line was
announced in 1991, prior to FTB data availability.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Table 18. Transit Corridor Income Category Model of Rail Stations Five Years after Opening on
Neighborhood Out Mobility Rates
Treatment
Variable
5 Years after Rail Station Opening
Rail
Corridor
Red/Purple Gold Green Blue
Model Difference
-in-
Difference
Fixed
Effects
Difference
-in-
Difference
Fixed
Effects
Difference
-in-
Difference
Fixed
Effects
Difference
-in-
Difference
Fixed
Effects
Income =
<30% AMI 0.038***
0.030**
* 0.040***
0.029**
* 0.042***
0.047**
* 0.079***
0.060**
*
(0.006) (0.003) (0.005) (0.005) (0.004) (0.005) (0.016) (0.003)
Income = 30-
50% AMI 0.033***
0.029**
* 0.037***
0.029**
* 0.040***
0.042**
* 0.064***
0.051**
*
(0.007) (0.003) (0.005) (0.005) (0.006) (0.004) (0.014) (0.004)
Income = 50-
80% AMI 0.022***
0.022**
* 0.024***
0.022**
* 0.021**
0.026**
* 0.033***
0.035**
*
(0.007) (0.004) (0.004) (0.004) (0.007) (0.004) (0.010) (0.004)
Treatment *
Income =
<30% AMI 0.011 0.004 0.001 -0.003
(0.012) (0.009) (0.031) (0.019)
Treatment *
Income = 30-
50% AMI 0.012 0.004 -0.015 0.013
(0.015) (0.010) (0.029) (0.017)
Treatment *
Income = 50-
80% AMI 0.013 0.019* -0.019 0.021
(0.014) (0.010) (0.034) (0.015)
Treatment *
Income =
>80% AMI 0.032*** 0.045*** -0.042 0.031**
(0.011) (0.010) (0.035) (0.011)
Post *
Income =
<30% AMI -0.036*** -0.003 -0.053*** -0.103***
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
85

(0.006) (0.008) (0.004) (0.014)
Post *
Income = 30-
50% AMI -0.029*** -0.005 -0.055*** -0.096***
(0.008) (0.009) (0.005) (0.013)
Post *
Income = 50-
80% AMI -0.016** 0.001 -0.044*** -0.076***
(0.006) (0.009) (0.008) (0.011)
Post *
Income =
>80% AMI -0.016*** -0.007 -0.049*** -0.070***
(0.005) (0.007) (0.008) (0.010)
Treatment *
Post *
Income =
<30% AMI 0.008 0.001 0.011 -0.002 -0.018 -0.015 0.040** 0.026*
(0.008) (0.003) (0.008) (0.004) (0.012) (0.009) (0.016) (0.015)
Treatment *
Post *
Income = 30-
50% AMI 0.003 -0.000 0.021** 0.004 -0.005 -0.016 0.026* 0.027*
(0.010) (0.004) (0.009) (0.006) (0.012) (0.009) (0.014) (0.015)
Treatment *
Post *
Income = 50-
80% AMI 0.012
0.016**
* 0.021** 0.015 -0.014 -0.019* 0.031** 0.040**
(0.008) (0.005) (0.010) (0.009) (0.016) (0.009) (0.014) (0.014)
Treatment *
Post *
Income =
>80% AMI 0.020***
0.038**
* 0.026** 0.025** 0.003 -0.014 0.036***
0.055**
*
(0.007) (0.006) (0.012) (0.011) (0.016) (0.009) (0.011) (0.014)
Constant
(baseline
mobility rate) 0.217***
0.289**
* 0.164***
0.259**
* 0.231***
0.277**
* 0.232***
0.262**
*
(0.004) (0.007) (0.008) (0.009) (0.018) (0.007) (0.009) (0.009)
Year fixed
effect
No Yes No Yes No Yes No Yes
Neighborhoo
d fixed effect
No Yes No Yes No Yes No Yes
Adjusted R
2
0.153 0.752 0.080 0.773 0.306 0.866 0.342 0.788
AIC -8445.6 -11429 -9600.7 -14009 -5852.0 -8907.9 -10464 -14043
BIC -8358.8 -11342 -9504.0 -13894 -5785.8 -8841.6 -10367 -13916
F-test 31.58 29.33 80.42 18.13 31.53 21.6 116.71 80.24
Prob > F 0 0 0 0 0 0 0 0
Number of
Observations 2400 2400 3120 3120 1840 1840 3119 3119
Model types: Difference-in-Difference and Fixed Effects; Weighted by Adjusted Baseline Population in
Neighborhood; Standard errors clustered by treatment-control pair
Reference Category: Control Neighborhoods with Income above 80% in Years where Rail Stations Were
not Open in Paired Treatment Neighborhoods
Notes: Expo Phase I excluded because it opened in 2012, with five years after in 2016, but our data only
goes through 2013
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
86

* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

1-G. Regression Results for Model Extensions

Table 19. Downtown Exclusion: Rail Effect on Neighborhood Out Mobility
Treatment
Variable
Rail Station
Announcement
Rail Station Opening 5 Years after Rail Station
Opened
Model Difference-
in-
Difference
Fixed
Effects
Difference-
in-
Difference
Fixed
Effects
Difference-
in-Difference
Fixed
Effects
Treatment
0.014 0.009 0.008

(0.011) (0.010) (0.008)
Post
0.000 -0.010* -0.016***

(0.008) (0.006) (0.005)
Treatment * Post
0.001 0.002 0.008 0.005 0.012 0.006*

(0.012) (0.004) (0.009) (0.003) (0.007) (0.003)
Constant (baseline
mobility rate) 0.208** 0.299** 0.215** 0.299** 0.216*** 0.299***

(0.006) (0.004) (0.007) (0.004) (0.006) (0.004)
Year fixed effect No Yes No Yes No Yes
Neighborhood
fixed effect
No Yes No Yes No Yes
Adjusted R
2

0.020 0.883 0.025 0.883 0.035 0.884
AIC
-8430.94 -14249.810 -8444.51 -14259.818 -8472.6 -14268.1
BIC
-8407.37 -14131.938 -8420.94 -14141.947 -8449 -14150.2
F-test
47.07 1.71 46.01 2.04 4.72 46.46
Prob > F
0 0.1722 0 0.1163 0.0047 0
Number of
Observations 2680 2680 2680 2680 2680 2680
Model types: Difference-in-Difference and Fixed Effects; Weighted by Adjusted Baseline Population in
Neighborhood; Standard errors clustered by treatment-control pair
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Table 20. Overlapping Station & Downtown Exclusion: Rail Effect on Neighborhood Out Mobility
Treatment
Variable
Rail Station
Announcement
Rail Station Opening 5 Years after Rail Station
Opened
Model Difference-
in-
Difference
Fixed
Effects
Difference-
in-
Difference
Fixed
Effects
Difference-
in-Difference
Fixed
Effects
Treatment
0.020 0.013 0.011
(0.012) (0.011) (0.010)
Post
0.003 -0.007 -0.015**

(0.009) (0.007) (0.006)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
87

Treatment * Post -0.006 0.004 0.003 0.004 0.006 0.005

(0.014) (0.004) (0.011) (0.004) (0.009) (0.004)
Constant (baseline
mobility rate) 0.204** 0.299** 0.212** 0.300** 0.215*** 0.300***

(0.007) (0.004) (0.008) (0.004) (0.008) (0.004)
Year fixed effect No Yes No Yes No Yes
Neighborhood
fixed effect
No Yes No Yes No Yes
Adjusted R
2

0.020 0.883 0.022 0.884 0.034 0.884
AIC
-6686.95 -11382.742 -6692.38 -11383.741 -6718.6 -11386.3
BIC
-6664.24 -11269.185 -6669.67 -11270.183 -6695.9 -11272.7
F-test
1.75 41.09 1.21 39 2.96 37.74
Prob > F
0.168 0 0.3135 0 0.0404 0
Number of
Observations 2160 2160 2160 2160 2160 2160
Model types: Difference-in-Difference and Fixed Effects; Weighted by Adjusted Baseline Population in
Neighborhood; Standard errors clustered by treatment-control pair
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Table 21. Subway versus Light Rail Comparison: Rail Effect on Neighborhood Out Mobility
Treatment Variable Rail Station
Announcement
Rail Station Opening 5 Years after Rail Station
Opened
Model Difference-
in-
Difference
Fixed
Effects
Difference-
in-
Difference
Fixed
Effects
Difference-
in-Difference
Fixed
Effects
Treatment
0.016 0.009 0.008

(0.011) (0.012) (0.011)
Post
-0.003 -0.003 -0.012*

(0.010) (0.007) (0.006)
Treatment * Post
0.007 -0.001 0.018 -0.003 0.025** 0.007

(0.014) (0.004) (0.013) (0.004) (0.010) (0.006)
Subway
0.117** 0.055** 0.033***

(0.008) (0.008) (0.009)
Subway *
Treatment 0.014 0.009 0.010

(0.011) (0.018) (0.016)
Subway * Post
-0.092** -0.038** -0.014*

(0.012) (0.008) (0.008)
Subway *
Treatment * Post -0.015 -0.001 -0.011 0.015* -0.017 0.006

(0.018) (0.011) (0.015) (0.006) (0.013) (0.007)
Constant (baseline
mobility rate) 0.205** 0.300** 0.204** 0.299** 0.208*** 0.299***

(0.006) (0.004) (0.007) (0.004) (0.008) (0.004)
Year fixed effect No Yes No Yes No Yes
Neighborhood fixed
effect
No Yes No Yes No Yes
Adjusted R
2

0.107 0.872 0.126 0.873 0.119 0.874
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
88

AIC
-9700 -16000 -9600 -16000 -9638 -15573.2
BIC
-9600 -15000 -9500 -15000 -9590 -15447.2
F-test
465.4 37.64 25.95 40.47 9.28 40.35
Prob > F
0 0 0 0 0 0
Number of
Observations 2980 2980 2980 2980 2980 2980
Model types: Difference-in-Difference and Fixed Effects; Weighted by Adjusted Baseline Population in
Neighborhood; Standard errors clustered by treatment-control pair
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Table 22. Highway Median versus In-the-Neighborhood: Rail Effect on Neighborhood Out Mobility
Treatment
Variable
Rail Station
Announcement
Rail Station Opening 5 Years after Rail Station
Opened
Model Difference-
in-
Difference
Fixed
Effects
Difference-
in-
Difference
Fixed
Effects
Difference-
in-Difference
Fixed
Effects
Treatment
0.015 0.014 0.016**

(0.011) (0.009) (0.008)
Post
-0.001 -0.004 -0.010

(0.007) (0.007) (0.006)
Treatment * Post
0.016 -0.001 0.021* 0.004 0.024*** 0.014***

(0.012) (0.004) (0.010) (0.004) (0.008) (0.003)
Highway Median
0.016 0.057* 0.034*

(0.016) (0.026) (0.018)
Highway Median *
Treatment 0.006 -0.008 -0.019

(0.027) (0.038) (0.029)
Highway Median *
Post -0.002 -0.050* -0.030***

(0.023) (0.019) (0.011)
Highway Median *
Treatment * Post -0.052 -0.003 -0.041* -0.020 -0.038** -0.030***

(0.034) (0.006) (0.023) (0.013) (0.015) (0.008)
Constant (baseline
mobility rate) 0.206** 0.300** 0.208** 0.299** 0.210*** 0.299***

(0.007) (0.004) (0.007) (0.004) (0.006) (0.004)
Year fixed effect No Yes No Yes No Yes
Neighborhood
fixed effect
No Yes No Yes No Yes
Adjusted R
2

0.072 0.872 0.104 0.872 0.106 0.877
AIC
-9483.01 -15518.279 -9587.94 -15534.966 -9593.1 -15637.6
BIC
-9435.01 -15392.286 -9539.54 -15408.972 -9545.1 -15511.6
F-test
3.68 39.75 7.52 36.54 15.01 39.36
Prob > F
0.0017 0 0 0 0 0
Number of
Observations 2980 2980 2980 2980 2980 2980
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
89

Model types: Difference-in-Difference and Fixed Effects; Weighted by Adjusted Baseline Population in
Neighborhood; Standard errors clustered by treatment-control pair
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
90

Table 23. Transit Network Variables: Rail Effect on Neighborhood Out Mobility
Rail Station Announcement Rail Station Opening 5 Years after Rail Station Opened
Variable of
Interest
System
Size
Transfer
Station
Stops to
CBD
All System
Size
Transfer
Station
Stops to
CBD
All System
Size
Transfer
Station
Stops to
CBD
All
Treatment 0.013 0.014 0.014 0.013 0.010 0.009 0.010 0.011 0.012* 0.011 0.012 0.013*
(0.011) (0.011) (0.011) (0.011) (0.009) (0.010) (0.010) (0.009) (0.007) (0.008) (0.008) (0.007)
Post 0.021* 0.002 -0.002 0.017** 0.013* -0.008 -0.012* 0.011* 0.011* -0.015*** -0.018*** 0.009
(0.008) (0.008) (0.008) (0.007) (0.006) (0.006) (0.006) (0.006) (0.006) (0.005) (0.005) (0.007)
Treatment *
Post 0.012 0.010 0.010 0.012 0.018* 0.018* 0.018* 0.017* 0.020*** 0.021*** 0.020*** 0.019***
(0.012) (0.012) (0.012) (0.011) (0.009) (0.009) (0.009) (0.009) (0.007) (0.007) (0.007) (0.007)
System Size
-0.00014**
-
0.00014*** -0.00017**

-0.00016*** -0.00014***
-
0.00013***
(0.000)

(0.000) (0.000)

(0.000) (0.000)

(0.000)
Number of
Stops to
Nearest
Transfer 0.001

-0.000 0.001

0.000 0.000 0.000
(0.001)

(0.002) (0.001)

(0.002) (0.001) (0.002)
Number of
Stops to
Downtown
Los Angeles 0.00124* 0.001 0.0013* 0.001 0.00135** 0.001
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Constant
(baseline
mobility rate) 0.203** 0.205** 0.199** 0.274*** 0.286** 0.282** 0.211** 0.278*** 0.255*** 0.214*** 0.206*** 0.247***
(0.008) (0.008) (0.007) (0.009) (0.007) (0.008) (0.009) (0.009) (0.006) (0.008) (0.007) (0.007)
Adjusted R
2
0.053 0.072 0.067 0.182 0.226 0.172 0.056 0.234 0.233 0.061 0.080 0.240
AIC 9424 9486 9469 9859.1 10025 9824.17 9434 10054 10055 9451 9511 10077
BIC 9394 9456 9439 9817.1 9995 9794 9404 10012 10025 9421 9481 10035
F-test 3.84 6.57 5.87 43.67 88.97 51.94 3.99 62.94 107.11 7.04 9.41 73.12
Prob > F 0.0066 0.000 0.00 0 0 0 0.005 0 0 0.000 0 0
Number of
Observations 2980 2980 2980 2980 2980 2980 2980 2980 2980 2980 2980 2980
Model types: Difference-in-Difference; Weighted by Adjusted Baseline Population in Neighborhood; Standard errors clustered by treatment-
control pair
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
91

Chapter 2: Do Rail Transit Station Openings Displace Low-Income
Households?

Author: Seva Rodnyansky

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
92

Abstract
Governments undertake local and regional infrastructure investments to improve the lives of
their constituents. Transportation agencies build public transit systems, for example, to improve
residents’ access to jobs and amenities, within their neighborhoods and metropolitan areas. These
investments have direct effects, including changes in ridership, service levels, access, travel
times, and other characteristics directly related to its operation. But, they may bring additional
externalities, such as changes in local air quality, noise, property values, neighborhood
composition, and residential mobility rates. There are winners and losers from these indirect
effects. The literature on externalities is extensive. In this paper, I examine whether negative
externalities from new rail transit station openings cause existing low-income residents to move
away at higher than average rates, a process termed residential displacement, using Los Angeles
County as a study area.
Residential mobility can be positive or negative. Yet prior research has shown that too much
mobility can lead to residential instability, which decreases health, educational, and behavioral
outcomes in children and adults. These effects are exacerbated for at-risk populations, such as
low-income households, minorities, households with children, and elderly households. I examine
whether these sub-populations are displaced by new rail station openings.
To do so, I use an innovative dataset from the California Franchise Tax Board that provides
location and demographic data for the approximately four million state income tax filers residing
in Los Angeles County from 1993 to 2013. I construct a measure of mobility from these 140
million records and test whether the opening of a nearby rail station increases a household’s
probability to move. I perform the analyses along two rail transit lines in Los Angeles County
which opened during the time period.
The evidence from neighborhoods surrounding two of Los Angeles County Metropolitan Transit
Authority’s (L.A. Metro) rail corridors suggests that a household’s probability of moving
increases slightly when rail stations open nearby. However, the evidence does not strongly
suggest that move probabilities increase for low-income households, households with children,
elderly households, or young households. Compared to similar control households, move
probability appears to increase for middle and higher-income households (with incomes above
80% of AMI) along both Red/Purple and Gold Lines. Relative to baseline mobility rates, higher-
income households increase move probability by 5-6% after stations open. The evidence does
not suggest that move rates for low income and lower-middle income households (with incomes
below 30% of AMI and between 30-80% of AMI, respectively) living near the Red/Purple Line
are significantly different from control households not living near rail. The effect on lower-
income households is less clear for the Gold Line: specifications disagree on the direction and
magnitude of the effect. Post-rail station opening move probability also increases for households
without dependents by 4-7% relative to baseline, with no effect for households with dependents.
The overall results suggest that building new rail stations does affect a nearby household’s
likelihood to move, but that the effect magnitude is not very high relative to baseline move
likelihood. This affect does appear to be larger for lower-income households, households with
children, young households, or elderly households, all groups who would be considered at-risk of
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
93

perils stemming from housing instability. This indicates that lower-income households may be
adopting to higher housing costs in other ways, including doubling up, selling assets, consuming
less other goods, or relying on assistance or donations. These research findings suggest that these
methods of coping deserve greater future research attention.

Introduction
Government actors undertake infrastructure projects to serve the public good. Policymakers and
planners at all levels of government, for example, provide public transit systems, to improve
residents’ access to jobs and amenities, within their neighborhoods and metropolitan areas.
Public transportation infrastructure investments have direct effects, including changes in
ridership, service levels, access, travel times, and other characteristics directly related to its
operation. However, new public transit systems can also have positive or negative indirect
effects or externalities, including changes in local air quality, noise, property values,
neighborhood composition, and residential mobility rates. There are winners and losers from
these indirect effects. This paper examines whether negative externalities from new rail transit
station openings cause existing low-income residents to move away at higher than average rates,
a process termed residential displacement, using Los Angeles County as a study area.
Why would new rail station openings displace existing low-income households? I hypothesize
that displacement would occur through a process of changing neighborhood composition coupled
with rising housing costs, together termed gentrification. In this process, the new transportation
amenity increases access to and from a neighborhood, which increases local housing demand,
leading to population in-movement. Absent a commensurate increase in housing supply (for
which there is ample evidence in Los Angeles County: see California Legislative Analyst’s
Office (LAO), 2015; Los Angeles Department of City Planning (LADCP), 2013; Collinson,
2011), housing prices increase in the area and previously residing households can no longer
afford the cost of housing. This is particularly salient for low-income households who already
face higher housing cost burdens. These residents are faced with several options: paying more
10
,
doubling-up with others, moving to smaller and thus cheaper units if available, moving out of the
neighborhood, or becoming temporarily or permanently homeless. While each of these options
deserves careful examination, this paper focuses on the moving option.
Understanding these effects is crucial from both the household and public agency perspectives.
From the household perspective, displacement-related moves may unintentionally break existing
ties to other people and institutions in the neighborhood. Moving may increase commute times,
especially for households reliant on public transit, if they move farther away from transit. For
families with children, instability created by a displacement-related move may negatively affect

10
Households may choose to pay more to stay in their current neighborhood. For neighborhoods with newly
improved public transit options, this option may be particularly salient for lower-income, minority, and foreign-born
populations, who tend to use transit more (Boarnet et al., 2015). This mitigating strategy could constrain
neighborhood displacement.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
94

children’s educational performance (Goldsmith, Britton, Reese, & Velez, 2017), and health and
behavioral outcomes (Jelleyman & Spencer, 2008). For public agencies and urban planners, a
disproportionate loss of low-income households may reduce ridership rates of the new system,
since low-income households tend to use public transit more than high-income households
(Santos, McGuckin, Nakamoto, Gray, & Liss, 2011). Moreover, displacement may impact local
businesses if existing residents move away and are not replaced with households with similar
tastes or preferences.
Despite these concerns, evidence for displacement has been uneven at best (Rayle, 2015). On
one side, advocacy groups have challenged plans for new public transit and the development of
residential and commercial structures in neighborhoods immediately surrounding the stations,
termed transit-oriented developments (TOD), on the basis of potential displacement (Pollack et
al., 2010; Pendall, Gainsborough, Lowe, & Nguyen, 2012; Young, 2009; Puget Sound Sage,
2012). However, in academic studies, quantitative evidence on the effects of rail station openings
on residential mobility has been scant and contains measurement issues (Zuk, Bierbaum, Chapple,
Gorska, & Loukaitou-Sideris, 2017; Zuk et al., 2015). Prior studies looking at rail transit and
neighborhood change have focused on gentrification, not displacement (e.g., Kahn, 2007).
Studies focusing on displacement have not focused on rail (e.g., Freeman, 2005). Additionally,
methodological constraints stemming from inadequate data, unclear measurement, and a lack of
baseline to compare the impact of effects constrain the results of previous studies.
In contrast to previous work, I directly measure the effect of rail transit station openings on
household mobility and innovate on prior methodological shortcomings. To do so, I use a unique
administrative dataset of over 100 million tax file records in Los Angeles County from 1993-
2013 from the California Franchise Tax Board (FTB). I geocode tax filing locations and measure
mobility using change in filing location. With this data, I answer the following research
questions:
• To what extent does the opening of a nearby rail-station affect a household’s probability
of moving out of their neighborhood?
o Are low-income households more impacted than high-income households?
o Are households with children or elderly dependents more impacted?
o Are households with young heads of households more impacted?

To assess these questions, I set up a quasi-experimental research design, where I compare
mobility rates for households in rail-proximate neighborhoods to similar households in non-rail-
proximate areas, before and after rail stations open in Los Angeles County. I control for other
mobility determinants available in the dataset, including income, age, marital status, and number
of dependents, as well as annual changes in these variables. Los Angeles County is an ideal
laboratory for this study, due to its significant investment in new rail transit over the past 30
years and its housing affordability crisis marked by high and rising housing cost burdens for
renters (Collinson, 2011) and homeowners (LADCP, 2013) underpinned by an undersupply of
housing to demand (LAO, 2015).

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
95

This research develops a unique dataset particularly well-suited to answer germane research
questions on the relationships between public investment, household mobility, and neighborhood
change. Analyses using this dataset can contribute to data-driven decision-making and policy
development in improving outcomes for households and neighborhoods.
The remainder of the paper proceeds as follows. First, I situate this research in the prior
literature, which informs theory and methodology. Next, I describe the study area and dataset.
Then, I present the relevant models, and in the final section I discuss the results.

Literature Review

This paper tests whether households move out of their neighborhood at higher rates after a rail
station opened than they had before. In considering this question, I examine the literature on why
moving might be detrimental for households and then on why households move in general. Then,
I propose a mechanism for why households, especially low-income ones, may move at higher
rates after rail stations open, and consider the relevant literature. I also examine past empirical
studies looking at residential displacement.

What’s the Problem with Moving?

Households change residences in all parts of the world and moving is a prominent feature of
urban life (Jelleyman & Spencer, 2008). Residential mobility is not a priori good or bad. Moves
can be ‘upward’ and associated with an “advancement of lifestyle through improvements to
housing, neighborhoods, or employment” (Morris, Manley, & Sabel, 2018). However too many
or too frequent moves can be detrimental, leading to residential instability. While there are
multiple definitions for what counts as residential instability – including too many moves or
addresses reported over a certain period of time, eviction or foreclosure, doubling up or living
with others – each of them may represent a form of housing insecurity (Cox, Henwood,
Rodnyansky, Wenzel, & Rice, 2017). A large literature has found significant adverse effects on
health and educational outcomes from residential instability (Morris et al., 2018; Jelleyman &
Spencer, 2008; Goldsmith et al., 2017).

Regarding health outcomes, Jelleyman and Spencer (2008) found that childhood residential
mobility was associated with behavioral and emotional problems, including “behavioral
disturbance, poorer emotional adjustment, increased teenage pregnancy rates, earlier illicit drug
use, drug-related problems, and teenage depression” in a review of 22 studies on health and
residential mobility (Jelleyman & Spencer, 2008, p.590). Residential stability as a child was
found to be strongly positively associated with self-rated health as an adult, due to difficulties of
forming new relationships and lower investment in social networks (Bures, 2003).

Behaviorally, residential instability for adolescents (age 16-19) was positively associated with
crime and delinquency (Sampson & Groves, 1989) and for early adolescents (age 10-15) with
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
96

lifetime alcohol use (Ennett, Flewelling, Lindrooth, & Norton, 1997). In children from poor
households, frequent moves have been associated with poor behavioral outcomes like attention
problems and externalizing behaviors, especially for households who moved three or more times
within a five-year period (Ziol-Guest & McKenna, 2014). Residential instability has also been
found to deter development of children’s executive function and processes that facilitate goal-
directed behaviors (Schmitt, Finders, & McClelland, 2015).

Residential instability appears to negatively affect children’s and adolescent’s educational
outcomes. “Children and adolescents who move frequently perform worse on measures of math,
reading, and educational expectations” (Schmitt et al., 2015, p.191). Frequent mobility
compounds risk for poor educational outcomes for children in poverty (Cutuli et al., 2013).
Living in high residential mobility neighborhoods has also been associated with lower
educational outcomes and less time spent doing homework, especially for low-income
adolescents (Ainsworth, 2002). Residential mobility has also been found to disrupt daily routines
and social networks in children (Evans & Wachs, 2010), reducing educational outcomes in the
future. Additionally, individuals who live in a neighborhood for less time have been found less
likely to engage in local civic participation (Kang & Kwak, 2003).

Research has been mixed on whether the impact of residential mobility depends on destination
quality. According to a large literature review by Leventhal and Brooks-Gunn (2000), the role of
neighborhoods accounted for 5-10% of variance in outcomes for children and adolescents. Roy,
McCoy, & Raver (2014) found that moves during early or middle childhood were associated
with increased behavioral and cognitive dysregulation in 5
th
grade. When conditioning for
neighborhood poverty, the effects of moving were found to be detrimental only for moves from
low-poverty to high-poverty neighborhoods, but moves from high-poverty to low-poverty
neighborhoods were found to be beneficial (Roy et al., 2014). In contrast, Metzger, Fowler,
Anderson and Lindsay (2015) found that frequent movers among adolescents (age 16-19) are
50% less likely to earn a high-school degree by age 25, regardless of whether they move to a
poor or rich neighborhood. Similarly, Goldsmith et al. (2017) found that adolescent moves to
higher-income neighborhoods did not significantly change educational attainment outcomes
(graduating high school, enrolling in college, or completing a bachelor’s degree). Moves within
larger changes in neighborhood income had more negative effects on educational attainment than
moves with smaller changes in neighborhood income. By race, adolescent moves to higher-
income neighborhoods were associated with higher high school completion and college
enrollment for Latinos (compared to Whites) and lower for African Americans (compared to
Whites) (Goldsmith et al., 2017).

In addition, policymakers may be particularly interested in understanding displacement effects
on households headed by older or younger household heads. Elderly households may be on fixed
incomes and may not be able to easily adjust to rising housing costs, making them more
vulnerable to displacement (Clark & Davies, 1990). Moving has been found to be more difficult
and detrimental to health at advanced ages, especially for elderly with existing health issues
(Chen & Wilmoth, 2004). Young households, who may have recently formed and are just
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
97

starting out, may not be able to cope with increased housing costs either. The higher likelihood
of renting rather than owning housing among young households also makes them more prone to
moving than middle-aged or older households.
The above literature summary shows the potential perils of moving, and specifically frequent
moving, and identifies the groups most at risk from residential instability. Next, I examine why
households would move, given the health and welfare risks of residential mobility.

Why Do Households Move?
Scholars from a multitude of social science traditions have theorized on why households move
within a city since at least the middle of the 20
th
century. Understanding the determinants of
mobility drew scholars initially, because excess mobility was seen as linked to social problems
and as an inherent pathology by Chicago school sociologists and because of a governmental
belief in ‘upward mobility’ to improve neighborhoods (Rossi & Shlay, 1982).

Multiple theories have been proposed as to what spurs households to move: lifecycle events that
lead to changes in demand for housing space consumed (Rossi, 1955; Sabagh, van Arsdol, &
Butler 1969), changes in employment (Brown, 1975), stress from dissatisfaction with current
housing unit or neighborhood (Speare, 1974; Sabagh et al., 1969), preference for a different level
of local amenities and public goods (Tiebout, 1956) or social and/or civic participation (Sabagh
et al., 1969), general economic cost-benefit calculation of moving versus staying (Quigley &
Weinberg, 1977; Lee, 1966; Fredland, 1974), non-optimal equilibrium between commute costs
and desired bundles of housing services (Alonso, 1964; Muth, 1969; Kain, 1968), and
technological changes enabling different communication means and transportation modes
(Sheller & Urry, 2006). These theories explored reasons at multiple levels of analysis, ranging
from individuals and households, to neighborhoods, to metropolitan area. Each of these theories
about residential mobility are constrained by necessary conditions relating to the actual moving
process, such as the search for new housing, the availability of residences of the type sought by
the household, and the availability of resources to move (Sabagh et al., 1969; Quigley &
Weinberg, 1977).

Empirical studies in the past few decades have confirmed that each of these considerations do in
fact increase household move likelihoods. Mobility rates have been shown to increase due to
life-cycle changes including changes in marital status, number of dependents, and retirement
(Clark & Davies Withers, 1999; van der Vliest, Gorter, Nijkamp, & Rietveld, 2002; Ioannides &
Kan, 1996; Clark & Huang, 2003; Weinberg, 1979), job changes (Clark & Davies Withers, 1999;
van der Vliest et al., 2002; Ioannides & Kan, 1996; Weinberg, 1979)
11
, dissatisfaction with
current housing unit (van der Vliest et al., 2002)
12
, and increases in commuting distance (van
Ommeren, Rietveld, & Nijkamp, 1999). Additionally, mobility was found to decrease via several
macro-level variables: high real interest rates (Ioannides & Kan, 1996) and housing market

11
But not van Ommeren et al. (2000) who did not find that job changes cause residential moves unless commuting
distances increase due to the job change
12
Though Varady (1983) did not find dissatisfaction with housing to be associated with increased mobility
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
98

tightness (Weinberg, 1979; Varady, 1983). Moving transaction costs were not found to unduly
deter mobility (Ioannides & Kan, 1996).

In contrast to most of the household-level changes above, the evidence on the effect of
neighborhood-level characteristics on move probabilities has been rather mixed. On one hand,
neighborhood dissatisfaction (Clark & Huang, 2003), perceived neighborhood level of turnover
(Lee, Oropesa, & Kanan, 1994), perceptions of worsening neighborhood quality were found to
increase move probability (Boehm & Ihlanfeldt, 1996). On the other hand, Varady (1983) found
that dissatisfaction with the neighborhood or with local public services did not increase
household move probability, in study using the American Housing Survey. Moreover, in a study
of Nashville households, among neighborhood racial mix, income mix, tenure mix, density,
vacancy rate, mobility rate, population change, and perceptions of neighborhood physical
change, social change, and overall sentiment, none were found to influence household mobility
(Lee et al., 1994), similar to earlier research on Rhode Island households (Speare, 1974).

In addition to variables that increase or decrease the likelihood of moving, empirical studies have
established that mobility baselines vary and that certain groups are less likely to move away from
their neighborhood than others. These include homeowners (Rossi, 1955; South & Crowder,
1998; Ioannides & Kan, 1996; van Ommeren et al., 1999; Clark & Huang, 2003; Lee et al., 1994;
Varady, 1983), married households (Ioannides & Kan, 1996; van Ommeren et al., 1999; Varady,
1983), older households (South & Crowder, 1998; van der Vliest et al., 2002; Ioannides & Kan,
1996; van Ommeren et al., 1999; Clark & Huang, 2003; Lee et al., 1994; Weinberg, 1979;
Varady, 1983), households with children (South & Crowder, 1998; Ioannides & Kan, 1996),
large household size (Clark & Davies Withers, 1999)
13
, households with lower education (van
der Vliest et al., 2002; van Ommeren et al., 1999)
14
, individuals living without roommates and
not sharing kitchen or bath facilities (van Ommeren et al., 1999), households not living in
overcrowded conditions (Clark & Huang, 2003; Varady, 1983)
15
, households living in older
housing units (Varady, 1983), long-term neighborhood dwellers (Lee et al., 1994; Boehm &
Ihlanfeldt, 1996; Varady, 1983), housing living in perceived high quality neighborhoods (Boehm
& Ihlanfeldt, 1996), households living in suburban locations (Varady, 1983). Some studies have
also shown that mobility baselines differ across metropolitan areas and by household location
within a metropolitan area: homeowners moved more in the suburbs and renters more in the city
than metropolitan area averages (van der Vliest et al., 2002).

Despite the potential adverse effects, households move for many reasons. The literature above
has shown that move probabilities change due to a variety of family, economic, housing, and
neighborhood reasons. In my analysis, I add a new neighborhood reason – the opening of a new
rail station nearby – to this canon. Prior studies have found that mobility rates differ for different
groups. Hence, I control for differences between groups to the extent enabled by variables in my

13
Though Weinberg (1979) found that household size is positively correlated with residential mobility
14
Though Varady (1983) found that high education is positively associated with lower mobility probability
15
Computed as households not experiencing room-stress: a mismatch between actual housing space and required
housing space ((actual rooms / required rooms) – 1) (Clark & Huang, 2003, p.328)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
99

dataset. Next, I provide a mechanism for why households might move in response to a new rail
station opening in their neighborhood.

Rail Stations and Displacement

Why would new rail station openings displace existing low-income households? I hypothesize
that displacement would occur through a process of changing neighborhood composition coupled
with rising housing costs, together termed gentrification. Namely, when a new rail station opens
in a neighborhood, households perceive it as a positive amenity due to increased access to and
from a neighborhood. This in turn increases local housing demand, leading to population in-
movement, a fact empirically confirmed for a variety of U.S. metropolitan areas: San Francisco
Bay Area (Chapple, 2006; Chapple, 2009), Chicago (Lin, 2002), Dallas (Heilmann, 2017),
Washington D.C. (Dawkins & Moeckel, 2016), Toronto and Montreal (Grube-Cavers &
Patterson, 2015), and in multi-city analyses (Zuk et al., 2017; Zuk et al., 2015; Kahn, 2007). The
gentrification of rail station neighborhoods follows a similar patterns to gentrifying
neighborhoods in general. Households into gentrifying neighborhoods have been found to be
wealthier, higher educated, and whiter on average (Zuk et al., 2017; Freeman, 2005). However,
certain studies have found mixed results on race and on income depending on metropolitan area
and the degree of residential development in the neighborhood (Baker & Lee, 2017; McKinnish,
Walsh & White, 2008; Kahn, 2007; Ellen & O’Regan, 2010; Grube-Cavers & Patterson, 2015).

This influx of new, often more wealthy residents increases the competition for existing housing
in the neighborhood, pushing up rents and property values, absent a commensurate increase in
housing supply. Scholars have generally confirmed that residential real estate appreciates to
varying degrees after rail stations open in numerous U.S. cities large and small: Phoenix
(Atkinson-Palombo, 2010; Golub, Guhathakurta, & Sollapuram, 2012), Buffalo (Hess and
Almeida, 2007), Atlanta (Immergluck, 2009), San Diego (Duncan, 2011), Minneapolis (Pilgram
& West, 2017), and in multi-city analyses (Higgins & Kanaroglou, 2016; Bartholomew &
Ewing, 2011)
16
. Eventually, low-income existing residents can no longer afford the housing cost.
They face several options: paying more, doubling-up with others, moving to smaller and thus
cheaper units if available, moving out of the neighborhood, or becoming temporarily or
permanently homeless. While each of these options deserves careful examination, this paper
focuses on the moving option.

Other studies have considered household displacement in the context of gentrification generally,
using survey data. Using neighborhood-level measures of mobility, Vigdor, Massey and Rivlin
(2002) and Freeman and Braconi (2004) found a slow-down of low-income household mobility
in gentrifying neighborhoods in Boston and New York City. Using the same data as Freeman and
Braconi (2004), Newman and Wyly (2006) focused on New York City renters and found that
gentrification increases mobility rates by 6-10%. More relevant to my analysis, Freeman (2005)

16
However, Dong (2017) does not find evidence of increased housing prices or costs in a study of gentrifying rai-
station neighborhoods in suburban Portland, OR. This may be due to the focus on suburban areas.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
100

tested household level move propensities for a U.S. national sample and concluded that the
probability of being displaced from living in a gentrifying neighborhood was 0.9-1.4%, or 0.5%
higher than in a non-gentrifying neighborhood. Ellen and O’Regan (2010) used individual
housing unit data for the U.S. nationally and found “no evidence of heightened exit rates for
renter or for poor households, even among initial residents” in gentrifying neighborhood (Ellen
& O’Regan, 2010, p.24). Most recently, Martin and Beck (2018) tested whether increased
property taxes in gentrifying neighborhoods displaces homeowners. They found that while
property tax mechanisms can trigger forced homeowner moves, these displacements were no
more common in gentrifying than non-gentrifying neighborhoods (Martin & Beck, 2018). In
sum, most of these studies concluded that evidence for a gentrification to displacement link is
mixed at best. However, they had not considered the introduction of new amenities such as rail
stations, which could have strengthened this linkage.

Very few papers have examined the link between new rail stations and displacement (Zuk et al.,
2015; Zuk et al., 2017). At the neighborhood level, Ong, Zuk, Pech, and Chapple (2017)
developed a simulation model to test the effect of proximity to rail stations, the proportion of old
housing, and employment density for census tracts on the likelihood of gentrification and on the
loss of low-income households (equated with displacement) in the San Francisco Bay Area.
Their simulation model correctly predicted actual gentrified tracts for 73 of 85 tracts (86%)
which gentrified from 2000-2013, but incorrectly predicted gentrification in 383 of 512 tracts
(75%) that did not gentrify during the same time period. On displacement, the model correctly
predicted 470 of 537 tracts (88%) which experienced displacement of low-income households,
but incorrectly predicted 769 of 1009 tracts (76%) which did not experience displacement. A
different study used the geocoded version of Panel Study of Income Dynamics (PSID) to
measure the effect of new rail station openings on displacement (Delmelle & Nilsson, 2018).
They tracked about 1000 households across 55 metropolitan areas that have built new rail
stations since 1970, measuring move likelihoods using a logit model. They found that new rail
station openings were not associated with a higher propensity to move, controlling for household
and neighborhood characteristics (Delmelle & Nilsson, 2018).

Overall, there is evidence for a process of rail-induced gentrification. However, the evidence for
a gentrification to displacement link is inconclusive at best and the evidence for a rail-induced
gentrification link to displacement is nascent. Ong et al.’s (2017) simulation model overpredicted
rail-induced displacement as often as it predicts it correctly and focused on legacy rail
transportation access in the San Francisco Bay Area during 2000-2013, a time period where few
new rail stations were opened. Ong et al. (2017) used census data and not household-level data,
precluding household-level conclusions. Their study also did not provide a baseline measure for
average mobility even at a neighborhood level. Delmelle and Nilsson (2018) provided the closest
evidence to date on the question, looking at a longitudinal dataset over across many cities.
However, their small sample size stretched across four decades and fifty-five cities and was very
sensitive to the specific households in the survey. While they did provide a national baseline
mobility, mobility rates differ by neighborhood and by metropolitan area, possibly making their
impact assessment too general.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
101

Literature Gaps & Research Questions
In trying to understand the rail-displacement connection, the majority of the literature reviewed
above suffers from one of four deficiencies: no data to measure household-level mobility, no
baseline to compare mobility rates, not looking at rail stations specifically, or small sample size.
To correct for these issues, I use a rich, household-level dataset for Los Angeles County to
examine the role of neighborhood rail station openings on predicting household moves. Based on
prior findings about the detriments of residential instability for certain groups, I focus my
analysis on low-income households, families with children, young households and elderly
households. My study tests the following hypotheses couched within the overall theme of
displacement:

1. Household move probability increases when a new rail station opens nearby.
2. Low-income household move probability increases when a new rail station opens nearby,
compared to high-income household move probability.
3. Households with children move probability increases when a new rail station opens
nearby, compared to household without children move probability.
4. Households with young heads of household move probability increases when a new rail
station opens nearby, compared to household with elderly head of household move
probability.

The next sections describe the dataset and methods used in testing these hypotheses.

Data and Study Area

This study uses a rich household-level longitudinal dataset to analyze whether nearby rail station
openings increase or decrease household move likelihood, for two rail lines – the Red/Purple and
Gold – in the Los Angeles County context. I measure mobility as the change in a household’s
year to year tax filing location. Then I analyze whether move probability changes for households
before and after nearby rail stations open, controlling for other household characteristics and year
fixed effects. To ensure that I capture the effect of rail stations and not of other metropolitan
area-wide or economy-wide effects, I compare residents near rail stations (treatment households)
to residents of similar neighborhoods which could have, but did not receive a rail station (control
households) as a counterfactual. The research design thus compares treatment and household
before and after rail stations open, similar to a difference-in-difference analysis. This remainder
of this section summarizes the dataset, geocoding strategies, the study area, and control
neighborhood selection. The next section introduces the empirical model.

Data
This type of analysis requires longitudinal data and selection of treatment and control
neighborhoods. I draw household-level mobility, life course, and demographic data from a data
universe of all individuals who have ever filed California taxes from 1993 to 2013 in Los
Angeles County – a dataset of over 100 million records. This dataset, obtained from the
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
102

California Franchise Tax Board (FTB), includes records on all households who moved in or out
of the county; records on any years in which a household lived outside of Los Angeles County
and filed California tax returns, even if from outside of California, are also in the dataset.
Household-specific identifiers enable year-to-year tracking.
17
The longitudinal dataset includes
information available on the California tax return including the households’ income, state taxes
paid, and approximate location, in every year the household is in the data. The data also includes
information on marital status, age of filer, and the number of dependents.

To protect tax filers’ privacy and confidentiality, FTB aggregated household location data at the
9-digit U.S.P.S. zip code level (zip-9), reflecting a block-, block-face-, or building-level
geographic precision. FTB used U.S.P.S. zip codes (zip-5) in areas where zip-9 codes had too
few observations for adequate confidentiality. I geocoded household locations in every year for
which household data exist using correspondence files obtained from Geolytics, Inc, which
contain latitude and longitude of zip-9 centroids (Geolytics, Inc 2015). For data with only zip-5
locations, 5-digit zip code centroid latitude and longitude were obtained from Boutell.com’s
”Free Zip Code Latitude and Longitude Database” and from Zip-code.com’s “California ZIP
Code database” (Boutell.com 2016, Zip-code.com 2016). Observations with neither 5- nor 9-
digit zip codes were dropped from the analysis (ranging from a 0.12-2.83% of total observations
depending on the year); the incidence such observations appears to be spatially random and
diminishes over time. The geocoding strategy is described in more detail in Appendix 2-A and in
Boarnet, Bostic, Rodnyansky, Santiago-Bartolomei, Williams, and Prohofsky (2017).
Having geocoded tax filing locations, I define a move as a household’s change in tax filing
location from year t to year t +1. This definition of moving is subject to two conditions deriving
from the nature of tax data and from geocoding constraints: 1) the household exists in the dataset
for both years and 2) the move is at least 0.5 miles from the initial location.

The first condition is that a filing household must exist in the dataset for consecutive years t and t
+1. If a household violates this condition, it may appear that the household moved, while it in
fact may have stayed in place, in essence a false positive. A household may not show up in the
FTB data for several reasons: non-filing, death, or other circumstances. It is estimated that 85-
90% of all California households file taxes (FTB, 2006; FTB, 2017). Households with incomes
below a certain threshold are not technically required to file taxes (FTB, 2013), but at least 75%
of lower-income California households do file taxes based on data on households who claims the
Earned-Income Tax Credit (EITC) (IRS, 2013). I am comfortable that the dataset thus represents
a near-population-level sample. For more details on non-filing of taxes, see Appendix 2-B and
Boarnet et al. (2017).

The second condition ensures that I track actual moves and not simply changes in postal
geography designations. Specifically, the location (latitude and longitude of a zip-9 centroid) of a
particulate 9-zip is subject to change based on U.S.P.S. business priorities, as zip-9s represent

17
Certain households file in earlier or later years; these were detrended to their nominal filing year to obtain a more
balanced panel. Duplicate entries were likewise removed.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
103

mail delivery routes. These locations tend not to change by more than a few meters if at all: 0.2-
1% of all California zip-9 centroids changed by more than 1 kilometer in any given year
(Boarnet et al., 2017). See Appendix 2-A for details. However, certain years have more changes
than others. To avoid geocoding-related complications, I define a move as a tax location change
of at least 0.5 miles. While moves of less than 0.5 miles are normatively relevant for the moving
household, my dataset currently precludes me from considering these very short distance moves
in the analysis.

Together, these sample restrictions reduce the potential sample for the two transit corridors by an
average of 16-18% across the 1993-2013 time period. Given that lower-income households have
slightly lower filing rates, sample restrictions affect lower income households to a higher extent
than higher income households. Sample restrictions reduce the sample by an average of 23-27%
for households with annual incomes below 30 percent of Area Median Income (AMI) compared
to only 7-10% of households with annual incomes above 80 percent of AMI. Refer to Appendix
2-C for the effect of the restrictions by specific year and transit line.

Study Area
In the past three decades, the Los Angeles County Metropolitan Transportation Authority (L.A.
Metro) has built six different, inter-connected transit lines, for a total of 93 stations. These have
opened at different times, include both light and heavy rail, run through multiple jurisdictions,
and have stops with a variety of land uses. In order to most directly address the research
questions, I restrict the analyses in this paper to three of these transit lines for which I have
sufficient household tax filing data before and after rail station opening and which run through
dense urban neighborhoods: the Red and Purple Subway lines and the Gold Light Rail line. The
Purple Line runs on the same track and serves the same stations as the Red Line for all but two
stations. For this analysis I combine the Red and Purple Lines (Red/Purple from now on). The
excluded lines – the Green and Expo Phase I – do not have enough pre-opening or post-opening
data available for this analysis.

The Red/Purple Line connects a number of neighborhoods in the central portion of the City of
Los Angeles and part of the San Fernando Valley (see Appendix 1-D, Figure 15 for a map). The
openings of the 16 different stations along these lines were staggered from 1990 to 2000 as the
lines were being built out. I only have data from 1993-2013 which covers 15 of 16 stations.
18
Of
the in-sample stations, four opened in 1993, three in 1996, five in 1999, three in 2000 (Appendix
2-D, Table 32). The Gold Line was built in two segments to connect Pasadena and Northeast Los
Angeles neighborhoods to Downtown Los Angeles and then connect East Los Angeles and
Boyle Heights to Downtown Los Angeles (see Appendix 1-D, Figure 16 for a map). Of the 20
in-sample stations, 12 were opened in 2003 and 8 in 2009.

18
The 16
th
and excluded station on the Red/Purple line is 7
th
and Metro Center and it was open prior to the
Red/Purple line (in 1990) as part of a different line (Blue Line).
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
104

These train lines run through dense, urban neighborhoods compared to the rest of Los Angeles
County (Table 24). The neighborhoods along these transit corridors are characterized by a very
high proportion of renters (90% along Red/Purple Line and 65-75% along Gold Line) compared
to 52% countywide and about 35% nationally. These neighborhoods also have large non-white
populations, over and above county averages, and the Red/Purple line has a significant foreign-
born population. Renter, minority, and foreign-born populations tend to be more reliant on transit
(Boarnet et al., 2015). Their displacement away from transit-proximate neighborhoods may thus
reduce their access to jobs and amenities and/or drive up their transportation costs. From a
system perspective, their displacement may reduce system ridership, if they are replaced by
households less likely to ride transit (Manville, Taylor, & Blumenberg, 2018).

Table 24. Neighborhood Description for Study Area

Gold:
Pasadena
Branch
Gold: Boyle
Heights
Branch
Red & Purple
Line
Los Angeles
County average
Opening year 2003 2009 1993, 1996,
1999, 2000

Train type Light Rail Light Rail Underground
Subway

Number of
stations
11 8 15

Population
Density
~13,000 / sq.
mile
~15,000 / sq.
mile
~24,500 / sq.
mile
2,419 / sq. mile
Renter % ~65% ~76% ~90% ~52%
Foreign-born % ~10% ~11% ~18% ~35%
Non-white % ~74% ~95% ~69% ~73%
Note. Data below reflects averages of station area neighborhood descriptive statistics derived from
Boarnet et al. (2015).
Source: U.S. Census 2010, ACS 2009-2013, Boarnet et al. (2015)

Treatment and Control Household Selection and Descriptive Statistics

I define a train-proximate household (treatment household) as one who lives within a 0.5 mile
Euclidean distance of a train station (treatment neighborhood) on the Red/Purple or Gold Line.
This distance reflects at most a 15-20 minute walk to the rail station from any point in the
neighborhood.

To understand whether rail station openings have a causal effect on household move probability,
I test treatment households against counterfactual households in similar neighborhoods that did
not receive a rail station during the time period. Compared to Los Angeles County (see table 24),
households living in treatment neighborhoods are more likely to be renters, to use transit more
frequently, to have annual incomes below $50,000, to be more racially and ethnically diverse, to
be foreign-born, and to not own a vehicle (Boarnet et al., 2015). All of these characteristics
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
105

indicate that treatment neighborhood households are likely more transit dependent than the
average Los Angeles County household. Therefore, holding other considerations constant, they
are likely to welcome the opening of a transit station in their neighborhood. Moreover, if the new
rail station makes their neighborhood more desirable and increases housing costs, the existing
households may place significant value on paying extra to live next to the new rail station;
though, this may be bounded by their relatively low household incomes.

However, what if households living in treatment neighborhoods choose to live in treatment areas
in expectation of rail transit, or rail transit is built because these particular households live in
these particular neighborhoods? Because I do not know households’ explicit preferences, I have
to assume the both households living nearer to and further from a new transit station have a
priori the same preferences for living near transit. I thus use a quasi-experimental identification
strategy that compares households in neighborhoods within 0.5 miles of the station location
(treatment households) to similar households not near the station (control households). Each
treatment neighborhood receives a paired control neighborhood.

Control households are households who reside in control neighborhoods. Control neighborhood
selection follows Chapter 1. Briefly, I select control households that are not walking distance a
rail station, but in a similar part of the County, usually within 0.5 – 3 miles of the rail station.
Following Schuetz, Giuliano, and Shin (2018), my control neighborhoods are centered around a
busy intersection, since rail stations are often built near busy intersections. To choose between all
potential control intersections, I examine the demographic characteristics of the neighborhoods
within a half-mile radius of the intersection, including income, race, ethnicity, housing tenure
and educational status using the U.S. Census and American Community Survey. I then pick the
intersection with demographics most similar to the treatment neighborhood. To avoid overlap
issues, I ensure that the central point of the control neighborhood is at least 1 miles away from all
other non-paired stations, other control neighborhoods, and other L.A. Metro rail stations not in
the analysis.
19

Since I am only focused on transitions out of neighborhoods in which train stations open (or
could open) in this paper, I include in my sample (a) only households who have lived in a
treatment or control neighborhood for at least one year during the study timespan (1993-2013),
and (b) only for the years during which they lived in treatment or control neighborhoods. Thus,
the dataset is not a fully balanced panel; rather, a collection of housing spells during which
households lived in the neighborhoods of interest. The unit of analysis for the study is thus the
household-year pair. The resulting treatment sample size is 1,614,478 Red/Purple Line and
1,294,773 Gold Line household-year observations and the control sample size is 1,364,413
Red/Purple Line and 1,043,105 Gold Line household-year observations from 1993-2013 (Table
25).

19
In a few cases, finding a control neighborhood that is of the correct distance parameters creates a challenges
especially in the transit-dense areas of Downtown L.A. and Central L.A. In these cases, I lower the centroid to
centroid distance requirement of 1 mile to 0.9 miles. For instances where this occurs, see Appendix 2-D.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
106

To measure moves, I compute whether a household moved from their neighborhood from year t
to year t+1, a distance of at least one half-mile (based on geocoding constraints described
above
20
). The unadjusted probability that a treatment households i moves out of a station area in
year t is 26% and 22% for Red/Purple line and Gold line treatment households, respectively
(Table 25). The unadjusted probability that a household i moves out of their neighborhood is
slightly higher for a treatment household than a control household across the 20-year time
period: 26% treatment vs. 24% control for Red/Purple, 20% treatment vs. 18% control for Gold.
These rates are slightly above and below, though very similar to the overall Los Angeles County
21% out-move rate for the same time period (Boarnet et al., 2017). Note that, I keep all
household movers for this analysis, even if they moved to a different transit-station
neighborhood. Though the destination neighborhood still has transit service, the household may
have lost its network or the new transit service may not be as advantageous for their
transportation patterns. I leave the distinction by destination for future research.

To ensure broad similarity between treatment and control households and to set the context for
the data, I briefly summarize the remaining descriptive statistics for both transit lines, in turn
(data in Table 25). Aggregate household characteristics (age
21
, proportion with dependents
22
,
change in number of dependents, proportion married
23
, and proportion experiencing marital
status changes) are very similar between treatment and control households. The treatment group
has slightly higher proportion of households with annual household income below 50% of AMI
(62 percent for treatment versus 55 percent for control), while the control group has slightly
more households with incomes above 120 percent of AMI (14 percent for control versus 10
percent for treatment). The treatment and control groups have a comparable number of
households who experience large upward or downward income swings in any year. These
descriptive statistics imply that overall control and treatment neighborhoods are reasonably
similar and that controls serve as a good counterfactual for the Red/Purple Line study area. For
detailed station-pair descriptive statistics, see Appendix 2-D.

For the Gold Line sample, as with the Red/Purple Line above, the age distribution and proportion
of married households and those experiencing marital status changes are similar between
treatment and control groups. A slightly higher proportion (52 percent) of control households has
dependents present than the treatment group (48 percent), though changes in the number of
dependents were equally likely for both groups. The income profiles of the treatment and control
groups are very similar as is their propensity to experience large upward or downward income
changes. Descriptive statistics suggest that treatment and control groups are quite similar for the
Gold Line study area and will serve as a good counterfactual. For detailed station-pair
descriptive statistics, see Appendix 2-D.

20
As a reminder, I restrict the tax dataset to households filing in at least two consecutive years to effectively
measure year-to-year mobility. I define the dependent variable movedout as a binary operator which equals 1 if
household i moves out of the treatment neighborhood j in time period t and 0 otherwise.
21
Households who do not report an age (6-7%) or who are aged below 17 (0.2%) are excluded from the analysis.
22
Households can claim dependents on their tax returns if those dependents are living in and/or being fully
supported by the household. Dependents are usually children but can include elderly or others.
23
Defined as a “married, filing-jointly” tax filing status
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
107

Analyses of amenity locations can run into issues of endogeneity. In this case, train stations may
be placed in areas expected to have more mobile households or to encourage certain households
to move out and others to move in. It is certainly possible that Red/Purple Line or Gold Line
train station locations are endogenous. However, Schuetz et al. (2018) find that determining
where to build a rail station is a result of multiple factors, including socioeconomic, financial,
engineering, and political considerations using the Los Angeles Metro case, allaying some of the
endogeneity concerns. In addition, I control for household income and age which are both
inversely related to the propensity to use transit.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
108

Table 25. Descriptive Statistics of Household-Year Data for Treatment and Control Households by Transit Line
Red/Purple Line Gold Line

Number of households % of households Number of households % of households
Category Variable Name Treatment Control Total Treatment Control Treatment Control Total Treatment Control

Sample Size
(household-
year)
1,614,478 1,364,413 2,978,891

1,294,773 1,043,105 2,337,878

Sample Size of
Who Moved
(movedout)
423,026 326,187 749,213 26.2% 23.9% 279,295 203,198 482,493 19.8% 17.8%

Sample Size of
Before and
After Train
Station Open
(post)
1,409,847 1,098,678 2,508,525 87.3% 80.5% 600,464 479,050 1,079,514 46% 46%
Age
24
age 1-17 2,437 3,742 6,179 0.2% 0.3% 3,553 3,861 7,414 0% 0%

age 18-25 131,934 110,773 242,707 8.2% 8.1% 149,041 132,267 281,308 12% 13%

age 26-40 682,945 541,604 1,224,549 42.3% 39.7% 470,076 358,255 828,331 36% 34%

age 40-54 425,558 367,833 793,391 26.4% 27.0% 339,167 277,223 616,390 26% 27%

age 55-70 207,936 188,276 396,212 12.9% 13.8% 184,542 153,359 337,901 14% 15%

age 70+ 56,275 68,541 124,816 3.5% 5.0% 63,198 53,184 116,382 5% 5%

No age listed 107,393 83,644 191,037 6.7% 6.1% 85,196 64,956 150,152 7% 6%
Dependents
Dependents
present
660,164 555,828 1,215,992 40.9% 40.7% 617,555 542,170 1,159,725 48% 52%

Change in
number of
dependents
191,281 156,861 348,142 11.8% 11.5% 178,679 155,177 333,856 14% 15%
Marital
Status
Married 419,337 379,097 798,434 26.0% 27.8% 397,455 355,400 752,855 31% 34%

change in
marital status
88,283 68,555 156,838 5.5% 5.0% 71,834 60,263 132,097 6% 6%

24
Households who do not report an age (6-7%) or who are aged below 17 (0.2%) are excluded from the analysis.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
109

Income
2526

<30% of AMI
(<$15,000)
605,204 455,131 1,060,335 37.5% 33.4% 416,701 328,285 744,986 32% 31%

30-50% of
AMI ($15,000-
25,000)
387,361 300,022 687,383 24.0% 22.0% 290,798 241,929 532,727 22% 23%

50-80% of
AMI ($25,000-
40,000)
301,432 254,276 555,708 18.7% 18.6% 247,231 206,589 453,820 19% 20%

80-120% of
AMI ($40,000-
60,000)
164,310 158,784 323,094 10.2% 11.6% 160,571 126,523 287,094 12% 12%

120-240% of
AMI ($60,000-
120,000)
113,919 135,869 249,788 7.1% 10.0% 133,440 102,215 235,655 10% 10%

>240% of AMI
(>$120,000)
42,252 60,331 102,583 2.6% 4.4% 46,032 37,564 83,596 4% 4%

>=25%
increase in
income
441,212 357,090 798,302 27.3% 26.2% 313,596 245,162 558,758 24% 24%

>=25%
decrease in
income
251,565 210,783 462,348 15.6% 15.4% 175,357 136,198 311,555 14% 13%
Source: Author calculations on FTB data

25
I divide income according to fractions of Area Median Income (AMI) for the Los Angeles-Long Beach Metropolitan Statistical Area from 1993-2013, from the
U.S. Department of Housing and Urban Development, because AMI helps determine affordability requirements for various affordable housing programs and
because it is a regional indicator that is calculated annually, fitting will with our annual income data. Appendix 1-C shows AMI from 1993-2013 for the Los
Angeles-Long Beach MSA.
26
Income is shown on this table as converted from 2013 AMI.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
110

Methods

Model Setup
My research design tests hypotheses relating to the question: To what extent does the opening of
a nearby rail-station affect a household’s probability of moving out of their neighborhood? Using
a difference-in-difference setup, I test the probability of moving before and after rail station
openings for treatment and control households. Since move probability in a given year is a binary
consideration, I estimate two models which take binary dependent variables: the linear
probability model (LPM) (Equation 1) and the logit (Equation 2). Though the LPM is easy to
estimate and to interpret, it has known issues (see below). As a result the logit model, which uses
a maximum likelihood estimation setup and a logistic function, provides a good comparison and
robustness check. To control for time-invariant neighborhood-specific effects, I also estimate a
year and neighborhood fixed effects model (Equation 3).

Equation 1: Linear Probability Model
𝑃𝑟 (𝑚𝑜𝑣𝑒𝑑𝑜𝑢𝑡 𝑖𝑗𝑡 = 1|𝑋 ) = 𝛼 + 𝛽 ∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛾 ∗ 𝑝𝑜𝑠𝑡 + 𝛿 ∗ (𝑡𝑟𝑒𝑎𝑡 𝑚 𝑒𝑛𝑡 ∗ 𝑝𝑜𝑠𝑡 ) + 𝜽 ∗
𝑿 𝑖𝑡
+ 𝜇 ∗ 𝑌𝑒𝑎𝑟 𝑡 + 𝘀 𝑖𝑗𝑡

Equation 2: Logit Model
𝑃𝑟 (𝑚𝑜𝑣𝑒𝑑𝑜𝑢𝑡 𝑖𝑗𝑡 = 1|𝑋 ) =
𝑒 (𝛼 +𝛽 ∗𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝑗 +𝛾 ∗𝑝𝑜𝑠𝑡 𝑡 +𝛿 𝑡 (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝑗 ∗𝑝𝑜𝑠𝑡 𝑡 )+𝜽 ∗ 𝑿 𝒊𝒕
+𝜇 ∗𝑌𝑒𝑎𝑟 𝑡 )
(1 + 𝑒 (𝛼 +𝛽 ∗𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝑗 +𝛾 ∗𝑝𝑜𝑠𝑡 𝑡 +𝛿 𝑡 (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝑗 ∗𝑝𝑜𝑠𝑡 𝑡 )+𝜽 ∗ 𝑿 𝒊𝒕
+𝜇 ∗𝑌𝑒𝑎𝑟 𝑡 )
)

Equation 3: Linear Probability Model for with Neighborhood-Area Fixed Effects
𝑃𝑟 (𝑚𝑜𝑣𝑒𝑑𝑜𝑢𝑡 𝑖𝑗𝑡 = 1|𝑋 ) = 𝛼 + 𝛽 ∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛾 ∗ 𝑝𝑜𝑠𝑡 + 𝛿 ∗ (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑝𝑜𝑠𝑡 ) + 𝜽 ∗
𝑿 𝑖𝑡
+ 𝜇 ∗ 𝑌𝑒𝑎𝑟 𝑡 + 𝜔 ∗ 𝑁𝑒𝑖𝑔 ℎ𝑏𝑜𝑟 ℎ𝑜𝑜𝑑
𝑗 + 𝘀 𝑖𝑗𝑡

I define the dependent variable movedout as a binary operator which equals 1 if household i
moves out of the treatment neighborhood j in time period t and 0 otherwise. As a reminder, I
restrict the tax dataset to households filing in at least two consecutive years, to effectively
measure year-to-year mobility. The dependent variable for both models, 𝑃𝑟 (𝑚𝑜𝑣𝑒𝑑𝑜𝑢𝑡 = 1|𝑿 ),
measure the probability that movedout is equal to 1, given some value for each of the other
regressors X. In both models, I regress movedout on a set of household-specific regressors, rail
station-related neighborhood characteristics, and year fixed effects.

Following the difference-in-difference setup, each models above contains a binary treatment
variable indicating whether a household lives in a treatment or control neighborhood in year t
and a binary post variable indicating whether the rail station is open in year t. The coefficient
interaction of these two variables – treatment*post – is the variable of interest, i.e., as it estimates
the average treatment effect on the treated households.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
111

Following the mobility literature, I include a matrix of household-level demographic covariates
Xit for each household-year. Using the available household-level covariates in the FTB dataset, I
include log income, age, number of dependents, and marital status into the regression. From
these covariates, I construct additional life course change variables (major change in income,
change in dependents, change in marital status) each of which have been found to influence
mobility likelihood (South & Crowder, 1997; South & Crowder, 1998; Clark & Davies Withers,
1999). This also helps control for possibly differing average mobility rates between different
population groups (Goldstein 1958, 1964).

I include year-fixed effects to control for non-neighborhood specific exogenous events. Year-
fixed effects year are binary variables for each year t and are added to control for exogenous
events affecting households across neighborhoods in a particular year in each model. The
neighborhood fixed effects model also includes a binary for every neighborhood j to control for
neighborhood-specific idiosyncrasies that do not vary over time. I do not add neighborhood fixed
effects into the logit and LPM estimates (equation 1 and 2) because of the bootstrapping method
I employ to correct for possible measurement error, non-linearity, heteroskedasticity, and
autocorrelation (see Linear Probability Model and Bootstrapping section, below).

Sub-population Models

I am most concerned with the effect of new rail stations on low-income populations, households
with children, and young households. I hypothesize that these groups will be most affected by a
nearby rail station opening. I test the differential effects by income (equation 4), number of
dependents (equation 5), income and number of dependents (equation 6), and age (equation 7).
The equations below build on the LPM model above (equation 1), setting up a difference-in-
difference estimation with year fixed effects. The dependent variable is the same as equation 1.

To test for differential effects of new rail station openings by income group, Equation 4
introduces dummy variables for household income with 3 categories k, with k = {income below
30% AMI, income between 30-80% AMI, and income above 80% AMI}, that are interacted with
the rail-station variables.
27
‘Income above 80% AMI’ is the reference category. The coefficient
of interest is ρ*θ, the interaction between the income category, treatment, and post.

Equation 4: LPM with Income Category Dummies
𝑃𝑟 (𝑚𝑜𝑣𝑒𝑑𝑜𝑢𝑡 𝑖𝑗𝑡 = 1|𝑋 ) = 𝛼 + 𝛽 ∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛾 ∗ 𝑝𝑜𝑠𝑡 + 𝛿 ∗ (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑝𝑜𝑠𝑡 ) +
𝜌 ∗ 𝑖𝑛𝑐𝑜𝑚𝑒 𝑘 ∗ (1 + 𝛽 ∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛾 ∗ 𝑝𝑜𝑠𝑡 + 𝛿 ∗ (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑝𝑜𝑠𝑡 )) 𝜽 ∗ 𝑿 𝑖𝑡
+
𝜇 ∗ 𝑌𝑒𝑎𝑟 𝑡 + 𝘀 𝑖𝑗𝑡

27
I also ran a regression specification similar to Equation 4 but with k=4 income categories: k = {income below
30% AMI, income between 30-50% AMI, income between 50-80% AMI, and income above 80% AMI}. The results
did not materially differ from the 3 income category results below.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
112

To test for differential effects of new rail station openings on households with children or elderly
dependents, Equation 5 introduces dummy variables for number of dependents with 4 categories
m, with m = {no dependents, 1 dependent, 2 dependents, and 3 or more dependents}, that are
interacted with the rail-station variables. ‘No dependents’ is the reference category. The
coefficient of interest is π*θ, the interaction between the number of dependents category,
treatment, and post.

Equation 5: LPM with Number of Dependents Category Dummies
𝑃𝑟 (𝑚𝑜𝑣𝑒𝑑𝑜𝑢𝑡 𝑖𝑗𝑡 = 1|𝑋 ) = 𝛼 + 𝛽 ∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛾 ∗ 𝑝𝑜𝑠𝑡 + 𝛿 ∗ (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑝𝑜𝑠𝑡 ) +
𝜋 ∗ 𝑑𝑒𝑝𝑒𝑑𝑒𝑛𝑡𝑠 𝑚 ∗ (1 + 𝛽 ∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛾 ∗ 𝑝𝑜𝑠𝑡 + 𝛿 ∗ (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑝𝑜𝑠𝑡 )) 𝜽 ∗ 𝑿 𝑖𝑡
+
𝜇 ∗ 𝑌𝑒𝑎 𝑟 𝑡 + 𝘀 𝑖𝑗𝑡
Residential instability tends to have a more negative outcome on children from low-income
households than high-income households (Jelleyman & Spencer, 2008; Morris et al., 2018). I
thus test for the joint effect of income and number of dependents on the probability that a
household moves out after a new rail station opens. Using the same categories for income and
number of dependents as equations 4 and 5, equation 6 interacts income and dependents
categories with each other and with the rail-station variables. No dependents and household
income over 80% of AMI is the reference category. The coefficient of interest is ρ*π*θ, the
interaction between the income category, the number of dependents category, treatment, and
post.

Equation 6: LPM with Number of Dependents and Income Category Dummies
𝑃𝑟 (𝑚𝑜𝑣𝑒𝑑𝑜𝑢𝑡 𝑖𝑗𝑡 = 1|𝑋 ) = 𝛼 + 𝛽 ∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛾 ∗ 𝑝𝑜𝑠𝑡 + 𝛿 ∗ (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑝𝑜𝑠𝑡 ) + 𝜌 ∗
𝑖𝑛𝑐𝑜𝑚𝑒 𝑘 ∗ (1 + 𝜋 ∗ 𝑑𝑒𝑝𝑒𝑑𝑒𝑛𝑡𝑠 𝑚 ∗ (1 + 𝛽 ∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛾 ∗ 𝑝𝑜𝑠𝑡 + 𝛿 ∗ (𝑡𝑟𝑒𝑎𝑡𝑚𝑒 𝑛 𝑡 ∗
𝑝𝑜𝑠𝑡 )) 𝜽 ∗ 𝑿 𝑖𝑡
+ 𝜇 ∗ 𝑌𝑒𝑎𝑟 𝑡 + 𝘀 𝑖𝑗𝑡

To test for differential effects of new rail station openings by age, Equation 7 introduces dummy
variables for five age of filer categories n, with n = {18-25 years old, 26-40 years old, 41-54
years old, 55-70 years old, and above 70 years old}, that are interacted with the rail-station
variables. ‘Middle age (40-54 years old)’ is the reference category. The coefficient of interest is
τ*θ, the interaction between the number of dependents category, treatment, and post.

Equation 7: LPM with Age Category Dummies
𝑃𝑟 (𝑚𝑜𝑣𝑒𝑑𝑜𝑢𝑡 𝑖𝑗𝑡 = 1|𝑋 ) = 𝛼 + 𝛽 ∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛾 ∗ 𝑝𝑜𝑠𝑡 + 𝛿 ∗ (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑝𝑜𝑠𝑡 ) +
𝜏 ∗ 𝑎𝑔𝑒 𝑛 ∗ (1 + 𝛽 ∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛾 ∗ 𝑝𝑜𝑠𝑡 + 𝛿 ∗ (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑝𝑜𝑠𝑡 )) 𝜽 ∗ 𝑿 𝑖𝑡
+ 𝜇 ∗ 𝑌 𝑒𝑎𝑟 𝑡 +
𝘀 𝑖𝑗𝑡

Taken together, these specifications test the displacement potential of rail stations in numerous
ways and for a variety of populations. I run the above specifications separately for Gold and
Red/Purple Line study areas.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
113

Linear Probability Model and Bootstrapping
I use the linear probability model (LPM) for operational and interpretational simplicity, despite
some known issues with this model. Here, I briefly describe some of the common criticisms of
the LPM and how I allay the potential issues.

There are a few common scenarios when non-linear models may be preferred to linear models in
the case of a binary dependent variable:
1) non-linear models are more unbiased and more consistent: there are unbiasedness and
consistency concerns with LPM since ordinary least squares (OLS) is a biased and
frequently inconsistent estimator of the LPM, because predicted probabilities may be
outside of [0,1] (Horrace and Oaxaca, 2007). Trimming observations whose predicted
probabilities lies outside of [0,1] and re-estimating may reduce bias for finite samples.
2) the underlying probability model is known to be non-linear, and a non-linear model
such as the logistic model is a better fit.
3) possibly inappropriate linear significance tests, because the homoscedasticity
assumption is not satisfied (Hellevik, 2007).
4) random sampling error (Hellevik, 2007; Long, 1997).
5) measurement error in the dependent variable, (Hausman et al., 1998).

I discuss each of the above issues and my solutions in turn. First, my predicted probabilities
range between 5-55% across all specifications and do not lie outside of extreme values [0,1],
even for an individual observation. This indicates that in my case, there are fewer consistency
concerns and bias is likely to be insignificant (Hellevik, 2007; Horrace and Oaxaca, 2007).
Second, I have no a priori hypothesis that the underlying model is non-linear. If it were so, and a
logistic model would be preferred, the log odds are a linear function of my regressors but the
probability is not, leading to a non-linear relationship between the probability and the log odds
(Long, 1997). Since the predicted probabilities are not extreme, it is likely that the log odds are a
linear function of the probability (Long, 1997) and so the LPM and the logistic model yield very
similar results. Third, I correct for heteroskedasticity by clustering standard errors at the
neighborhood level (30 clusters for Red/Purple line; 40 clusters for Gold Line)
28
. Fourth, my
sample size is quite large, so I can reject random sampling error as the cause of a particular
association (Hellevik, 2007).

I recognize the potential issue of measurement error in the dependent variable in my dataset. The
data come from an administrative dataset not initially designed to measure household mobility.
As a result, there is the potential for mis-coding a move or a non-move, where in reality, a
household did not file state income taxes in a particular year, but did not actually move. I control
for this by constructing a dataset of households who exist in the dataset for at least two
consecutive years, as far as the data show. This approach may drop certain households who
appear in year t, do not file taxes in year t+1, and reappear again in year t+2, possibly at the
same geographic location. Because I am uncertain of the reasons for non-filing and can not

28
Where bootstrapping methods and bootstrapped standard errors are used (see below) they also correct for
heteroskedasticity.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
114

adequately ascertain that this pattern does not represent moves in two consecutive years, I drop
these observations. Hausman et al. (1998) maintain that adjusted maximum likelihood
procedures can correct for the measurement error in the dependent variable, but the specified
error distribution may be incorrectly assumed and still lead to inconsistent estimates. They
recommend a semiparametric approach that does not specify an error distribution to address both
error types.

To further ensure the robustness of my model to measurement error and several of the other
concerns above, I implement a bootstrapping strategy. Bootstrapping approximates the
distribution of a statistic using a Monte Carlo simulation which samples from a distribution of
the observed data (Cameron and Trivedi, 2009 p.357-388). Specifically, I run all of my
specifications 50 times, drawing 8,000-10,000
29
observations (across all years) from each
neighborhood stratum with replacement, to re-estimate bootstrapped standard errors. The
simulations via the bootstrapping method downplay the effect of any one (or multiple) miscoded
observations in the dependent variable on the error term. The bootstrap also generally provides
correct and consistent standard errors for non-linear models (Gonçalves and White, 2005). Thus,
in case I have mis-specified the model, the bootstrapping will help correct the standard errors.
Finally, bootstrapped standard errors yield better estimates for both autocorrelated and
heteroskedastic errors (Gonçalves and White, 2005). Given the time-series-like nature of my
dataset, the correcting for autocorrelation errors alone makes the bootstrap valuable. I provide
the results of bootstrapped and non-bootstrapped models below.

Results:
This section presents the results of the above models on the effects of rail station openings on
mobility. First, I examine the effects or rail station opening generally on Red/Purple and Gold
Lines without attention to specific populations. Then, I report the effects on other covariates.
Next, I look at results for each sub-population. I also report on two additional analyses: a)
comparing two branches of the Gold line to isolate specific study area effects and b) examine the
effect of rail station opening over time on predicted mobility rates.
Rail Station Opening Effects
Broad-level results across the whole sample indicate a small positive association between rail
station openings and move probability. This association is statistically significant for the Gold
Line using the Bootstrapped LPM and Logit models (Table 26). For the Red / Purple Line, only
the Logit and Neighborhood Fixed Effects model show a positive association between rail station
opening and moving out of the neighborhood. Bootstrapped LPM point estimates (coefficients
on treatment*post variable) reflect a 1.3 percentage point (p.p.) increase in move probability for
the Gold Line, reflecting a 6.9% increase over baseline. From the Logit, the odds of moving
derived from the opening of a nearby rail station increase by 6.0%% for Red/Purple and 9.5% for
the Gold Lines. The Neighborhood Fixed Effects model estimates a 0.8 p.p. increase in move

29
Depending on the minimum sample size per neighborhood. The smallest Red/Purple neighborhoods enable
resampling of 10,000 households; the smallest Gold Line neighborhoods are slightly smaller and enable resampling
of 8,000 households.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
115

probability for Red/Purple Line, reflecting a 2.5% increase over baseline, while Gold Line
estimates are insignificant. Results across models and study areas suggest a general increase of
0-1.5 p.p. in the probability of moving across the whole sample.
Additionally, coefficients on the treatment variable (Table 26) suggest that households living in
treatment areas have slightly higher move propensities than those living in control areas,
especially in Gold Line neighborhoods, controlling for household characteristics and year fixed
effects. Moreover, coefficients on post variable suggest that move probabilities increase for all
households (treatment and control) by about 1-3 percentage points after rail stations open,
especially in Gold Line neighborhoods.
From a specification perspective, LPM and Logit bootstrapped estimates appear to be very
similar, in the direction of effect, statistical significance, and in many cases magnitude (Table
26). Neighborhood Fixed Effects estimates are generally similar to the bootstrapped LPM, but
slightly more conservative on the rail-related variables. F-tests and Chi-squared tests confirm
that all the variables have a jointly significant effect on the probability of moving in every
specification. The AIC and BIC are very similar for both the Bootstrapped LPM and
Neighborhood Fixed Effects models, and slightly lower for the Logit model, indicating a slightly
better fit. Overall, bootstrapping provides a more statistically significant outcome, a better
estimate for autocorrelated and heteroskedastic standard errors, more correct and consistent
standard errors in a non-linear model, and helps allay issues related to any possible measurement
error, and provides. The LPM provides very similar results to the Logit and aids easy
interpretation of the results. Given these considerations, I rely mostly on the bootstrapped LPM
for the remaining analyses.
To compare rail station effect magnitudes, I consider the impact of demographic and life course
variables on moving probability. Results on these variables are consistent with prior findings
from the literature for both transit lines (Table 26). Specifically, the probability of moving
decreases with age, with the number of dependents, and with being married. Moving probability
increases with income, with positive and negative income changes, with changes in the number
of dependents, and changes in marital status. These effects have consistent signs in all
specifications and nearly all are statistically significant. The largest increases in mobility
probability derive from negative changes in income (+ 5.0-5.2 percentage points), positive
changes in income (+ 2.6 p.p.), changes in marital status (+ 4.3-5.2 p.p.), and changes in number
of dependents (+ 2.4-2.8 p.p.). Most of these life course change variables have a higher impact
on mobility than rail station openings, across income groups and rail lines.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
116

Table 26. Regression Results for Household Mobility Determinants for Gold and Red/Purple Lines.
Model LPM BSE Logit LPM
Neighborhood
Fixed Effects
LPM BSE Logit LPM
Neighborhood
Fixed Effects
Standard Errors Bootstrapped Bootstrapped Robust Bootstrapped Bootstrapped Robust
Clustering Household Household
Transit Line Red/Purple Red/Purple Red/Purple Gold Gold Gold
Income All Income All Income All Income All Income All Income All Income
Treatment 0.006 1.034 0.029*** 0.013*** 1.092*** 0.043***
(0.015) (0.028) (0.005) (0.002) (0.016) (0.004)
Post -0.006 0.962 0.002 0.030*** 1.241*** 0.000
(0.007) (0.025) (0.002) (0.003) (0.024) (0.001)
Treatment*post 0.009 1.060** 0.008*** 0.013*** 1.095*** -0.003**
(0.010) (0.031) (0.002) (0.003) (0.027) (0.001)
Age -0.004*** 0.975*** -0.004*** -0.003*** 0.980*** -0.003***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Number of
Dependents -0.010*** 0.950*** -0.008*** -0.004*** 0.979*** -0.001**
(0.002) (0.005) (0.000) (0.001) (0.004) (0.000)
Change in
Dependents 0.028*** 1.171*** 0.029*** 0.024*** 1.170*** 0.027***
(0.001) (0.020) (0.001) (0.002) (0.018) (0.001)
Married -0.014*** 0.923*** -0.014*** -0.023*** 0.845*** -0.020***
(0.003) (0.010) (0.001) (0.002) (0.010) (0.001)
Change in Marital
Status 0.043*** 1.246*** 0.045*** 0.052*** 1.338*** 0.054***
(0.002) (0.027) (0.001) (0.003) (0.029) (0.002)
Log Income 0.014*** 1.092*** 0.011*** 0.016*** 1.132*** 0.010***
(0.002) (0.004) (0.000) (0.001) (0.006) (0.000)
+25% Change in
Income 0.026*** 1.154*** 0.025*** 0.026*** 1.175*** 0.024***
(0.001) (0.014) (0.001) (0.002) (0.013) (0.001)
-25% Change in
Income 0.052*** 1.359*** 0.048*** 0.050*** 1.413*** 0.043***
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
117

(0.003) (0.019) (0.001) (0.002) (0.022) (0.001)
Constant 0.299*** 0.419*** 0.324*** 0.189*** 0.199*** 0.247***
(0.020) (0.021) (0.005) (0.009) (0.014) (0.004)
Year Fixed Effects Yes Yes Yes Yes Yes Yes
Neighborhood Fixed
Effects
No No Yes No No Yes
Observations 2,737,206 2,737,206 2,737,206 2,160,604 2,160,604 2,161,783
Adjusted/Pseudo R
2
0.0248 0.025 0.031 0.022 0.023 0.029
χ
2
-test/ F-test 10579.30 5185.51 3884.48 9174.19 6456.7 2426.86
Prob > χ
2
/ Prob >F 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
AIC 2,922,103 2992386 2,906,661 1,982,118 2,011,192 1,966,830
BIC 2,922,475 2992784 2,907,417 1,982,508 2,011,582 1,967,698
Note. Three specifications shown: LPM with Bootstrapped and Bias Corrected Standard Errors (BSE), LPM with no bootstrapping and errors
clustered at station level (VCE), and Logit model with Bootstrapped and Bias Corrected Standard Errors; Bootstrapped standard error in
parenthesis, with 10,000 observations per neighborhood for Red/Purple Line and 8,000 observation per neighborhood for Gold Line
30
, simulated
50 times.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations on FTB data

30
Difference in bootstrap sampling reflects differences in the minimum station area populations between Red/Purple and Gold lines.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
118

Rail Station Effects on Key Sub-Populations

Low-Income Households
A more nuanced picture arises when drilling into the rail station effect by income group (Table
27). Table 27 provides coefficient estimates for each income category interacted with rail-related
variables. For the Red/Purple line, the rail station effect (coefficient on treatment * post and
income group * treatment * post) is statistically significant only for the highest income group
(above 80% of AMI), increasing by 2.0 pp, or a 4.6% increase over baseline. The results are not
significantly different from zero for the other income groups. Results are similar using the
Neighborhood Fixed Effects specification for households with incomes above 80% AMI (+1.9
p.p. or increase of 3.9% over baseline). Lowest-income households (below 30% of AMI)
increase move rates by 2.2% over baseline, slightly below higher-income households using the
Neighborhood Fixed Effects model.
31
Taken together, these results suggest that Red/Purple Line
station openings may affect households in different parts of the income distribution differently.
Higher-income households increase move rates at slightly higher rates than lower income
households.
The study time period starts in 1993, the same year as the first four stations on the Red Line
opened in and near Downtown Los Angeles. This means that the pre-rail period for these four
stations is only one year, possibly skewing the results. To test for this, I run a regression
excluding Red Line stations opened in 1993, the first year for which I have data available.
Regression result coefficients are very similar to the model with all stations included for all
incomes (Appendix 2-E, Table 39) and for the regression by income groups (Appendix 2-E,
Table 40) regardless of specification (Bootstrapped LPM, Logit, or Neighborhood Fixed
Effects). The evidence does not suggest that excluding those four stations meaningfully changes
Red/Purple Line results.
For the Gold Line, the rail station opening effect increases the mobility rate for households with
incomes over 80% of AMI, by 2.2 p.p. (6% upward impact compared to baseline) using
Bootstrapped LPM estimates and 0.7 p.p. (2% impact) using Neighborhood Fixed Effects. The
specifications disagree on the effect on lower-income households. Bootstrapped LPM results
show slight increases in move probability for households with incomes 30-80% of AMI (2.5%
increase over baseline) and households with incomes less than 30% AMI (1% increase over
baseline). Conversely, the Neighborhood Fixed Effects results show a 2.7% (AMI 30-80%) and
1.7% (AMI less than 30%) decrease in move probability. Similar to the Red/Purple Line, the
Gold Line evidence suggests that train station openings slightly increase move rates for middle
and higher income households, while the effect on lower-income households is less clear.

31
I interpret the relative magnitude of coefficients on income*treatment*post as the sum of partial derivatives of
equation 4, with respect to treatment*post divided by the constant. For example, for the neighborhood fixed effects
model results for the Red/Purple Line in Table 4, the relative magnitude of rail opening on households with incomes
below 30% AMI is (0.019 + (-0.008))/0.490 = 0.022.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
119

Table 27. Income Category Regression Results for Gold and Red/Purple Lines
Model LPM: Income Neighborhood
Fixed Effects:
Income
LPM: Income Neighborhood
Fixed Effects:
Income
Standard Errors Bootstrapped Robust Bootstrapped Robust
Clustering Households Households
Transit Line Red/Purple Red/Purple Gold All Gold All
Treatment 0.018 0.033*** 0.032*** 0.050***
(0.012) (0.005) (0.005) (0.004)
Post -0.001 0.007*** 0.029*** -0.003*
(0.008) (0.002) (0.004) (0.002)
Treatment*post 0.020* 0.019*** 0.022*** 0.007***
(0.012) (0.004) (0.005) (0.002)
Income = <30% AMI -0.006 0.000 -0.016*** -0.010***
(0.009) (0.003) (0.004) (0.002)
Income = 30-80% AMI 0.001 0.005** -0.003 0.001
(0.009) (0.002) (0.004) (0.002)
Income = <30% AMI *
Treatment -0.023 -0.020*** -0.024*** -0.010***
(0.014) (0.004) (0.005) (0.002)
Income = 30-80% AMI
* Treatment -0.014 -0.011*** -0.024*** -0.010***
(0.014) (0.004) (0.006) (0.002)
Income = <30% AMI *
Post -0.013 -0.012*** 0.005 0.005**
(0.010) (0.003) (0.004) (0.002)
Income = 30-80% AMI
* Post -0.005 -0.004 0.003 0.005**
(0.009) (0.003) (0.004) (0.002)
Income = <30% AMI *
Post * Treatment -0.007 -0.008* -0.019*** -0.017***
(0.015) (0.005) (0.006) (0.003)
Income = 30-80% AMI
* Post * Treatment -0.016 -0.017*** -0.013* -0.013***
(0.014) (0.004) (0.007) (0.003)
Constant 0.435*** 0.490*** 0.353*** 0.361***
(0.007) (0.004) (0.006) (0.003)
Other Covariates Age, Number of Dependents, Marital Status, Δ Marital Status, ±Δ 25%
Income
Year Fixed Effects Yes Yes Yes Yes
Neighborhood Fixed
Effects
No Yes No Yes
Observations 2,787,854 2,787,854 2,186,504 2,187,750
Adjusted r2 0.026 0.032 0.021 0.030
χ
2
-test / F-test 9364.69 2403.82 13,768.00 1476.62
Prob > χ
2
/ Prob > F 0.0000 0.0000 0.0000 0.0000
AIC 2,973,847 3133617 2,005,540 1,990,247
BIC 2,974,335 3134464 2,006,019 1,991,204
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
120

Note. LPM with Bootstrapped and Bias Corrected Standard Errors (BSE) specification shown. Income
reference category: >80% AMI (middle-higher income); Bootstrapped standard error in parenthesis, with
10,000 observations per neighborhood for Red/Purple Line and 8,000 observations per neighborhood for
Gold Line, simulated 50 times.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations on FTB data
The Gold Line is made up of two branches that feature different socioeconomic and demographic
characteristics. The Pasadena branch from Downtown Los Angeles, northeast through Northeast
LA, to Pasadena opened in 2003 and the Boyle Heights branch from Downtown Los Angeles
east through Boyle Heights to East Los Angeles opened in 2009. These two lines connect to each
other in Downtown Los Angeles at the Union Station stop. The Boyle Heights branch has a
higher renter concentration and minority percentage, and slightly higher population density than
the Pasadena branch (see Table 24). The Pasadena branch also has slightly higher incomes (See
Appendix 2-D Table 36) and higher median gross rents (Boarnet et al., 2015), while households
near the Boyle Heights branch are slightly more likely to be married and have dependents (See
Appendix 2-D, Table 33, 34, 35). Sections of both lines have seen residential real estate
development activity from 2000-2010, but rents for new residences are 1.5-2 times higher for
Pasadena branch than for Boyle Heights branch, reflecting prior differences in rents (Boarnet et
al., 2015).
Due to these differences, I also examine these two branches of the Gold Line separately, to see if
there are any heterogeneous effects on move probability between the branches (Table 28). Rail
station opening increases the probability of moving among higher-income households by 6.2%
(Boyle Heights) and 4.2% (Pasadena) relative to baseline. Along the Boyle Heights branch,
results for low-income (below 30% of AMI) households are not statistically different for
treatment and control households. Lower-middle income households (30-80% AMI) are 0.3%
less likely to move after stations open relative to baseline – or very similar to no difference.
Along the Pasadena branch, low income households (less than 30% AMI) are 3.8% less likely to
move after households open and lower-middle income households (30-80% AMI) are 0.6% less
likely to move. Results for the two parts of the Gold Line are broadly similar to the Gold Line
results overall: slightly higher move rates for households with incomes above 80% AMI after
stations open and little to no difference for hosueholds with incomes below that threshold.
Table 28. Income Category Regression Results Separated for Two Branches of the Gold Line
Model LPM: Income LPM: Income
Standard Errors Bootstrapped Bootstrapped
Transit Line Gold: Boyle Heights
Branch
Gold: Pasadena Branch
Treatment 0.012** 0.040***
(0.005) (0.006)
Post -0.112*** -0.065***
(0.011) (0.008)
Treatment*post 0.020* 0.015**
(-0.01) (0.008)
Income = <30% AMI 0.001 -0.026***
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
121

(0.005) (0.004)
Income = 30-80% AMI 0.004 -0.004
(0.004) (0.005)
Income = <30% AMI * Treatment -0.023*** -0.008
(0.008) (0.008)
Income = 30-80% AMI *
Treatment -0.021*** -0.01
(0.006) (0.008)
Income = <30% AMI * Post 0.01 0.012**
(0.009) (0.005)
Income = 30-80% AMI * Post 0.007 0.006
(0.008) (0.006)
Income = <30% AMI * Post *
Treatment -0.017 -0.029***
(0.014) (-0.01)
Income = 30-80% AMI * Post *
Treatment -0.021* -0.017*
(0.012) (0.009)
Constant 0.324*** 0.359***
(0.009) (0.008)

Other Covariates Age, Number of Dependents, Marital Status, Δ Marital
Status, ±Δ 25% Income
Year Fixed Effects Yes Yes
Observations 822,291 1,364,213
Adjusted r2 0.021 0.023
χ
2
-test 5170.50 4753.62
Prob > χ
2
0.0000 0.0000
AIC 668,482 1,324,786
BIC 668,912 1,325,234
Note. Income reference category: >80% AMI (middle-higher income); Bootstrapped standard error in
parenthesis, with 8,000 observations per neighborhood for Gold Line, simulated 50 times.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations on FTB data

Households with Dependents
Estimates of the rail station opening effect on households with dependents are not statistically
significant in nearly every case for the Red/Purple Line and Gold Line (Table 29, panels 2 and
3). The lone exception entails a 1% decrease over baseline in the move probability for
households with at least three dependents in Gold Line neighborhoods. On the contrary, rail
station openings increase move probability for households with no dependents: by 1.2 p.p. and
1.4 p.p. for the Red/Purple and Gold Lines respectively. This reflects a 4% and 7% increase in
move probability over baseline. In general, all Gold Line households with dependents have
higher mobility rates than those without dependents; though, treatment households with
dependents have lower mobility probabilities. Along both transit corridors, both control and
treatment households with dependents have lower likelihoods of moving after nearby train
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
122

stations open, but households with no dependents are more likely to move regardless of
treatment or control designation. These findings are consistent with prior literature which shows
that mobility decreases with household size and with the presence of children. The evidence is
thus not supportive of the hypothesis that rail station openings disproportionately increase move
rates for households with children.
I am particularly interested in effects on low-income households with children, given the adverse
effects of residential instability on this group. Panels 4 and 5 of Table 29 show results for the
interaction of income groups and numbers of dependents. There is no statistically significant rail
station opening effect on low or lower-middle income households with any number of
dependents along either the Gold or Red/Purple Lines. This is true of high-income households
with dependents as well. The evidence from the regression again does not suggest that lower
income households with children are disproportionately affected by new rail station openings.

Table 29. Dependents Category Regression Results and Dependents with Income Interaction Regression
Results for Gold and Red/Purple Lines
Model LPM:
Dependents
LPM:
Dependents
LPM: Income &
Dependents
LPM: Income
& Dependents
Standard Error Bootstrapped Bootstrapped Bootstrapped Bootstrapped
Transit Line Red/Purple Gold All Red/Purple Gold All
Treatment 0.008 0.022*** 0.018 0.038***
(0.007) (0.003) (0.015) (0.007)
Post 0.011** 0.045*** 0.011 0.039***
(0.005) (0.004) (0.007) (0.006)
Treatment*post 0.012* 0.014*** 0.018 0.017**
(0.007) (0.004) (0.015) (0.008)
1 Dependent 0.027** 0.016*** 0.006 -0.005
(0.011) (0.004) (0.016) (0.008)
2 Dependents 0.017* 0.017*** 0.002 -0.008
(0.010) (0.004) (0.019) (0.007)
3+ Dependents 0.013 0.013*** -0.015 -0.01
(0.014) (0.004) (0.021) (0.009)
1 Dependent *
Treatment -0.008 -0.011** -0.010 -0.002
(0.015) (0.005) (0.028) (0.011)
2 Dependents *
Treatment -0.007 -0.016*** -0.002 -0.017
(0.018) (0.006) (0.038) (-0.01)
3+ Dependents *
Treatment -0.005 -0.024*** 0.017 -0.027**
(0.020) (0.006) (0.030) (0.011)
1 Dependent * Post -0.042*** -0.025*** -0.028 -0.014
(0.011) (0.005) (0.018) (0.012)
2 Dependents * Post -0.038*** -0.030*** -0.038* -0.023**
(0.010) (0.006) (0.022) (-0.01)
3+ Dependents * Post -0.042*** -0.024*** -0.031 -0.020*
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
123

(0.014) (0.007) (0.020) (0.012)
1 Dependent * Post *
Treatment -0.002 -0.004 0.013 0.000
(0.016) (0.007) (0.030) (0.016)
2 Dependents * Post *
Treatment -0.006 -0.006 0.017 0.007
(0.019) (0.008) (0.039) (0.014)
3+ Dependents * Post *
Treatment -0.015 -0.016* -0.013 -0.01
(0.021) (0.009) (0.029) (0.016)
Income = <30% AMI -0.020* -0.036***
(0.011) (0.006)
Income = 30-80% AMI -0.009 -0.015***
(0.010) (0.005)
Income = <30% AMI *
Treatment -0.019 -0.024***
(0.019) (0.008)
Income = 30-80% AMI
* Treatment -0.012 -0.019**
(0.017) (0.007)
Income = <30% AMI *
Post -0.009 0.005
(0.011) (0.007)
Income = 30-80% AMI
* Post 0.005 0.01
(0.011) (0.007)
Income = <30% AMI *
Post * Treatment -0.001 -0.01
(0.020) (-0.01)
Income = 30-80% AMI
* Post * Treatment -0.009 -0.006
(0.019) (0.009)
1 Dependent * Income
= <30% AMI 0.035 0.044***
(0.023) (-0.01)
2 Dependents * Income
= <30% AMI 0.031 0.048***
(0.030) (0.009)
3+ Dependents *
Income = <30% AMI 0.041 0.038***
(0.037) (0.013)
1 Dependent * Income
= 30-80% AMI 0.016 0.016*
(0.020) (0.009)
2 Dependents * Income
= 30-80% AMI 0.014 0.025***
(0.026) (-0.01)
3+ Dependents *
Income = 30-80% AMI 0.038 0.026***
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
124

(0.027) (-0.01)
1 Dependent * Income
= <30% AMI *
Treatment -0.002 -0.01
(0.038) (0.014)
2 Dependents * Income
= <30% AMI *
Treatment -0.012 0.000
(0.048) (0.016)
3+ Dependents *
Income = <30% AMI *
Treatment -0.037 0.003
(0.053) (0.017)
1 Dependent * Income
= 30-80% AMI *
Treatment 0.004 -0.013
(0.030) (0.012)
2 Dependents * Income
= 30-80% AMI *
Treatment -0.004 -0.002
(0.052) (0.014)
3+ Dependents *
Income = 30-80% AMI
* Treatment -0.030 0.001
(0.037) (0.014)
1 Dependent * Income
= <30% AMI * Post -0.011 -0.009
(0.024) (0.014)
2 Dependents * Income
= <30% AMI * Post 0.005 0.003
(0.030) (0.014)
3+ Dependents *
Income = <30% AMI *
Post -0.004 0.011
(0.038) (0.016)
1 Dependent * Income
= 30-80% AMI * Post -0.021 -0.013
(0.020) (0.015)
2 Dependents * Income
= 30-80% AMI * Post -0.004 -0.01
(0.028) (0.015)
3+ Dependents *
Income = 30-80% AMI
* Post -0.024 -0.01
(0.028) (0.014)
1 Dependent * Income
= <30% AMI *
Treatment * Post -0.018 -0.009
(0.040) (0.019)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
125

2 Dependents * Income
= <30% AMI *
Treatment * Post -0.024 -0.018
(0.051) (0.022)
3+ Dependents *
Income = <30% AMI *
Treatment * Post 0.008 -0.01
(0.056) (0.022)
1 Dependent * Income
= 30-80% AMI *
Treatment * Post -0.017 -0.004
(0.032) (0.018)
2 Dependents * Income
= 30-80% AMI *
Treatment * Post -0.027 -0.017
(0.055) (0.018)
3+ Dependents *
Income = 30-80% AMI
* Treatment * Post 0.001 -0.006
(0.037) (0.021)
Constant 0.283*** 0.207*** 0.178*** 0.352***
(0.009) (0.009) (-0.01) (0.007)

Other Covariates Age, Log Income, Marital
Status, Δ Marital Status, ±Δ 25%
Income
Age, Marital Status, Δ Marital
Status, ±Δ 25% Income
Year Fixed Effects Yes Yes Yes Yes
Observations 2,737,206 2,160,604 2,787,854 2,186,504
Adjusted r2 0.026 0.022 0.022 0.022
χ
2
-test 22518.78 31772.60 23926.17 39867.17
Prob > χ
2
0.0000 0.0000 0.0000 0.0000
AIC 2,921,026 1,980,805 1,980,805 2,003,251
BIC 2,921,564 1,981,334 1,981,334 2,003,868
Note. Income reference category: No dependents. Bootstrapped standard error in parenthesis, with 10,000
observations per neighborhood for Red/Purple Line and 8,000 observations per neighborhood for Gold
Line, simulated 50 times.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations on FTB data

Young and Elderly Households
To assess the effects of rail station openings on elderly and young households, I test for changes
in mobility probability in five separate age categories. I find no effect of rail station openings on
any age group along the Red/Purple Line. Along the Gold Line, however, results indicate that
move probability increases by 16% for households headed by 26-40 year-olds and by 13% for
households headed by 18-25 year-olds (Table 30). These results may reflect households
responding to increasing housing prices near Gold Line stations by moving away from the
neighborhood.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
126

Table 30. Age Category Regression Results for Gold and Red/Purple Lines
Model LPM: Age LPM: Age
Standard Errors Bootstrapped Bootstrapped
Transit Line Red/Purple Gold All
Treatment 0.004 0.012***
(0.007) (0.003)
Post -0.035*** 0.011***
(0.006) (0.004)
Treatment*post 0.005 0.003
(0.008) (0.004)
18-25 years old 0.041*** 0.028***
(0.011) (0.005)
26-40 years old 0.028*** 0.035***
(0.008) (0.003)
55-70 years old -0.093*** -0.051***
(0.011) (0.004)
70+ years old -0.116*** -0.069***
(0.012) (0.006)
18-25 years old * Treatment 0.019 0.007
(0.015) (0.007)
26-40 years old * Treatment 0.008 0.003
(0.011) (0.005)
55-70 years old * Treatment -0.012 -0.005
(0.017) (0.005)
70+ years old * Treatment -0.018 -0.013*
(0.016) (0.008)
18-25 years old * Post 0.040*** 0.034***
(0.013) (0.006)
26-40 years old * Post 0.049*** 0.026***
(0.010) (0.006)
55-70 years old * Post 0.047*** 0.019***
(0.012) (0.005)
70+ years old * Post 0.046*** 0.031***
(0.012) (0.007)
18-25 years old * Treatment * Post 0.006 0.015*
(0.019) (0.008)
26-40 years old * Treatment * Post -0.002 0.019***
(0.012) (0.006)
55-70 years old * Treatment * Post 0.014 0.006
(0.017) (0.007)
70+ years old * Treatment * Post 0.017 0.007
(0.016) (-0.01)
Constant 0.173*** 0.118***
(0.009) (0.009)

Other Covariates Log Income, Number of Dependents, Marital
Status, Δ Marital Status, ±Δ 25% Income
Year Fixed Effects Yes Yes
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
127

Observations 2,917,588 2,301,884
Adjusted r2 0.024 0.023
χ
2
-test 10406.27 11573.17
Prob > χ
2
0.0000 0.0000
AIC 3,158,180 2,159,769
BIC 3,158,773 2,160,351
Note. LPM with Bootstrapped and Bias Corrected Standard Errors (BSE) specification shown. Age
reference category: 40-54 years old. Bootstrapped standard error in parenthesis, with 10,000 observations
per neighborhood for Red/Purple Line and 8,000 observations per neighborhood for Gold Line, simulated
50 times.
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations on FTB data

Longer-term and Anticipation Effects
In the evidence presented thus far, there is no indication of direct rail-station induced
displacement of low-income households, households with children, elderly, or young
households. However, the above analyses considered changes in move likelihood one year after
nearby rail stations opened. Perhaps, the mechanism driving increases in housing costs takes
some time to adjust, as landlords can not immediately raise rents once stations open. Or perhaps
the new amenity attracts wealthier households over time, leading to slower changes in the
neighborhood. Schuetz et al. (2018) find evidence of an effect lag of three to ten years from rail
station opening on employment density changes in neighborhoods around L.A. Metro’s
Red/Purple and Gold Lines. Banzhaf and Walsh (2008) similarly find that wealthier households
move to neighborhoods which have been recently cleaned of pollution with at least a three-year
lag.
At the same time, new rail stations could have anticipation effects on housing costs and
neighborhood composition, and thus potentially on move likelihood, even before stations open.
Landlords could anticipate the new amenity and increase rents; higher-income households could
take advantage of relatively lower rents near a soon to open train station and move in advance.
Such anticipation results have been found for property price appreciation in advance of rail
station openings in Atlanta (Immergluck, 2009). However, the opening effect may be
confounded by the desire to move due to construction-related disruption.
In this section, I explore the possibility of lag and anticipation effects on move probability in my
sample. I use predicted values from regression estimates by income group (equation 4 and table
27, panels 2-3) and by number of dependents interacted with income group (equation 6 and table
27, panels 4-5). I predict values for each year available in the data before and after rail stations
open: 15 years before to 10 years after for Gold Line and 6 years before to 20 years after for
Red/Purple Line. I draw bootstrapped bias-corrected confidence intervals around predicted
values since.
Figures 19.A-C and 20.A-C show predicted mobility probability for treatment versus control
households for Red/Purple and Gold Lines, by income group. These figures show several trends
which align with the regression estimates above. First, there is little discernible difference in
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
128

move probability between treatment and control households with incomes below 80% of AMI. In
fact, their confidence bands overlap for much of the study time period, before and after rail
opening. Overlapping confidence bands indicate that there is not statistically significant
difference between the estimates. The only meaningful departure from this trend is a slight
uptick in move probability for Red/Purple Line lower-income treatment households in years in
the first two years after the station opens, compared to control households (Figures 19.A-B).
While this trend appears to be statistically significant for that period, it is brief and treatment
household mobility likelihoods return to their previous pattern – indistinguishable from those of
control households. This small uptick is not present for low-income Gold Line households.
A second trend is the visible increase in move likelihood on treatment households with income
over 80% of AMI directly following train station openings for both transit corridors (Figures
19.C, 20.C). After the initial increase, mobility rates stay higher than for control households for
the remainder of the study period.
A third trend is the decreasing mobility found in the years before rail stations open, visible to
some degree for all income groups for both transit corridors in both treatment and control
households. This trend accords generally with decreasing mobility rates in Los Angeles County
and nationwide. In addition, it could be an indication of anticipation effects, whereby households
are more likely to stay put and not move, awaiting the opening of the new rail station and the
increased access it brings.
Figure 19. A-C: Predicted Mobility Rates by Income Category for Red/Purple Line
Note: Predictions derived from Dependents Regression Results in Table 27, column 2

15%
20%
25%
30%
35%
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Predicted Mobility Rate
Years Before / After Station Opening
Red/Purple: Income <30% AMI
Control Confidence Interval Treatment Confidence Interval
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
129

Figure 20. A-C: Predicted Mobility Rates by Income Category for Gold Line
Note: Predictions derived from Dependents Regression Results in Table 27, column 3
15%
20%
25%
30%
35%
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Predicted Mobility Rate
Years Before / After Station Opening
Red/Purple: Income 30-80% AMI
Control Confidence Interval Treatment Confidence Interval
15%
20%
25%
30%
35%
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Predicted Mobility Rate
Years Before / After Station Opening
Red/Purple: Income >80% AMI
Control Confidence Interval Treatment Confidence Interval
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
130

10%
15%
20%
25%
30%
35%
-15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Predicted Mobility Rate
Years Before / After Station Opens
Gold Line: Income <30% AMI
Control Confidence Interval Treatment Confidence Interval
10%
15%
20%
25%
30%
35%
-15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Predicted Mobility Rate
Years Before / After Station Opens
Gold Line: Income 30-80% AMI
Control Confidence Interval Treatment Confidence Interval
10%
15%
20%
25%
30%
35%
-15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Predicted Mobility Rate
Years Before / After Rail Station Opens
Gold Line: Income >80% AMI
Control Confidence Interval Treatment Confidence Interval
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
131

I next examine the effects over time for households with and without dependents, comparing
those with incomes below 30% of AMI and above 80% of AMI. Red/Purple Line treatment
households with incomes below 30% of AMI and no dependents increase their move probability
for about 3 years after rail station opening, compared to only about 1 year for same income
households with at least 1 dependent (Figures 21.A-B). In fact, treatment household predicted
mobility rates are virtually indistinguishable from those of control households for those with
incomes below 30% of AMI and with at least one dependent. The same pattern does not exist for
the Gold Line: predicted mobility rates are nearly identical for treatment and control households
with or without dependents (Figures 22.A-B).
Treatment households with incomes above 80% of AMI have higher predicted mobility rates
than control households with or without dependents, before and after rail stations open (Figures
21.C-D, 22.C-D). For the Red/Purple Line, dependent-less households in this income bracket
increase move probability after rail stations open more than control households; the rates
eventually return to pre-opening patterns after 15 years (Figure 21.C). For Red/Purple Line
households with at least 1 dependent, the move rates also increase more for treatment households
in the year after rail stations open, and they stay at an elevated level for at least 15 years
afterwards (Figure 21.D). Gold Line higher-income treatment households without dependents
also increase move probabilities as rail stations open; these remain at an elevated level relative to
controls for at least the next 10 years (Figure 22.C). Gold Line treatment households in this
income bracket with at least one dependent have very similar move rates to control households
before rail stations open, but then increase their move probability after opening; it remains
elevated for at least the next 10 years (Figure 22.D).
These results show that differences between higher and lower-income households matter more
for rail station effects on the likelihood of moving than differences between households with and
without dependents. This result is interesting and has not been studied previously. These figures
also underscore the longer-term effect on move probabilities from rail station openings: in the
higher income brackets, post-opening differences between treatment and control households
persist even 10-15 years after the event.

Figure 21. A-D: Predicted Mobility Rates by Number of Dependents by Income Category for Red/Purple
Line
Note: Predictions derived from Dependents by Income Regression Results in Table 27, column 4
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
132

15%
20%
25%
30%
35%
40%
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Predicted Mobility Rate
Years Before / After Station Opening
Red/Purple: Income <30% AMI and No Dependents
Control Confidence Interval Treatment Confidence Interval
15%
20%
25%
30%
35%
40%
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Predicted Mobility Rate
Years Before / After Station Opening
Red/Purple: Income <30% AMI, with at least 1 Dependent
Control Confidence Interval Treatment Confidence Interval
15%
20%
25%
30%
35%
40%
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Predicted Mobility Rate
Years Before / After Station Opening
Red/Purple: Income >80% AMI and No Dependents
Control Confidence Interval Treatment Confidence Interval
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
133

Figure 22. A-D: Predicted Mobility Rates by Number of Dependents by Income Category for Gold Line
Note: Predictions derived from Dependents by Income Regression Results in Table 27, column 5

15%
20%
25%
30%
35%
40%
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Predicted Mobility Rate
Years Before / After Station Opening
Red/Purple: Income >80% AMI and at least 1 Dependent
Control Confidence Interval Treatment Confidence Interval
10%
15%
20%
25%
30%
35%
-15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Predicted Mobility Rate
Years Before / After Station Opens
Gold Line: Income <30% AMI and No Dependents
Control Confidence Interval Treatment Confidence Interval
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
134

10%
15%
20%
25%
30%
35%
-15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Predicted Mobility Rate
Years Before / After Station Opens
Gold Line: Income <30% AMI and at least 1 Dependent
Control Confidence Interval Treatment Confidence Interval
10%
15%
20%
25%
30%
35%
-15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Predicted Mobility Rate
Years Before / After Station Opens
Gold Line: Income >80% AMI and No Dependents
Control Confidence Interval Treatment Confidence Interval
10%
15%
20%
25%
30%
35%
-15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Predicted Mobility Rate
Years Before / After Station Opens
Gold Line: Income <30% AMI and at least 1 Dependent
Control Confidence Interval Treatment Confidence Interval
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
135

Discussion and Conclusion
This paper examines whether neighborhood level amenity improvements change the probability
that households residing nearby will move. I test whether household move probabilities change
before and after rail transit stations open in their neighborhood in comparison to households in
nearby neighborhoods without rail stations openings for two rail lines in Los Angeles County.
Given the known adverse effects of residential instability, I focus the analysis on particular sub-
populations at risk of displacement by changing housing costs related to the new train stations:
low-income, households with children, elderly households, and young households.
Summary of Results
The evidence from neighborhoods surrounding two of Los Angeles County Metropolitan Transit
Authority’s (L.A. Metro) rail corridors suggests that a household’s probability of moving
increases slightly when rail stations open nearby. However, the evidence does not strongly
suggest that move probabilities increase for low-income households, households with children,
elderly households, or young households. Compared to similar control households, move
probability appears to increase for middle and higher-income households (with incomes above
80% of AMI) along both Red/Purple and Gold Lines. Relative to baseline mobility rates, higher-
income households increase move probability by 5-6% after stations open. The evidence does
not suggest that move rates for low income and lower-middle income households (with incomes
below 30% of AMI and between 30-80% of AMI, respectively) living near the Red/Purple Line
are significantly different from control households not living near rail. The effect on lower-
income households is less clear for the Gold Line: specifications disagree on the direction and
magnitude of the effect. Post-rail station opening move probability also increases for households
without dependents by 4-7% relative to baseline, with no effect for households with dependents.
Examining the effects of rail station openings over a longer time horizon, I find similar results:
very minimal effects if any on low-income households or low-income households with
dependents from rail station openings. When effects do exist, they revert to the mean after 1-3
years. However, mobility rate increases for higher-income households occur right after rail
stations open and persist for 10-15 years.
The general trend of these findings follows the displacement literature: little evidence of large-
scale displacement across various study areas and analytical methods (Freeman, 2005; Vigdor et
al., 2002; Ellen & O’Regan, 2010; Freeman & Braconi, 2004; Newman & Wyly, 2006; Martin &
Beck, 2018). The results are also consistent with Delmelle and Nilsson (2018) who find no
evidence of rail-related displacement of low-income households in a national study using Census
data. My findings indicate that household-level lifecycle changes, such as change in the number
of dependents or in marital status, have a higher impact on the likelihood of moving, as do large
changes in income, consistent with the broader literature on why households move (Quigley &
Weinberg, 1977; Rossi & Shlay, 1982; Speare, 1974; Sabagh et al., 1969). These variables have
a higher impact than rail station openings for higher income households, with income above 80%
of AMI, indicating the primacy of household-level variables over neighborhood-level variables
in move decisions.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
136

This study contributes to the literature on residential mobility by utilizing a novel dataset to
assess transit system externalities on sub-populations at risk of housing instability. The
methodology measures moves directly, as reflected in change in tax filing. The dataset provides a
ready baseline computed at the neighborhood and rail corridor level to assess the relative
magnitude of change in move probability. The tax data also allows for comparisons along the
between income groups, age groups, and numbers of dependents. The results suggest that rail
station openings are statistically correlated with moving across the general population of
households living near rail stations. Potentially, housing costs increase in these neighborhoods
increasing the likelihood of moving. Evidence suggests that these trends do not
disproportionately affect sub-populations at risk of housing instability generally, including
lower-income households, households with children, and households with young or elderly heads
of household. This suggests that these potentially at-risk households may be coping with
increased housing costs in other ways. Thus, this research guides the literature toward exploring
these coping mechanisms, discussed in some detail below.

Discussion of Results
There are several potential explanations for why findings do not suggest that lower-income
households move out at higher rates after rail stations open. I explore each in turn.
First, the presence deed-restricted affordable housing may deter mobility and insulate residents
from market-rate rent increases. Lowest-income households have access to supply-side deeded
affordable housing, including public housing, low-income housing tax credit properties, section 8
project housing, among other programs. If they live in this form of housing, they may be less
likely to move, given the relative scarcity of deeded affordable housing. Even for demand-side
affordable housing such as Section 8 housing choice vouchers, households may be less likely to
move once they find a landlord willing to rent to them.
Second, rent control policies also may deter mobility and insulate residents from market-rate rent
increases. The city of Los Angeles has a rent stabilized policy which covers a large proportion of
multi-family rentals. All of the Red and Purple line stations as well as a portion of Gold line
stations are located inside the city as are their corresponding control neighborhoods. Lower-
income households living in rent-stabilized units may be less likely to move given the lack of
housing choice at their desired price points and given that other lower-income households
engage in the same behavior.
Third, housing search costs may deter households from moving. Housing search is a costly
process, and considerably more so before the widespread penetration of the internet (Sabagh et
al., 1969). Lower-income households in particular may feel the cost from the housing search
process more acutely, due to their relative lack of access to resources, lack of time, and more
limited means and housing budget (Boyd, Edin, Clampet-Lundquist, & Duncan, 2010;
Winstanley, Thorns, & Perkins, 2002; Rosenbaum, Reynolds, & DeLuca, 2002). Hence, they
may be less likely to move due to housing search costs. Together, these three issues may help
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
137

explain the lack of significant findings about rail station openings pushing out low-income
households.
Fourth, households may choose to pay more to avoid moving. Lower-income households may be
willing to pay more for housing to avoid moving, despite the new rail amenity and possible
increases in housing cost. Possibly, lower-income households value the new access that the
transit system provides, and it lowers their cost of transportation, potentially offsetting any
increases in rent. Or, other factors pull households to stay in their existing neighborhoods
including social networks, civic organizations, familiarity, language or ethnic group homophily,
specialty retail and grocery stores. Households with children may pay more in housing to avoid
moving children from neighborhood public schools.
Fifth, households may adjust in ways other than moving. For example, low-income households
may double up with other households in the same neighborhood, invite more members of
extended family to live with them, downsize to smaller apartment to the extent possible, seek
housing assistance from public and non-profit organizations, sell their car, take on extra jobs or
more hours at existing jobs, consume less food or other goods, etc. While these adjustments may
temporarily forestall a move away from their neighborhoods and prevent them from becoming
homeless or transient, the nevertheless lower households’ quality of life. On net, they likely
represent a welfare loss for the household.
Sixth, my proposed mechanism for displacement is rising housing cost due to gentrification;
however, perhaps the rising housing cost is also due to housing undersupply relative to demand.
Without changing zoning near stations, developers would not be able to create more units on the
same amount of land in newly transit-served neighborhoods (Schuetz, Giuliano, & Shin, 2017).
There is evidence that despite a growing population, the City of Los Angeles has downzoned
major portions of its land area from 1960 to 1990 (Morrow, 2013, p.3). In 1960, the city of Los
Angeles had 2.5 million households and population capacity (the number of possible housing
units if each parcel was built up to its allowable limit) of 10 million, meaning the city was full to
25% of capacity. By 1990, the City population had increased to 3.5 million, but the population
capacity decreased to 3.9 million, so that the City was full up to 88% of capacity. From 1990-
2010, population capacity crept up slowly to keep up with population increases and by 2010, the
population increased to 4.0 million and the capacity ratio tightened to 92% full (Morrow, 2013).
This tightened zoning likely increased the competition for adequately zoned space for residential
development and increased housing prices.
This downzoning has covered many portions of the city. In a detailed station-area analysis of
zoning and subsequent development, Schuetz et al. (2017) find that Los Angeles neighborhoods
near the Gold and Red Lines had development after transit opened, while others did not. They
also find that compatible zoning is a necessary, but not sufficient condition for new development
around transit stations (Schuetz et al., 2017). Boarnet and Houston’s (2014) analyses finds
similar heterogeneity in development patterns by station, along Los Angeles’ Gold and Expo
Lines. They also find that pre-existing market conditions and historical urban form patterns drive
development activity more than expanded transit access (Boarnet & Houston, 2014). Finally, a
Los Angeles City policy adopted in 2017 incentivizes the development of Transit Oriented
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
138

Communities by increasing allowable density above established limits in local land use plans,
within 1 mile of well-served transit stations in exchange for a percentage of units to be set aside
for affordable housing (Los Angeles Department of City Planning, 2018). This ordinance may
help align land use with demand for housing near transit lines for all incomes, potentially
alleviating displacement pressure. Because it was only enacted in 2017, this ordinance does not
interfere with my study’s time period that ends in 2013.
Seventh, rail station opening effects may reach further than 0.5 miles. It is possible that the
effects of new rail station openings on mobility are evident beyond my definition of treatment
households as those whose filing address falls within 0.5 miles of a station. I use 0.5 mile
because it is a 15-20 walking distance at most to get to the station from anywhere in the
neighborhood, reflecting what is thought of by transportation planners as the upper bound for the
time most Americans will walk to get to transit. However, my results on post variables indicate
that at least for higher-income households, station openings increase the likelihood of moving for
both treatment and control households. Perhaps a wider catchment area (1 mile instead of 0.5
mile) would encompass this effect better.
The higher move out rates observed for middle and upper income households also deserves
discussion. Compared to lower income households, middle and upper income households have
fewer policy protections (no access to means-tested or deed-restricted affordable housing in most
cases). Housing search costs are also not prohibitive for this group. They can also afford a
broader range of available housing. Higher-income households are also less likely to be transit-
dependent than lower-income households; thus, staying in a neighborhood with new transit
service but increased prices may have less value for them.
Another possible explanation is that stigma associated with public transit may be influencing
higher-income households to move away from newly transit-served neighborhoods. Anecdotal
evidence suggests that higher-income and white households in the U.S. tend to oppose expanding
transit service into their neighborhoods as they believe transit will increase crime, decrease
property values, and bring in more homeless people. While these concerns have resurfaced in
debating the stigma of local buses, relative to trains or bus-rapid transit (e.g., Hess, 2012; Spross,
2017), they are not limited to just buses. Schweitzer (2014), for example, found that social media
user comments reflect negative sentiment about public transit agencies, service, and patrons
compared to other public services. In Los Angeles, stigma-related such concerns were partially
responsible for delaying the Purple Line extension through Beverly Hills to Westwood
(Brightwell, 2013) and for rerouting the Red Line on Vermont Avenue instead of through the
wealthier Hancock Park and Mid-City neighborhoods (Taylor, Kim, & Gahbauer, 2009). Hence,
it may be the case that higher-income households act on such sentiments and move once new rail
stations open. At the same time, however, the Gold and Red/Purple Lines were routed through
relatively lower-income neighborhoods, with high proportions of renters and minorities, before
and after stations opened. Thus, I would argue that households with such sentiments would have
already had reason to move before transit was introduced, so the stigma issue would be partially
mitigated.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
139

Finally, an additional explanation involves the long duration of station-area effects for higher-
income households and the anticipation effects. After stations open, the long duration of
increased mobility likelihood for higher-income households may indicate a longer term effect on
either the housing market or on neighborhood composition in these newly rail-served
neighborhoods, following an initial upward jolt in move rates, relative to surrounding
neighborhoods. At the same time, for both transit lines, move probabilities decrease in
anticipation of rail station openings and increase after stations open. This is possibly explained
by landlords’ inability to raise rents during construction periods or to attract new tenants in the
pre-opening period. This effect could also be mirroring a general downward trend in residential
mobility nationally and in Los Angeles County (Boarnet et al., 2017).
This plethora of possible explanations beckons additional research into the mechanisms behind
household mobility and alternatives for how households adjust to new rail stations and
neighborhood change.

Limitations and Next Steps
While my results and very clear and consistent, I acknowledge several potential limitations.
First, my dataset is limited to households that file taxes. This censoring is an issue because the
distribution of non-filers skews toward the bottom of the income distribution. In California,
about 11% of households who should be filing, do not file taxes (FTB, 2006; FTB, 2017). But,
households with incomes below $25,125 for families and $12,562 for individuals in 2013 dollars
are not required to file California income taxes (FTB, 2013). At the same time, evidence from
EITC compliance suggests that at least 75% of low-income Californians file taxes, even for those
who are not mandated (IRS, 2013). Thus there is a potential gap of up to 14% (89% - 75%) of
California tax filing across the distribution.
There are several circumstances that possibly mitigate this limitation. The data does not show a
spatial skewing of income-based non-filing. Therefore, I am no more likely to have fewer low-
income filers in one station area than another or in a station area versus a control area.
Additionally, over half of my sample are household-years in which households earned fewer than
50% of AMI ($25,000 in 2013) for both Red/Purple and Gold Line study areas, with over 30% of
household-years having incomes below 30% of AMI ($15,000 in 2013). Thus, while this
population may be filing at a slightly lower rate, they are overrepresented in my study area and
sample. I also control for income and for year fixed effects in all of my specifications, which
should partially allay the censoring concern.
A second potential limitation is the control selection method. I currently select control
households by drawing a control neighborhood and selecting households within it, rather than
selecting households directly. I could refine the control-selection scheme by pairing individual
households using a propensity score match, to get a clearer effect identification. However,
propensity score matching works best when propensity scores are calculated on variables that are
not used in the actual regression analysis of the effect. Due to the limited number of covariates in
my data (income, age, number of dependents, and marital status only), it may not be
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
140

methodologically sound to calculate a propensity score using those variables and then regress for
mobility differences using those same variables. An alternative solution would be to create
synthetic neighborhoods as control neighborhoods and select households that way. Yet, synthetic
controls may not improve the results and continue to use a neighborhood-level control selection
scheme, similar to my current method.
A third possible limitation is my methodology. Currently, I estimate move probability in a one-
stage analysis. This provides for clearer identification, but may not be able to clarify the
mechanism behind the effect. Given data on annual housing prices or housing supply changes, a
two-stage model incorporating housing prices or new units could confirm or reject hypothesized
the mechanism behind the rail station opening effect.
A fourth possible limitation concerns specification choice. I use the bootstrapped linear
probability model for my analyses and correct for known issues. I also test sensitivity to
specification by comparing the bootstrapped LPM to a bootstrapped logit model and a
neighborhood fixed effects model with standard errors clustered by household. However, I could
take further advantage of the panel-like structure of the dataset and use a survival-type model,
such as a discrete-time logistic or a Cox proportional. This would calculate hazard rates and
estimate the types of households more likely to move relative to their duration of stay.
Finally, I only look at two transit lines in one metropolitan area. While this greatly simplifies the
analysis and may give clearer effect identification, it may not generalize well to other cases. The
study area – the L.A. Metro Red/Purple and Gold Lines and the households living near them –
may have different mobility characteristics than other existing or potential transit corridors. I
make these choices due to data availability. In future research, I plan to expand this work with a
geographically expanded dataset.
While I will look to address these limitations in future analyses, the current findings reveal
several implications for future work. To further uncover the housing-market mechanism, I plan
to look at the interaction between mobility, housing supply, housing demand, and new rail
station openings. I will also look at whether rent control and/or low-income housing deter
mobility. From the perspective of residential mobility, I will examine where households move
and destination outcomes make these households better or worse off.
This paper reinforces the importance of considering mobility in understanding local processes,
policies, decision-making. Many decisions, including about public investment location, are made
using a static view of neighborhoods. The consideration that households move frequently (20-
30% in my sample; 21% in Los Angeles County) and that certain household types move more
than others can aid policymakers and urban planners in making more relevant choices.
Moreover, due to the adverse effects of residential instability, urban planners and policymakers
should seek solutions that promote stability, especially for at-risk sub-populations.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
141

Appendix

2-A. Geocoding
The geocoding strategy follows Boarnet et al. (2017, p.9-10).
Geocoding tax filing data is necessary to understand where households file taxes and whether
they more from one year to the next. The filer database from FTB does not contain filers’ filing
addresses, but does include a code for the taxpayer’s zip code. Following FTB’s confidentiality
guidelines, the level of zip code reported is chosen based on the number of taxpayers in the zip
code. About half of all records were in zip codes with enough taxpayers to be coded at the 9-
digit level. For most of the remaining records, zip code was coded at the 5-digit level. 9-digit zip
codes, frequently known as zip+4, are unique U.S. Postal Service designations for a set of
addresses within one city block, block-face, set of buildings, or individual building. The zip+4
identifies a household’s location within one city block.
I matched the 9-digit zip codes with latitude and longitude coordinates using conversion files
from Geolytics, a private provider of location data (Geolytics, Inc. 2015). 9-digit zip code
locations may change over time, based on U.S.P.S. needs. However, I measured the change in
centroid distance between years for 9-digit zip codes in 2000, 2004, 2007, 2009, 2012, 2013, and
2014, and found that it is greater than 1 kilometer in less than 0.2%-1% of cases.
32
We are thus
comfortable using geocodes from each of the years listed above. We match FTB data to
geocoded Geolytics data based on the closest year available.
33
For 9-digit zip codes not present
in any Geolytics year or for households without a 9-digit zip code, we used the latitude and
longitude of the centroid for the 5-digit zip code, which was provided for most of the
observations.
34
Observations with neither 5- nor 9-digit zip codes were dropped from the analysis
(ranging from 0.12-2.83% of total observations). The incidence of observations lacking any zip
code data appears to be random and diminishes over time. The incidence of 5-digit versus 9-digit
zip code identification also appears random, excepting confidentiality cases. As such, we are
comfortable that our sample restriction and geocoding approaches do not knowingly bias our
measurements.

2-B. Non-filing of Taxes
Filing in one year and dropping out in the next makes measurement of location change
impossible, as dropped out households do not show a location for the dropped out year. In
addition to moving out of California, reasons to drop out of the data set include marriage, death,

32
Depending on comparison year
33
For example, 2005 FTB data is geocoded using 2004 Geolytics data and so on.
34
5-digit zip code centroid coordinates for the U.S. and for California were obtained from Boutell.com’s ”Free Zip
Code Latitude and Longitude Database” (Boutell, 2016) and from Zip-codes.com’s “California ZIP Code database”
(Zip-codes.com, 2016)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
142

income dropping below the filing threshold, becoming a dependent of another filer, or other
circumstances. Federal estimates of tax non-compliance (non-filing) range from 10 – 16% in the
mid-2000s (IRS, 2010; IRS, 2012; IRS, 2017); in California, this is estimated at 11% during the
same time period (FTB, 2006; FTB, 2017). Households are only required to file taxes federally if
their annual incomes are above a certain level. In 2013, this threshold was $20,000 for families
and $10,000 for individuals federally (IRS, 2013), and $25,125 for families and $12,562 for
individuals in California (FTB, 2013). Income-based filing thresholds are indexed to inflation
and thus vary slightly from year to year. A study of federal tax non-filers using 2005 data
estimated that 77% households who did not file had annual incomes below $20,000, the federal
income mandate (Lawrence, Udell & Young, 2011). This shows that lack of mandate to file, not
blatant tax evasion, is the primary reason not to file taxes. I am not as concerned, however, that
the data is missing a significant portion of the lowest-income households, because 75% of
eligible California households claim the Earned Income Tax Credit (EITC), a tax credit for low
and lower-middle income working households often with children, which requires them to file a
tax return (IRS, 2013). Thus, many of the lowest-income households still file taxes, even if they
fall below the mandatory filing threshold.
Refer to Boarnet et al. (2017) pp.10-11 for more information.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
143

2-C. Sample Restriction Effects
Table 31 describes the effect of sample restrictions on the sample size by year, treatment status, income, and transit corridor. For each
cell, the denominator is the total number of households who live that neighborhood in that year in that income group. The numerator is
the number of households who are in the data for at least two consecutive years.
Table 31. Effects of Sample Restriction by Transit Line by Year for Households of All Incomes
Transit
Line
Red /Purple Gold
Household
Type
Treatment Control Treatment Control
Income All
<30%
AMI
>80%
AMI
All
<30%
AMI
>80%
AMI
All
<30%
AMI
>80%
AMI
All
<30%
AMI
>80%
AMI
1993 73% 63% 88% 78% 67% 91% 78% 67% 90% 78% 67% 91%
1994 75% 67% 88% 80% 70% 92% 80% 72% 90% 81% 70% 92%
1995 74% 66% 89% 79% 70% 92% 78% 68% 91% 78% 68% 93%
1996 78% 72% 89% 82% 74% 92% 81% 73% 91% 82% 74% 93%
1997 80% 74% 89% 83% 76% 92% 82% 75% 91% 83% 75% 93%
1998 82% 76% 89% 85% 78% 92% 84% 77% 92% 85% 78% 93%
1999 83% 77% 90% 85% 79% 93% 84% 77% 92% 85% 78% 93%
2000 83% 76% 91% 85% 78% 93% 84% 76% 92% 85% 77% 93%
2001 84% 78% 91% 86% 79% 93% 85% 77% 92% 86% 77% 94%
2002 81% 75% 90% 84% 77% 92% 82% 75% 90% 83% 75% 92%
2003 84% 77% 91% 87% 79% 93% 85% 77% 92% 86% 78% 93%
2004 84% 78% 91% 87% 80% 93% 86% 79% 92% 87% 79% 93%
2005 84% 78% 90% 87% 80% 93% 86% 78% 92% 87% 79% 94%
2006 85% 79% 90% 87% 80% 92% 86% 79% 92% 87% 79% 94%
2007 83% 75% 90% 86% 77% 92% 85% 75% 92% 85% 74% 93%
2008 84% 77% 91% 86% 78% 93% 86% 76% 93% 86% 76% 94%
2009 85% 79% 92% 87% 81% 93% 86% 79% 92% 87% 79% 94%
2010 85% 78% 91% 87% 80% 93% 86% 78% 93% 87% 79% 94%
2011 84% 77% 90% 86% 79% 92% 85% 78% 92% 86% 78% 94%
2012 83% 76% 89% 85% 78% 91% 84% 76% 91% 85% 77% 92%
Average 82% 75% 90% 84% 77% 92% 84% 76% 92% 84% 76% 93%
Source: Author calculations of FTB data
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
144

2-D. Station-Level Comparisons between Treatment and Station Neighborhoods
Table 32. Treatment and Control Neighborhood Descriptive Statistics: Distance, Year Opened, Sample Size, Mobility Rate
# Station Name Control Intersection Branch Station
Opening
Year
Distance (mi)
between Treatment
& Control Centroid
Sample Size Mobility Rate
Treatment Control Treatment Control
1 Civic Center / Grand
Park
1st / 2nd / Lucas /
Beverly / Glendale
Red
1993 1.0 23,600 51,044 28% 24%
2 Hollywood /
Highland
Fairfax / Santa Monica Red
2000 1.5 39,331 44,091 32% 29%
3 Hollywood / Vine Melrose / La Brea Red
1999 1.7 157,260 33,504 26% 22%
4 Hollywood /
Western
Wilton / Santa Monica Red
1999 0.9 49,138 47,115 33% 26%
5 North Hollywood Victory / Lankershim /
Colfax
Red
2000 1.4 165,777 170,814 26% 23%
6 Pershing Square San Pedro / 8th St Red
1993 0.9 57,706 23,476 25% 27%
7 Universal City /
Studio City
Ventura / Laurel Canyon Red
2000 1.9 16,852 164,454 37% 24%
8 Union Station Main / Griffin Red
1993 1.5 63,739 40,555 25% 16%
9 Vermont / Beverly Western / Beverly Red
1999 1.0 54,496 268,687 31% 22%
10 Vermont / Santa
Monica
Sunset / Silver Lake Red
1999 1.2 185,206 259,233 20% 23%
11 Vermont / Sunset Rowena / Hyperion Red
1999 1.4 49,617 34,053 21% 25%
12 Westlake /
MacArthur Park
Venice / Hoover Red
1993 1.0 154,375 57,761 21% 20%
13 Wilshire /
Normandie
Pico / Western Purple
1996 1.1 199,727 50,265 25% 24%
14 Wilshire / Vermont Beverly / Rampart Purple
1996 1.0 75,041 71,821 31% 22%
15 Wilshire / Western Wilshire / La Brea Purple
1996 1.0 322,613 47,540 24% 23%
16
Allen Washington / Allen
Gold:
Pasadena
2003 1.2 27647 34332 23% 14%
17
Atlantic Garfield / Riggin
Gold: Boyle
Heights
2009 1.4 34299 31792 13% 12%
18
Chinatown Sunset / Echo Park
Gold:
Pasadena
2003 1.5 82086 52408 22% 19%
19
Del Mar California / Allen
Gold:
Pasadena
2003 2 19620 9930 33% 24%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
145

20 East Los Angeles
Civic Center Beverly / Garfield
Gold: Boyle
Heights
2009 1.4 44668 37134 13% 16%
21
Fillmore
Huntington / Garfield /
Atlantic / Los Robles
Gold:
Pasadena
2003 2 20933 26937 31% 22%
22
Highland Park OR: York / Avenue 50;
Gold:
Pasadena
2003 1.1 199165 51885 23% 15%
23
Heritage Square Heritage / Soto
Gold:
Pasadena
2003 1.3 32242 13642 17% 21%
24
Indiana Olympic / Ditman
Gold: Boyle
Heights
2009 1.1 65874 42652 14% 14%
25
Lake Lake / Washington
Gold:
Pasadena
2003 1.2 34518 132792 27% 22%
26 Lincoln Heights /
Cypress Park Cypress / Division
Gold:
Pasadena
2003 1.9 100325 29566 22% 15%
27 Little Tokyo / Arts
District 7th and Santa Fe?
Gold: Boyle
Heights
2009 1.2 26173 1277 26% 48%
28
Memorial Park Fair Oaks / Washington
Gold:
Pasadena
2003 1.4 107426 103053 28% 21%
29
Mariachi Plaza Olympic / Lorena
Gold: Boyle
Heights
2009 2.1 121708 86657 21% 22%
30
Maravilla Olympic / Atlantic
Gold: Boyle
Heights
2009 1.3 43748 37764 15% 14%
31
Pico / Aliso Soto / 8th
Gold: Boyle
Heights
2009 1.4 16382 20225 22% 17%
32
Sierra Madre Villa California / Rosemead
Gold:
Pasadena
2003 1 19244 17458 14% 21%
33
Soto City Terrace / Pomeroy
Gold: Boyle
Heights
2009 1.3 145377 128210 19% 19%
34
South Pasadena Huntington / Main
Gold:
Pasadena
2003 1.4 128678 41880 21% 19%
35
Southwest Museum Eastern / Huntington
Gold:
Pasadena
2003 1.8 24660 143511 19% 20%

Table 33. Treatment and Control Neighborhood Descriptive Statistics: Age Category
Age Categories →
Treatment Neighborhood
Name ↓
Treatment Control

1-17 18-25 26-40 40-54 55-70 70+
No age
data 1-17 18-25 26-40
40-
54
55-
70 70+
No age
data
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
146

1 Civic Center / Grand Park
0% 7% 46% 22% 14% 5% 5% 0% 12% 42% 27% 11% 2% 6%
2 Hollywood / Highland
0% 7% 54% 22% 9% 3% 5% 0% 5% 48% 25% 11% 6% 4%
3 Hollywood / Vine
0% 8% 49% 23% 10% 3% 6% 2% 5% 40% 25% 14% 10% 5%
4 Hollywood / Western
0% 11% 44% 27% 11% 2% 4% 0% 12% 39% 28% 12% 2% 5%
5 North Hollywood
0% 9% 47% 24% 10% 3% 7% 0% 10% 38% 27% 13% 4% 7%
6 Pershing Square
0% 4% 43% 26% 14% 4% 9% 0% 4% 44% 29% 14% 3% 5%
7 Universal City / Studio City
1% 7% 45% 23% 12% 7% 5% 1% 5% 37% 26% 16% 9% 6%
8 Union Station
0% 7% 36% 25% 17% 8% 7% 0% 12% 40% 26% 14% 3% 5%
9 Vermont / Beverly
0% 10% 40% 27% 13% 3% 6% 0% 8% 38% 28% 15% 4% 7%
10 Vermont / Santa Monica
0% 10% 39% 28% 14% 3% 6% 0% 7% 42% 27% 13% 4% 7%
11 Vermont / Sunset
0% 9% 41% 26% 14% 4% 5% 1% 4% 40% 28% 14% 9% 4%
12 Westlake / MacArthur Park
0% 9% 40% 28% 13% 3% 8% 0% 12% 39% 28% 13% 2% 7%
13 Wilshire / Normandie
0% 8% 41% 27% 14% 3% 7% 0% 11% 37% 28% 16% 4% 5%
14 Wilshire / Vermont
0% 8% 44% 27% 13% 3% 6% 0% 10% 39% 28% 15% 4% 5%
15 Wilshire / Western
0% 7% 40% 28% 14% 4% 7% 1% 6% 46% 22% 13% 8% 5%
16
Allen 0% 11% 37% 27% 16% 5% 4% 1% 11% 30% 29% 18% 9% 4%
17
Atlantic 0% 14% 35% 25% 14% 8% 4% 0% 11% 29% 26% 20% 11% 3%
18
Chinatown 0% 10% 33% 26% 17% 6% 7% 0% 12% 41% 26% 13% 3% 5%
19
Del Mar 0% 6% 44% 24% 15% 6% 5% 4% 10% 20% 25% 23% 14% 4%
20 East Los Angeles Civic
Center 0% 16% 37% 25% 12% 4% 6% 0% 14% 35% 26% 15% 6% 4%
21
Fillmore 1% 6% 40% 26% 16% 7% 4% 2% 10% 26% 27% 19% 12% 4%
22
Highland Park 0% 12% 36% 27% 14% 4% 7% 0% 14% 36% 27% 15% 4% 4%
23
Heritage Square 0% 15% 37% 26% 15% 3% 4% 0% 16% 36% 27% 14% 3% 4%
24
Indiana 0% 15% 38% 25% 13% 3% 6% 0% 16% 38% 26% 12% 3% 6%
25
Lake 0% 11% 41% 22% 11% 10% 5% 0% 9% 31% 29% 17% 6% 8%
26 Lincoln Heights / Cypress
Park 0% 13% 36% 27% 14% 3% 7% 0% 15% 36% 26% 15% 3% 5%
27
Little Tokyo / Arts District 0% 6% 42% 26% 14% 6% 5% 0% 4% 50% 35% 7% 1% 2%
28
Memorial Park 0% 8% 43% 22% 13% 6% 8% 1% 12% 32% 27% 16% 6% 7%
29
Mariachi Plaza 0% 13% 35% 26% 13% 4% 8% 0% 14% 37% 26% 12% 3% 9%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
147

30
Maravilla 0% 16% 38% 25% 12% 3% 5% 0% 15% 36% 27% 12% 3% 6%
31
Pico / Aliso 0% 13% 35% 26% 13% 7% 6% 0% 15% 39% 25% 11% 2% 8%
32
Sierra Madre Villa 0% 12% 33% 28% 17% 6% 4% 2% 10% 25% 27% 20% 13% 4%
33
Soto 0% 14% 36% 26% 13% 3% 8% 0% 15% 37% 25% 12% 3% 8%
34
South Pasadena 1% 7% 30% 30% 18% 8% 7% 0% 11% 33% 28% 17% 6% 4%
35
Southwest Museum 0% 10% 34% 29% 18% 5% 4% 0% 13% 36% 25% 14% 4% 6%

Table 34. Treatment and Control Neighborhood Descriptive Statistics: Number of Dependents
Number of Dependents →
Treatment Neighborhood
Name ↓
Treatment Control Treatment Control

0 1 2 3+ 0 1 2 3+ Change in Dependents
1 Civic Center / Grand Park
75% 13% 9% 4% 38% 26% 21% 15% 7% 18%
2 Hollywood / Highland
83% 10% 5% 2% 87% 8% 4% 1% 5% 4%
3 Hollywood / Vine
74% 13% 9% 4% 75% 10% 6% 9% 8% 7%
4 Hollywood / Western
61% 19% 14% 6% 46% 25% 19% 10% 12% 15%
5 North Hollywood
66% 17% 11% 6% 53% 21% 17% 9% 10% 13%
6 Pershing Square
78% 11% 8% 4% 75% 12% 9% 5% 7% 8%
7 Universal City / Studio City
82% 10% 6% 2% 81% 11% 6% 2% 5% 5%
8 Union Station
68% 15% 12% 5% 47% 22% 19% 12% 9% 15%
9 Vermont / Beverly
50% 24% 17% 8% 53% 22% 16% 8% 15% 13%
10 Vermont / Santa Monica
53% 23% 16% 8% 61% 19% 13% 7% 14% 12%
11 Vermont / Sunset
62% 19% 13% 6% 78% 12% 7% 2% 11% 5%
12 Westlake / MacArthur Park
44% 25% 19% 11% 38% 28% 22% 12% 16% 18%
13 Wilshire / Normandie
53% 22% 17% 8% 47% 24% 19% 10% 13% 15%
14 Wilshire / Vermont
52% 24% 17% 8% 48% 25% 17% 10% 14% 15%
15 Wilshire / Western
55% 22% 16% 7% 76% 12% 9% 3% 13% 6%
16
Allen 61% 18% 13% 8% 62% 16% 15% 7% 11% 10%
17
Atlantic 46% 23% 18% 13% 62% 18% 13% 7% 16% 10%
18
Chinatown 64% 17% 13% 6% 57% 19% 15% 9% 10% 12%
19
Del Mar 80% 13% 6% 2% 68% 13% 13% 6% 6% 7%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
148

20 East Los Angeles Civic
Center 39% 24% 20% 17% 46% 21% 19% 14% 18% 15%
21
Fillmore 77% 13% 7% 3% 63% 16% 14% 7% 7% 8%
22
Highland Park 50% 21% 17% 11% 51% 21% 17% 11% 15% 14%
23
Heritage Square 47% 22% 18% 13% 43% 24% 19% 14% 16% 17%
24
Indiana 36% 24% 22% 18% 35% 24% 22% 19% 19% 19%
25
Lake 63% 18% 12% 7% 59% 18% 15% 8% 11% 11%
26 Lincoln Heights / Cypress
Park 46% 22% 19% 13% 45% 23% 19% 12% 16% 16%
27
Little Tokyo / Arts District 81% 10% 6% 3% 83% 8% 4% 4% 7% 7%
28
Memorial Park 76% 13% 7% 4% 48% 21% 19% 12% 7% 15%
29
Mariachi Plaza 37% 24% 22% 17% 35% 24% 22% 19% 19% 19%
30
Maravilla 37% 24% 21% 18% 37% 24% 21% 18% 19% 18%
31
Pico / Aliso 43% 22% 20% 15% 29% 25% 24% 22% 16% 20%
32
Sierra Madre Villa 58% 19% 14% 9% 65% 14% 14% 7% 12% 8%
33
Soto 36% 24% 22% 18% 38% 24% 21% 17% 19% 19%
34
South Pasadena 67% 16% 12% 5% 55% 21% 15% 8% 7% 11%
35
Southwest Museum 59% 19% 13% 8% 46% 23% 18% 13% 11% 16%

Table 35. Treatment and Control Neighborhood Descriptive Statistics: Marital Status
Marital Status →
Treatment
Neighborhood
Name ↓
Treatment Control Treatment Control

Single
Married
(Jointly
Filing)
Married
Filing
Singly
Head of
Household Widow Single
Married
(Jointly
Filing)
Married
Filing
Singly
Head of
Household
Wido
w
Change in Marital
Status
1 Civic Center / Grand
Park 60% 26% 3% 11% 0% 36% 28% 1% 35% 0% 3% 7%
2 Hollywood /
Highland 75% 16% 1% 8% 0% 77% 17% 1% 5% 0% 3% 2%
3 Hollywood / Vine
69% 16% 1% 14% 0% 61% 32% 2% 5% 0% 4% 2%
4 Hollywood / Western
55% 27% 1% 17% 0% 43% 29% 1% 28% 0% 5% 7%
5 North Hollywood
60% 20% 1% 19% 0% 47% 29% 1% 24% 0% 5% 6%
6 Pershing Square
63% 21% 6% 10% 0% 68% 17% 2% 13% 0% 4% 5%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
149

7 Universal City /
Studio City 68% 25% 2% 6% 0% 66% 26% 1% 6% 0% 2% 2%
8 Union Station
55% 30% 1% 13% 0% 43% 34% 0% 23% 0% 4% 6%
9 Vermont / Beverly
44% 30% 1% 25% 0% 46% 28% 1% 24% 0% 6% 6%
10 Vermont / Santa
Monica 48% 25% 1% 26% 0% 54% 23% 1% 21% 0% 6% 5%
11 Vermont / Sunset
52% 28% 1% 18% 0% 64% 28% 1% 7% 0% 5% 2%
12 Westlake /
MacArthur Park 42% 23% 1% 34% 0% 37% 26% 0% 36% 0% 7% 8%
13 Wilshire / Normandie
46% 27% 1% 25% 0% 40% 32% 1% 27% 0% 6% 6%
14 Wilshire / Vermont
44% 30% 1% 25% 0% 42% 29% 1% 28% 0% 6% 6%
15 Wilshire / Western
48% 28% 1% 23% 0% 63% 28% 1% 8% 0% 6% 2%
16
Allen 49% 36% 1% 14% 0% 46% 43% 1% 9% 0% 4% 3%
17
Atlantic 39% 34% 1% 27% 0% 45% 40% 1% 14% 0% 6% 4%
18
Chinatown 52% 32% 1% 15% 0% 51% 27% 1% 21% 0% 4% 5%
19
Del Mar 62% 28% 2% 9% 0% 43% 52% 2% 4% 0% 3% 1%
20 East Los Angeles
Civic Center 36% 33% 0% 31% 0% 38% 34% 1% 27% 0% 7% 6%
21
Fillmore 60% 31% 1% 8% 0% 46% 42% 1% 11% 0% 3% 3%
22
Highland Park 45% 29% 1% 25% 0% 42% 36% 1% 21% 0% 6% 5%
23
Heritage Square 43% 32% 1% 25% 0% 39% 32% 1% 28% 0% 6% 7%
24
Indiana 34% 32% 0% 33% 0% 34% 32% 0% 34% 0% 7% 7%
25
Lake 54% 28% 1% 16% 0% 47% 35% 1% 17% 0% 4% 4%
26 Lincoln Heights /
Cypress Park 42% 31% 1% 26% 0% 41% 32% 1% 26% 0% 6% 6%
27 Little Tokyo / Arts
District 69% 19% 2% 10% 0% 74% 18% 2% 7% 0% 4% 3%
28
Memorial Park 65% 23% 1% 11% 0% 43% 30% 1% 25% 0% 4% 6%
29
Mariachi Plaza 37% 27% 0% 36% 0% 34% 30% 0% 36% 0% 8% 8%
30
Maravilla 35% 33% 0% 32% 0% 35% 32% 0% 32% 0% 7% 7%
31
Pico / Aliso 41% 28% 1% 30% 0% 31% 29% 0% 39% 0% 7% 8%
32
Sierra Madre Villa 46% 37% 1% 15% 0% 44% 47% 1% 8% 0% 4% 2%
33
Soto 36% 28% 0% 36% 0% 36% 31% 0% 33% 0% 8% 7%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
150

34
South Pasadena 52% 35% 1% 12% 0% 43% 38% 1% 18% 0% 3% 4%
35
Southwest Museum 49% 33% 1% 18% 0% 41% 31% 1% 28% 0% 4% 7%

Table 36. Treatment and Control Neighborhood Descriptive Statistics: Income Categories
Income Category →
Treatment Neighborhood
Name ↓
Treatment Control Treatment Control Treatment Control
<30%
AMI
30-80%
AMI
>80%
AMI
<30%
AMI
30-80%
AMI
>80%
AMI
+25% Change in
Annual Income
-25% Change in Annual
Income
1 Civic Center / Grand
Park 25% 29% 46% 43% 46% 10% 28% 26% 16% 13%
2 Hollywood / Highland
29% 40% 31% 25% 35% 40% 31% 30% 19% 19%
3 Hollywood / Vine
36% 44% 20% 24% 29% 47% 29% 29% 17% 20%
4 Hollywood / Western
39% 43% 19% 43% 45% 11% 30% 27% 16% 14%
5 North Hollywood
30% 45% 25% 33% 46% 21% 27% 25% 15% 14%
6 Pershing Square
28% 27% 45% 38% 33% 29% 27% 29% 17% 18%
7 Universal City / Studio
City 21% 29% 50% 20% 29% 51% 30% 27% 19% 19%
8 Union Station
37% 35% 28% 41% 45% 13% 27% 26% 16% 14%
9 Vermont / Beverly
40% 46% 15% 37% 44% 19% 27% 26% 14% 15%
10 Vermont / Santa Monica
41% 45% 15% 33% 42% 24% 26% 25% 15% 15%
11 Vermont / Sunset
36% 42% 22% 20% 28% 53% 28% 26% 16% 17%
12 Westlake / MacArthur
Park 46% 44% 10% 47% 45% 7% 26% 25% 14% 13%
13 Wilshire / Normandie
40% 44% 16% 42% 44% 14% 27% 26% 15% 15%
14 Wilshire / Vermont
40% 44% 16% 38% 46% 17% 28% 26% 15% 13%
15 Wilshire / Western
38% 44% 19% 20% 31% 49% 27% 30% 16% 18%
16
Allen 25% 36% 39% 25% 32% 44% 24% 24% 14% 15%
17
Atlantic 31% 47% 22% 27% 38% 35% 22% 22% 12% 14%
18
Chinatown 40% 36% 24% 36% 43% 21% 27% 27% 15% 15%
19
Del Mar 15% 24% 61% 20% 16% 64% 26% 25% 15% 19%
20 East Los Angeles Civic
Center 34% 50% 16% 31% 45% 24% 23% 22% 12% 12%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
151

21
Fillmore 17% 26% 57% 24% 28% 48% 24% 25% 15% 15%
22
Highland Park 31% 45% 25% 29% 42% 29% 23% 24% 13% 13%
23
Heritage Square 36% 46% 18% 34% 48% 18% 25% 24% 13% 13%
24
Indiana 37% 50% 13% 37% 51% 12% 24% 24% 13% 13%
25
Lake 24% 39% 37% 24% 36% 40% 24% 22% 14% 13%
26 Lincoln Heights /
Cypress Park 39% 45% 16% 34% 46% 20% 25% 24% 14% 13%
27 Little Tokyo / Arts
District 32% 32% 35% 23% 23% 54% 27% 29% 20% 22%
28
Memorial Park 20% 33% 47% 31% 39% 30% 25% 24% 14% 13%
29
Mariachi Plaza 42% 47% 11% 39% 50% 11% 24% 24% 13% 13%
30
Maravilla 35% 50% 15% 36% 51% 14% 24% 24% 12% 13%
31
Pico / Aliso 41% 45% 15% 39% 51% 10% 25% 24% 14% 12%
32
Sierra Madre Villa 28% 38% 34% 24% 27% 49% 23% 25% 14% 16%
33
Soto 41% 48% 11% 36% 50% 14% 24% 23% 13% 12%
34
South Pasadena 19% 29% 52% 26% 38% 36% 23% 23% 14% 13%
35
Southwest Museum 29% 38% 33% 32% 47% 21% 23% 23% 14% 12%

Table 37. Treatment and Control Neighborhood Descriptive Statistics: Median Income
Income Category →
Treatment Neighborhood
Name ↓
Treatment Control

median 10th %ile 90th %ile median 10th %ile 90th %ile
1 Civic Center / Grand
Park $ 40,900 $ 6,800 $ 158,000 $ 18,600 $ 6,400 $ 45,700
2 Hollywood / Highland
$ 28,800 $ 6,100 $ 92,400 $ 35,400 $ 6,200 $ 116,300
3 Hollywood / Vine
$ 22,500 $ 5,700 $ 64,200 $ 40,900 $ 5,600 $ 166,000
4 Hollywood / Western
$ 21,400 $ 5,700 $ 64,200 $ 18,800 $ 6,100 $ 47,600
5 North Hollywood
$ 25,900 $ 7,200 $ 69,700 $ 23,200 $ 7,100 $ 62,400
6 Pershing Square
$ 38,600 $ 3,500 $ 207,100 $ 23,300 $ 4,400 $ 102,900
7 Universal City / Studio
City $ 45,200 $ 6,800 $ 169,800 $ 44,200 $ 7,000 $ 165,400
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
152

8 Union Station
$ 23,100 $ 5,600 $ 90,600 $ 19,700 $ 6,100 $ 51,300
9 Vermont / Beverly
$ 20,400 $ 6,200 $ 54,800 $ 21,300 $ 6,200 $ 65,300
10 Vermont / Santa Monica
$ 19,800 $ 5,900 $ 53,900 $ 24,000 $ 6,600 $ 73,600
11 Vermont / Sunset
$ 23,000 $ 5,900 $ 67,500 $ 47,000 $ 7,900 $ 154,000
12 Westlake / MacArthur
Park $ 17,600 $ 5,800 $ 44,300 $ 17,200 $ 6,100 $ 39,800
13 Wilshire / Normandie
$ 19,900 $ 5,800 $ 56,600 $ 19,500 $ 5,900 $ 54,400
14 Wilshire / Vermont
$ 20,100 $ 5,800 $ 56,900 $ 21,300 $ 6,600 $ 57,800
15 Wilshire / Western
$ 21,100 $ 5,800 $ 63,300 $ 42,900 $ 7,400 $ 149,400
16
Allen $ 33,500 $ 7,500 $ 110,500 $ 37,600 $ 7,600 $ 129,200
17
Atlantic $ 25,500 $ 8,100 $ 64,400 $ 31,200 $ 7,600 $ 89,300
18
Chinatown $ 20,800 $ 5,500 $ 80,000 $ 22,400 $ 6,400 $ 66,200
19
Del Mar $ 59,000 $ 10,900 $ 173,950 $ 77,300 $ 4,500 $ 409,050
20 East Los Angeles Civic
Center $ 22,500 $ 7,800 $ 53,400 $ 25,100 $ 8,000 $ 68,200
21
Fillmore $ 52,800 $ 9,500 $ 165,400 $ 41,300 $ 6,000 $ 191,600
22
Highland Park $ 24,700 $ 7,400 $ 68,200 $ 27,700 $ 7,600 $ 78,200
23
Heritage Square $ 22,400 $ 7,000 $ 60,300 $ 23,000 $ 7,700 $ 59,200
24
Indiana $ 20,800 $ 7,500 $ 49,700 $ 20,800 $ 7,500 $ 47,600
25
Lake $ 32,300 $ 8,600 $ 104,700 $ 32,700 $ 8,300 $ 109,700
26 Lincoln Heights /
Cypress Park $ 20,200 $ 6,300 $ 53,600 $ 23,650 $ 7,400 $ 63,900
27 Little Tokyo / Arts
District $ 28,500 $ 4,900 $ 119,000 $ 53,000 $ 5,200 $ 166,700
28
Memorial Park $ 40,600 $ 9,200 $ 122,900 $ 25,400 $ 7,700 $ 107,700
29
Mariachi Plaza $ 18,800 $ 6,500 $ 46,200 $ 19,400 $ 7,200 $ 45,000
30
Maravilla $ 22,200 $ 7,700 $ 52,000 $ 21,300 $ 7,700 $ 49,900
31
Pico / Aliso $ 20,000 $ 6,000 $ 54,500 $ 19,800 $ 7,400 $ 44,400
32
Sierra Madre Villa $ 30,000 $ 8,000 $ 97,600 $ 43,200 $ 6,400 $ 181,300
33
Soto $ 19,000 $ 6,700 $ 45,400 $ 21,000 $ 7,500 $ 50,400
34
South Pasadena $ 45,000 $ 8,200 $ 141,100 $ 31,900 $ 7,600 $ 100,100
35
Southwest Museum $ 29,200 $ 7,600 $ 100,750 $ 23,600 $ 7,500 $ 60,600
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
153

2-E. Robustness Tests
Table 38. Regression Results for Household Mobility Determinants for Gold and Red/Purple Lines
Model LPM LPM
Standard Errors Robust Robust
Clustering Neighborhood Neighborhood
Transit Line Red/Purple Gold
Income All Income All Income
Treatment 0.006 0.013
(0.005) (-0.01)
Post -0.006 0.030***
(0.004) (0.009)
Treatment*post 0.009 0.013
(0.005) (0.009)
Age -0.004*** -0.003***
(0.000) (0.000)
Number of Dependents -0.010*** -0.004**
(0.001) (0.002)
Change in Dependents 0.028*** 0.024***
(0.003) (0.001)
Married -0.014*** -0.023***
(0.002) (0.003)
Change in Marital Status 0.043*** 0.052***
(0.005) (0.002)
Log Income 0.014*** 0.016***
(0.001) (0.003)
+25% Change in Income 0.026*** 0.026***
(0.002) (0.002)
-25% Change in Income 0.052*** 0.050***
(0.002) (0.003)
Constant 0.299*** 0.189***
(0.010) (0.025)

Year Fixed Effects Yes Yes
Observations 2,737,206 2,160,604
Adjusted r2 0.026 0.022
χ
2
-test/ F-test 471.24 555.97
AIC 2,922,107 1,982,118
BIC 2,922,505 1,982,508
Note. LPM with no bootstrapping and errors clustered at station level (VCE).
* p<0.10 ** p<0.05 *** p<0.01
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
154

Table 39. Regression Results for All Incomes for Red/Purple Line excluding stations opened in 1993
(Civic Center, Pershing Square, Union Station, Westlake / MacArthur Park)
Model LPM BSE Logit LPM Neighborhood
Fixed Effects
Standard Errors Bootstrapped Bootstrapped Robust
Clustering Household
Income All Incomes All Incomes All Incomes
Treatment 0.005 1.029 0.071***
(0.004) (0.029) (0.004)
Post -0.000 0.996 0.005**
(0.006) (0.029) (0.002)
Treatment*post 0.011** 1.070* 0.008***
(0.004) (0.035) (0.002)
Age -0.004*** 0.975*** -0.004***
(0.000) (0.000) (0.000)
Number of Dependents -0.009*** 0.954*** -0.008***
(0.001) (0.005) (0.000)
Change in Dependents 0.027*** 1.166*** 0.028***
(0.003) (0.024) (0.001)
Married -0.013*** 0.929*** -0.014***
(0.002) (0.013) (0.001)
Change in Marital
Status 0.043*** 1.243*** 0.044***
(0.005) (0.035) (0.002)
Log Income 0.012*** 1.078*** 0.010***
(0.001) (0.005) (0.000)
+25% Change in
Income 0.027*** 1.155*** 0.026***
(0.002) (0.015) (0.001)
-25% Change in
Income 0.050*** 1.339*** 0.047***
(0.003) (0.020) (0.001)
Constant 0.325*** 0.189*** 0.340***
(0.011) (0.025) (0.005)

Year Fixed Effects Yes Yes Yes
Observations 2,305,767 2,305,767 2,305,767
Adjusted r2 0.026 0.022 0.031
χ
2
-test/ F-test 6979.66 5668.29 3333.65
AIC 2,476,711 2,412,857 2,463,963
BIC 2,477,103 2,413,250 2,464,608
* p<0.10 ** p<0.05 *** p<0.01 **** 0.01
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
155

Table 40. Regression Results by Income Group for Red/Purple Line excluding stations opened in 1993
(Civic Center, Pershing Square, Union Station, Westlake / MacArthur Park)
Model LPM: Income Neighborhood Fixed
Effects: Income
Standard Errors Bootstrapped Robust
Clustering Households
Transit Line Red/Purple Red/Purple
Treatment 0.017* 0.076***
(0.009) (0.005)
Post 0.003 0.010***
(0.007) (0.003)
Treatment*post 0.019^ 0.018***
(0.011) (0.004)
Income = <30% AMI -0.005 0.001
(0.007) (0.003)
Income = 30-80% AMI 0.001 0.005*
(0.008) (0.002)
Income = <30% AMI *
Treatment -0.023^ -0.020***
(0.012) (0.004)
Income = 30-80% AMI *
Treatment -0.014 -0.011**
(0.011) (0.004)
Income = <30% AMI * Post -0.009 -0.011***
(0.007) (0.003)
Income = 30-80% AMI * Post -0.001 -0.002
(0.008) (0.003)
Income = <30% AMI * Post *
Treatment -0.002 -0.004
(0.013) (0.005)
Income = 30-80% AMI * Post *
Treatment -0.014 -0.016***
(0.013) (0.004)
Constant 0.439*** 0.437***
(0.008) (0.004)
Other Covariates Age, Number of Dependents, Marital Status,
Δ Marital Status, ±Δ 25% Income
Year Fixed Effects Yes Yes
Neighborhood Fixed Effects No Yes
Observations 2,348,738 2,348,738
Adjusted r2 0.026 0.031
χ
2
-test / F-test 9051.14 2052.44
Prob > χ
2
/ Prob > F 0.0000 0.0000
AIC 2,520,896 2,508,020
BIC 2,521,377 2,508,755
* p<0.10 ** p<0.05 *** p<0.01 **** 0.01
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
156

Chapter 3: Does Living near Transit Beget Future Living Near
Transit?

Author: Seva Rodnyansky

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
157

Abstract
Los Angeles County has expanded its subway and light-rail system from 0 stations in 1989 to 93
today, through billions of dollars of federal subsidies and local sales tax revenue. Rail transit
provides faster and more frequent service than other modes. Rail stations are visible and legible,
often focal points for neighborhoods. Other transit modes like bus lines do not become a defining
neighborhood characteristic, despite similar service often at a lower cost to build and operate.
At the same time as the rail expansion, County-wide transit ridership is down over the same
period and most transit users are clustered in a few centrally-located, transit-friendly
neighborhoods (Manville, Taylor & Blumenberg, 2018). Given the simultaneous system
expansion, falling ridership, and transit user clustering, has Los Angeles County become more or
less transit-exposed from 1993-2013? To what extent do households sort into neighborhoods
along the dimension of proximity to transit?
There are at least 4 major reasons why understanding transit exposure is relevant for policy and
planning. First, households who live near transit are more likely to use transit. Second, living
near transit reduces car use and vehicle miles traveled (Cervero, 2007), even when not using
transit (Bailey, Mokhtarian & Little, 2008), which helps reduce greenhouse gas emissions. Third,
the presence of transit often invites additional residential and commercial development. Fourth,
living near transit tends to promote a more active lifestyle and improve individual and aggregate
health outcomes (Besser & Dannenberg, 2005), regardless of actually using transit (Frank &
Engelke, 2001). Hence, increasing transit exposure has benefits for households, for the transit
system operator, and the County as a whole. Due to these reasons, increased transit exposure can
thus be interpreted as a return on investment for financing and building Los Angeles County’s
transit system, even if ridership is stagnant or decreasing.
I explore the concept of transit exposure in terms of population flows using a longitudinal tax
filer database from the California Franchise Tax Board (FTB) available from 1993 to 2013.
Specifically, I test whether new transit lines change households’ move destination choices,
whether prior transit exposure increases the likelihood of moving to transit-proximate
neighborhoods, and what kinds of households move into and out of transit-proximate
neighborhoods.
Findings indicate that new transit lines do not alter move destination patterns. I also find that
living near transit strongly predicts future living near transit. I find that the typical households
moving into a transit-proximate neighborhood are younger, have income below County
median’s, and are non-married.
Taking stock of these findings, I conclude that on a household level, the phenomenon of sorting
into neighborhoods based on transit access is present in Los Angeles County. In aggregate,
however, the introduction of more rail lines and stations has not majorly shifted neighborhood to
neighborhood population flows. That said, Los Angeles County’s overall transit exposure is
increasing as the system expands, though this expansion will be limited until the number of
housing units near transit areas increases dramatically. Population shifts also take time to full
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
158

manifest and possibly we will see more pronounced differences in move patterns in the coming
decades.

Introduction and Motivation
Los Angeles County has expanded its subway and light-rail system from 0 stations in 1989 to 93
today and has added over 100 miles of new track in the process. Rail transit provides faster and
more frequent service than other modes. Rail stations are visible and legible, often focal points
for neighborhoods. Other transit modes like bus lines do not become a defining neighborhood
characteristic, despite similar service often at a lower cost to build and operate. At the same time,
public transit ridership has declined, on net, for both bus and rail service across Southern
California over this time period (Manville, Taylor, & Blumenberg, 2018). Los Angeles Metro
rail transit stations have expanded to many County neighborhoods, but 45% of transit riders in
the region live in neighborhoods that make up less than 1% of land area in the 6-county Southern
California Association of Governments (SCAG) region (Manville et al., 2018). Given the
simultaneous system expansion, falling ridership, and transit user clustering, is Los Angeles
County become more or less transit-exposed from 1993-2013?
35

There are at least 4 major reasons why understanding transit exposure is relevant for policy and
planning. First, living near transit has been shown to increase usage of the transit system
nationwide (Taylor & Fink, 2003) and in Los Angeles County (Boarnet et al., 2017; Chatman et
al., 2017). Prior research has shown that living near transit generally increases the number of
transit trips taken (Cervero and Gorham 1995; Spillar & Rutherford, 1990; Baum Snow & Kahn,
2000; Spears, Boarnet & Houston, 2017), transit mode share (Lin & Long, 2008; Khattak &
Rodriguez, 2005), and probability of taking a transit trip (Chen et al., 2008; Arrington &
Cervero, 2008; Cervero & Duncan, 2008). Second, living near transit reduces car use and vehicle
miles driven (Cervero, 2007), even when not using transit (Bailey, Mokhtarian & Little, 2008),
which helps reduce greenhouse gas emissions. In the Southern California context, this helps
regional governments comply with California Sustainable Communities and Climate Protection
Act of 2008 (SB 375). SB 375 and regional plans link environmental policy and local planning
by directing development to transit corridors, with the aim of reducing emissions (ARB n.d.,
SCAG 2012a). Third, where the presence of transit has catalyzed the development of new or
revitalized residential, commercial, retail, and institutional amenities, known as transit-oriented
developments (TOD), living near transit will make residents more likely to use these amenities.
Schuetz, Giuliano and Shin (2016) provide some evidence that employment density has risen
near Los Angeles County Metropolitan Transit Authority (L.A. Metro) stations five to ten years
after stations have opened. Boarnet & Houston (2014) provide case study evidence of increased

35
I use the term transit exposure as a broad category indicating that households are residing near transit. I
considered other terms including transit-dependent and transit-oriented. Transit-dependent generally refers to
populations who are dependent on transit for getting around either because they lack access to other forms of
mobility, i.e., do not have car or bicycle access, or have a disability which requires using transit. Transit-oriented
hearkens of transit-oriented development (TOD), a framework of residential and commercial development near
transit stations. Each of these terms would steer the conversations toward ridership or development respectively.
Instead, I retain the broader transit exposure term.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
159

residential development along two newer L.A. Metro rail corridors. These studies lend credence
that TOD development is occurring in at least some parts of Los Angeles County. Fourth, living
near transit can promote a more active lifestyle and improve individual and aggregate health
outcomes. Getting to / from transit increases physical activity for transit users (Besser &
Dannenberg, 2005). For those not using transit, living near transit still increases the likelihood of
active trips to use the amenities found in transit-proximate neighborhoods and TODs (Frank &
Engelke, 2001). Due to these reasons, increased transit exposure can be interpreted as an
additional measure of success regarding financing and building Los Angeles County’s transit
system, even if ridership is stagnant or decreasing.
Numerous ways exist to study the changes in transit exposure. I leverage access to a longitudinal
tax filer database from the California Franchise Tax Board (FTB) to answer this question in
terms of population flows from 1993 to 2013. I identify movers in the data by tracking which
households change filing location from year to year. Geocoding origin and destination locations
permits an assessment of where households move. I spatially join mover origins and destinations
to transit station maps to measure whether households moved to / from transit neighborhoods.
Using this dataset and methodology, I distill the idea of transit exposure to several testable
research questions:
1) Do movers’ location choices change before and after a new rail station opens in the
neighborhood?
2) Does prior exposure to (residence near) transit increase the probability of moving to a
transit-proximate neighborhood?
a. Does exposure to non-subway/light-rail modes have the same effect?
b. Does the service have to be operating or is it about the neighborhood type that
could or will support transit service?
3) What is the type of household who moves to / away from transit in Los Angeles County?
This approach yields intriguing findings. First, neighborhood to neighborhood spatial move
patterns are very similar before and after introduction of a new rail transit line. Second, living
near transit begets more living near transit. Living near transit of any type is an extremely strong
predictor of moving near transit in the future across transit models. Third, the majority of moves
in Los Angeles County originate outside of transit-proximate neighborhoods and end outside of
transit-proximate neighborhoods, but this is changing as the L.A. Metro rail transit system
grows. Fourth, the typical households moving into a transit-proximate neighborhood are
younger, have income below County median, and are non-married.

Geographic Move Patterns and New Rail Lines
Building a new light-rail or subway line represents a large physical and financial investment in
an urban area as a whole and in neighborhoods through which the line runs in particular (Kahn,
2007). The line produces both direct (tracks, stations, fences, signals, crossings, etc.) and indirect
(new development, demolitions, lighting, crosswalks, road reconfigurations, bicycle lanes, etc.)
changes to the built environment. These changes may rewire the spatial structure of the
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
160

neighborhood, affect how residents travel around the neighborhood, and attract more visitors to
the neighborhood. While they may affect current residents’ day to day existence, do they also
affect the location choice of movers from these neighborhoods?
Chapter 2 shows that new rail station openings increase move rates out of the neighborhood,
especially for middle- and higher-income households, but did not address move destination and
location choice. In this paper, I hypothesize that a new rail line will alter the pattern of where
households move. Specifically, once households are exposed to transit, I posit that they will be
more likely to move toward transit areas than prior to transit exposure. I hypothesize a stronger
effect for households with lower incomes who may be more transit-dependent (Santos,
McGuckin, Nakamoto, Gray, & Liss, 2011), though this may be moderated by the increased
housing prices near newly-opened transit stations (Higgins & Kanaroglou, 2016; Bartholomew &
Ewing, 2011).
I test this hypothesis by mapping aggregate move destinations at the 5-digit zip code level for
three L.A. Metro rail branches. These include the Gold Line – Pasadena Branch, the Gold Line –
Boyle Heights Branch, and the Red / Purple
36
Line (see map in Appendix 3-A), all of which had
stations which opened between 1993-2013. I omit the Green and Expo Phase I Lines from my
analysis because of insufficient pre-opening or post-opening observations. I define L.A. Metro
rail transit-proximate neighborhoods as areas within a 0.5 mile distance of the station location.
The number of tax filers in each of these 0.5 mile radius neighborhoods is derived by geocoding
tax filer locations at the 9-digit and 5-digit zip code levels (see Chapters 1 and 2).
I use the FTB data on Los Angeles County tax filers from 1993-2013 to identify households who
changed tax filing location from one year to the next. To ensure proper measurement of moves, I
include only households who file taxes for at least two consecutive years to avoid households
who drop out of the data (see Chapter 1 for a discussion of data dropouts) and who move a
distance of at least 0.5 miles to avoid geocoding issues (see Chapter 1 for a discussion of
geocoding issues). To see whether a household lives in a rail station neighborhood, I match the
household’s pre-move location to the 0.5 mile rail neighborhoods described above following
Chapter 2. These steps identify a subset of movers from transit-proximate areas. Next, I identify
the 5-digit zip code of the destination where transit-proximate households move. I map count of
movers moving to each zip code as a proportion of total movers from that transit-proximate
neighborhoods for each transit line, stratifying by pre-station-opening and post-station-opening. I
exclude destinations with fewer than 10 movers for confidentiality reasons.
This method produces a set of maps and accompanying tables, shown in Figures 23.A-C and
Appendix 3-B. Each map shows the top destination (by 5 digit zip codes) for movers from a
particular L.A. Metro rail corridor, at a particular income level. More common destinations are
symbolized by darker colors and less common destinations by lighter colors. The left panel of
each figure shows the pattern before rail stations opened and the right panel shows the patterns
after rail stations opened. The origin zip codes are outlined with a black line in each map. A table

36
The Purple Line has only two stations that are not co-located with the Red Line. I combine them for the analyses
in this paper.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
161

accompanying each set of figures provides the sample statistics for that map. Sample statistics
include the total number of movers from the rail corridor, the proportion of households moving
outside of Los Angeles County, the proportion of households moving less than 0.5 miles, and the
proportion of households not shown for confidentiality reasons. Moves of distances fewer than
0.5 miles are not shown because they may be related to geocoding error, rather than reflecting
actual moves, following Chapter 1 and 2. Destination zip codes with fewer than 10 movers are
not shown for confidentiality reasons.
Visual inspection of the resulting maps show little differences move destination patterns between
pre-opening and post-opening of the transit line. Households living in neighborhoods by the
Gold Line – Pasadena Branch move to similar destinations before and after the line opens
(Figure 23.A), this extends to very low income households as well (Figure 23.B). Pre/post
patterns for higher-income households differ slightly (Figure 23.C), but do not reflect a
wholescale change in move patterns. Gold Line – Boyle Heights Branch and Red / Purple Line
results (Appendix 3-B) are consistent with the Gold Line – Pasadena Branch: move patterns are
quite consistent pre/post rail line opening.
The maps strikingly reveal that the popular destination neighborhoods for households moving
from transit-proximate neighborhoods are in fact other transit-proximate neighborhoods. This
pattern is consistent across the three transit lines in different parts of Los Angeles County and for
both very low and higher-income households. In fact, transit-proximate neighborhoods visually
appear as popular for higher-income as for very low income households in aggregate, though
specific neighborhoods vary in popularity between high-income and low-income.
Summing up the maps visually, it appears that the introduction of a new rail line does not
materially change the move destination patterns. Yet, perhaps the new rail stations reinforce
households’ rootedness in these transit-proximate neighborhoods: transit exposure seems to
encourage rather than deter future transit exposure.
The mapping analysis provides a visual overview of changes in transit exposure in Los Angeles
County. However, these maps are limited in their ability to derive causal or associational
relationship between current and future transit exposure. For this, I turn to regression analysis in
the next section.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
162

Figure 23. A-C: Top Move Destinations (by 5 digit zip code) for Households Moving from L.A. Metro Gold Line – Pasadena Branch
neighborhoods, Before and After Transit service opened (2003)
Note: Each panel represents a different income stratification. Only zip codes with at least 10 moves, from 1993-2012 shown, for confidentiality
reasons.
Source: SCAG, LA County GIS, LA City Planning, Author calculations on California Franchise Tax Board data, Get, HERE, DeLorme,
MapMyIndia, OpenMapData contributors and the GIS User community. Created in ArcGIS.

87,402 Total Number of Movers 87,388
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
163

0.9% % Moved fewer than 0.5 miles 1.5%
15.1% % Moved Outside of LA County 16.2%
0.1% % Suppressed for Confidentiality Reasons 0.1%

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
164

27,719 Total Number of Movers 24,167
0.9% % Moved fewer than 0.5 miles 1.4%
13.7% % Moved Outside of LA County 14.0%
1.2% % Suppressed for Confidentiality Reasons 1.7%

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
165

13,243 Total Number of Movers 18,223
1.7% % Moved fewer than 0.5 miles 2.8%
17.9% % Moved Outside of LA County 19.8%
8.0% % Suppressed for Confidentiality Reasons 6.1%

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
166

Statistical Analysis of Transit Exposure
Having visually established that households in transit-proximate neighborhoods tend to move to
transit-proximate neighborhoods, I formalize this relationship and generalize it to all Los
Angeles County movers. I hypothesize that transit-proximity at origin will increase the
likelihood of moving to a transit-proximate neighborhood. I also hypothesize that this effect
extends to transit modes other than rail. I test these hypotheses using a two-stage discrete choice
model (Figure 24). In this model, household i in year t first makes the decision to move
(Mobility Choice) and then households who move choose whether to move to a neighborhood
with an open L.A. Metro rail station (Location Choice).

Figure 24. Two-Stage Discrete Choice Model of Mobility and Location Decisions
I estimate the model in two stages separately. In the first stage, I model the Mobility Choice as a
logit model of move likelihood on household-level characteristics to control for any systematic
differences between the population of movers and non-movers. I then use predicted values from
the Mobility Choice as a probability weight in the second stage Location Choice model. In the
second stage, I model the Location Choice as a function of whether the household currently
resides near a rail station, the distance moved, and household-specific variables. In both stages,
the unit of analysis is the household-year: each household i decides whether to move in each year
t, and then mover households choose where to move. Then the model repeats for each household
in the next year for which they consecutively file taxes. Since I am interested in the overall effect
of transit exposure on future transit exposure, I do not track households by their total number of
moves to separate frequent-mover households from infrequent-mover households.
My two-stage model is conceptually similar to several prior studies. Cervero & Duncan (2008)
jointly model the probability of living near a rail station, owning a car, and commuting by rail
using a three-stage nested logit (MNML) model. They find that residential self-selection (i.e.,
living near transit to use transit) explains 40% of the decision to commute by rail. My Location
Choice is similar in setup to Clark, Duerloo & Dieleman (2006) found that neighborhood quality
was a strong predictor of residential choice for a sample of Dutch movers. Levine & Frank
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
167

(2007) model whether stated preferences for a more walkable, transit-friendly neighborhood
drive changes in personal neighborhood choices in a hypothetical move. They find that survey
respondents from Atlanta with preferences for walkable environments have a strong desire to
change their current neighborhood and interpret this as an undersupply of walkable, transit-
friendly neighborhoods relative to demand (Levine & Frank, 2006).

Mobility Choice Model Setup
Households choose to move for a variety of reasons. Life course events including changes in
marital status, number of children, and household formation have been found to trigger moves
(Rossi, 1955; Sabagh, van Arsdol, & Butler 1969; Quigley & Weinberg, 1977). Changes in
income also influence the decision to move by expanding or restricting the choice set of possible
locations and housing units (Quigley & Weinberg, 1977). Changes in employment, the demand
for housing space, shifts in commute patterns, preference for particular neighborhood amenities
have all been associated with the decision to move (Tiebout, 1956; Alonso, 1964; Muth, 1969;
Kain, 1968; Brown, 1975; Rossi, 1955; Sabagh et al., 1969). In addition, certain types of
households move more often than other types, including renters, non-married households, young
households, households with no children, individuals living in shared housing or in overcrowded
conditions, urban households, among others (Chapter 2).
I model the Mobility Choice taking into account the differences in propensity to move based on
household characteristics and life course changes. I use the household-level variables available in
the FTB data: income, age, number of dependents, and marital status (Table 41) (see Chapter 2
for details). I define a move in the same way as in Chapters 1 and 2: a move of at least 0.5 miles
based on 9-digit and 5-digit geocoded coordinates among households who file taxes in at least
two consecutive years. This generates a dependent variable moveit, which takes the value 1 if
household i moves in year t and 0 otherwise. The logit model is a consistent estimator of
probability that household i moves in year t, given by
(1) 𝑀𝑜𝑏𝑖𝑙𝑖𝑡𝑦 𝐶 ℎ𝑜𝑖𝑐𝑒 : Pr(𝑀𝑜𝑣𝑒 𝑖𝑡
= 1|𝑋 ) =
𝑒 (𝛼 +𝜷 ∗𝑿 𝒊𝒕
+𝜇 𝑡 +𝘀 𝑧 )
1 + 𝑒 (𝛼 +𝜷 ∗𝑿 𝒊𝒕
+𝜇 𝑡 +𝘀 𝑧 )

Where Xit is the vector of independent variables associated with household i in year t (including
income, age, marital status, number of dependents, change in marital status, change in number of
dependents, and +/- 25% change in income, see Table 41), µt is the year fixed effect, and εz is the
error term, clustered by the 5-digit zip code in which the household currently resides.
Based on prior literature, I expect that changes in marital status and number of dependents will
increase move probability (Clark & Davies Withers, 1999; van der Vliest, Gorter, Nijkamp, &
Rietveld, 2002; Ioannides & Kan, 1996; Clark & Huang, 2003; Weinberg, 1979), because of
potential changes in demand for housing space and/or neighborhood amenities, or in moving to a
joint residence from two disparate residences. I expect that due to large increases [decreases] in
income, households may [no longer] be able to afford a larger or better apartment [no longer able
afford the current apartment], prodding them to move (Chapter 2). I also expect that marital
status, older age of household head, presence of dependents, and large household size will be
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
168

negatively associated with the probability of moving (South & Crowder, 1998; Ioannides & Kan,
1996; Clark & Davies Withers, 1999; Lee, 2014).

Location Choice Model Setup
I model a mover’s Location Choice by calculating the probability that a moving household
chooses to move to a neighborhood near an open L.A. Metro rail transit station, given a set of
household-specific variables, the distance moved, and whether the household currently lives in a
transit-proximate neighborhood. In this stage, the binary dependent variable yit takes the value 1
if household i moves to a transit-proximate neighborhood in year t and 0 if it moves to a different
neighborhood. To take into account the Mobility Choice in the first stage, I estimate the Location
Choices using a weighted maximum likelihood (WML) estimator (Manski & Lerman, 1977).
This maximizes the weighted log-likelihood for binary outcome models following Equation 2
(Cameron & Trivedi, 2005, p.479)
(2) 𝑊𝑀𝐿 𝐸𝑠𝑡𝑖𝑚𝑎𝑡𝑜𝑟 : 𝐿 𝑁 𝑊 (𝛽 ) = ∑ {(
𝑄 1
𝐻 1
) ∗ 𝑦 𝑖𝑡
∗ 𝑙𝑛 (𝐹 (𝑿 𝑖 ′
𝛽 )) + (
𝑄 0
𝐻 0
) ∗ (1 − 𝑦 𝑖𝑡
) ∗ ln (1 − 𝐹 (𝑿 𝑖 ′
𝛽 ))}
𝑁 𝑖 =1

where Q1 / H1 represents the probability that the household is a mover and Q0 / H0 represents the
probability that the household is not a mover. The Location Choice itself is modeled as a
function of the vector of independent variables 𝐹 (𝑿 𝑖 ′
𝛽 ), which includes whether the household
lives near an open rail station currently, the distance moved, year fixed effects and household-
specific characteristics:
(3) 𝐿𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝐶 ℎ𝑜𝑖 𝑐 𝑒 : Pr(𝑦 𝑖𝑡
= 1|𝑋 ) = 𝐹 (𝑿 𝑖 ′
𝛽 )
=
𝑒 (𝛼 +𝜷 ∗𝑿 𝒊𝒕
+𝛾 ∗𝑂𝑟𝑖𝑔𝑖𝑛𝑂𝑝𝑒𝑛𝑅𝑎𝑖𝑙 𝑖𝑡
+𝛿 ∗𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝐶𝑎𝑡𝑒𝑔𝑜𝑟𝑦 𝑖𝑡
+𝜇 𝑡 +𝘀 𝑧 )
1 + 𝑒 (𝛼 +𝜷 ∗𝑿 𝒊𝒕
+𝛾 ∗𝑂𝑟𝑖𝑔𝑖𝑛𝑂𝑝𝑒𝑛𝑅𝑎𝑖𝑙 𝑖𝑡
+𝛿 ∗𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝐶𝑎𝑡𝑒𝑔𝑜𝑟𝑦 𝑖𝑡
+𝜇 𝑡 +𝘀 𝑧 )

The WML estimator is a consistent estimator of the probability that household i moves to a
transit-proximate neighborhood in year t, given weighted by the likelihood that it moves in stage
1 (Cameron & Trivedi, 2005, p.828). In the location choice model, Xit is a vector of independent
variables associated with household i in year t including (income, age, marital status, number of
dependents). There are two additional independent variables of interest: OriginOpenRailit is a
binary variable equal to 1 if the household currently resides in a rail-proximate neighborhood and
0 otherwise and DistanceCategoryit is a categorical variable reflecting the distance the household
moves. As in the Mobility Choice model, µt is the year fixed effect, and εz is the error term,
clustered by the 5-digit zip code in which the household currently resides.

Measuring Transit Proximity
I define transit proximity in four ways for the Location Choice model:
1) Residence near Open L.A. Metro Rail stations
2) Residence near Open or Future L.A. Metro Rail stations
3) Residence in a Transit Priority Area (TPA)
4) Residence in a High Quality Transit Area (HQTA)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
169

Transit proximity measures 1 and 2 reflect access to high-frequency, higher-speed, and high-
capacity light rail and subway service provided by L.A. Metro. Measure 1 is a pure measure of
living within 0.5 miles of an open transit station. I use Measure 1 as the base model. Measure 2
takes into account that the L.A. Metro rail system grew from 22 stations prior to 1993 to 80 in
2013. Movers to / from these neighborhoods could have anticipated the rail system expansion;
or, these neighborhoods may have been attractive to transit-dependent or transit-seeking
households even before the stations were built or announced. Measure 3 and 4 are more general
transit access measures defined by the Southern California Association of Governments (SCAG)
in their 2012 regional plan (SCAG, 2012a) and reiterated in their 2016 regional plan (SCAG,
2016). Specifically, the Transit Priority Areas (TPA) “are defined as locations where two or
more high-frequency transit routes intersect” while the High Quality Transit Areas (HQTA) are
“defined as an area within one-half mile of a well-serviced fixed guideway transit stop, and it
includes bus transit corridors where buses pick up passengers every 15 minutes or less during
peak commute hours” (SCAG, 2016 p.22). The TPA is a more restrictive definition than the
HQTA. While both the TPA and HQTA were defined in 2012, at the later end of this study’s
time period, these measures reflect a broad measure of transit service accumulated over time.
The intersections that comprise TPAs and the corridors that comprise HQTAs did not simply
become well-transit-served in 2012 – many were likely to have relatively good transit services
before then. As a proxy for overall transit service over and above simply L.A. Metro Rail, I use
TPAs and HQTAs in my analysis.
These transit proximity measures represent different level and frequency of transit service. Thus,
they have different levels of attractiveness to potential movers. For example, households who are
fully transit dependent or who prefer to use public transit only will place a higher preference on
living near an open rail station, while households who are casual users or who view transit access
as a “nice to have” will suffice with living in a TPA or HQTA. Similarly, I posit that prior
experience living near a particular level of transit service will inform future expectations about
what transit accessibility should be. This could in turn inform preferences future move locations
in regards to transit access.
To operationalize these measures, I determine whether a mover lived near one (or more) of these
four transit neighborhood types before and after moving. I accomplish this by geocoding a filer’s
pre-move and post-move filing location using filers’ 9-digit and 5-digit zip codes and Geolytics,
Inc. correspondence files to obtain geographic coordinates for the 9-digit and 5-digit zip code
centroids (see Chapter 1 for a lengthy explanation of the geocoding methodology). I define a
neighborhood proximate to a L.A. Metro rail station by drawing a 0.5 mile buffer in ArcGIS
around station locations obtained from Los Angeles Metro. I determine whether a station is open
by comparing the opening year to the move year. I obtain shapefiles of TPAs and HQTAs
directly from SCAG (SCAG 2012b, SCAG 2012c). I next spatially join household filing
locations before and after moves to transit proximity areas.
Table 41 provides the sample size for each transit proximity measure. About 4% of movers move
to open rail station neighborhoods in any given year, while over 5% of movers move from open
rail station neighborhoods. A similar pattern with slightly higher incidence is observed for all
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
170

L.A. Metro Rail station neighborhoods, whether stations are open or not. Nearly one quarter of
movers move to TPAs, but slightly more move from TPAs. Similarly, while over 35% of
households move to HQTAs, nearly 40% move out of HQTAs in any given year.

Descriptive Statistics for Mobility and Location Choices
I draw the Mobility Choice sample from the total population of Los Angeles County tax filers
from 1993-2013, restricted to the households who appear in the data for at least two consecutive
years (Chapter 1). This results in 81,776,830 household-years (Table 41). In about 21% of
household-years, the household moved at least one-half mile. Table 41 shows the characteristics
of the samples for Mobility and Location Choices. The Mobility Choice sample skews toward
the lower end of the income distribution. Nearly a quarter of households experience an income
increase of at least 25% in any given year and 15% experience an income decrease of at least
25%. Almost half of the sample is single, another 38% is married, and the remainder are Heads
of Household, a tax-data designation for non-married households who are able to claim
dependents. About 4% of households experienced a change in marital status. Most households
are aged 26-54 and data does not exist for about 5% of households. Households with non-
existent age data and households with household heads aged below 18 are excluded from most
analyses. 43% of the sample have at least one dependent and about 12% of households
experience a change in the number of dependents in any given year.
The Location Choice sample is the subset of the Mobility Choice sample of households who
moved in a given year. This sample size is 17,030,828 household-years. The Location Choice
sub-sample is relatively similar to the Mobility Choice sample. Slightly more households with
incomes below 30% of AMI move in any given year than their proportion in the overall
population, the reverse is true for households with incomes above 120% AMI move. Single and
Head of Household households are slightly over-represented among movers, while married
households are slightly under-represented. Households below age 40 are more likely to be
movers, while households above 40 are less likely to be movers. The number of dependents is
similar between the filer sample and the mover sample.
Table 41 also reports characteristics of a particular move for the Location Choice. Most
households do not move very far. 37% of movers move within their neighborhood (a distance of
0.5 – 2 miles). Another 28% move within their area of the city (2-10 miles). 13% move within
the metropolitan area (10-25 miles) and 10% to slightly more distant or outlying parts of the
metropolitan area (25-100 miles). About 10% move more than 100 miles, likely reflecting an
inter-city or interstate move.

Table 41. Descriptive Statistics for Mobility Choice and Location Choice Models
Variable Mobility Choice Location Choice
# of Obs. % # of Obs. %
Sample Size 81,886,830

17,030,828

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
171

Number of Movers (Move) 17,030,828 20.8%

Movers to Open Rail Station Neighborhoods

675,482 4.0%
Movers from Open Rail Station Neighborhoods

914,639 5.4%
Movers to Rail Station Neighborhood

1,116,189 6.6%
Movers from Rail Station Neighborhood

1,488,421 8.7%
Movers to TPA (Transit Priority Area)

4,200,845 24.7%
Movers from TPA (Transit Priority Area)

4,795,672 28.2%
Movers to HQTA (High Quality Transit Area)

6,044,761 35.5%
Movers from HQTA (High Quality Transit Area)

6,750,464 39.6%
Income Category

<30% of AMI 22,121,779 27.0% 5,084,685 29.9%
30-50% of AMI 14,280,006 17.4% 3,324,354 19.5%
50-80% of AMI 14,597,715 17.8% 3,226,195 18.9%
80-120% of AMI 11,687,826 14.3% 2,306,491 13.5%
>120% of AMI 19,199,504 23.4% 3,089,103 18.1%
Income Increase > 25% 19,733,914 24.1%

Income Decrease > 25% 11,983,478 14.6%

Marital Status

Single 36,431,765 44.5% 8,366,873 49.1%
Married 31,158,564 38.1% 5,204,884 30.6%
Head of Household 14,296,501 17.5% 3,459,071 20.3%
Change in Marital Status 3,560,658 4.3%

Age

age 18-25 10,250,303 12.5% 2,921,661 17.2%
age 26-40 26,405,333 32.2% 6,638,761 39.0%
age 41-54 22,152,774 27.1% 3,723,805 21.9%
age 55-70 12,716,105 15.5% 1,706,744 10.0%
age 71+ 4,987,811 6.1% 587,478 3.4%
no age info provided 4,289,575 5.2% 1,336,768 7.8%
age below 18 806,811 1.0% 115,611 0.7%
Age Cohort

Millenial

2,338,651 13.7%
Gen X

6,390,004 37.5%
Baby Boom

5,121,033 30.1%
Silent

1,370,586 8.0%
Greatest

465,679 2.7%
Gen Z

8,107 0.0%
no age info provided

1,336,768 7.8%
Number of Dependents

No dependents 46,668,578 57.0% 9,834,498 57.7%
1 Dependent 14,850,641 18.1% 3,189,058 18.7%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
172

2 Dependents 12,557,323 15.3% 2,499,715 14.7%
3+ Dependents 7,810,288 9.5% 1,507,557 8.9%
Change in Number of Dependents 9,716,205 11.9%
Distance Moved

0.5-2 miles

6,348,143 37.3%
2-5 miles

3,030,466 17.8%
5-10 miles

2,134,298 12.5%
10-25 miles

2,148,247 12.6%
25-100 miles

1,683,851 9.9%
>100 miles

1,685,823 9.9%
Source: Author calculations on FTB data

Mobility Choice Results
Results from the first stage mobility model conform to theoretical expectations and mirror
previous findings. Households with dependents are less likely to move than those without
dependents; the more dependents, the lower the probability of moving is (see Table 42), in line
with South & Crowder (1998) Ioannides & Kan (1996), Clark & Davies Withers (1999), and Lee
(2014)
37
. Younger households under 40 years old have 55% higher odds of moving than middle-
aged households, while households older than 55 have at least 30% lower odds of moving, in line
with prior studies (Lee, 2014; South & Crowder, 1998; van der Vliest et al., 2002; Ioannides &
Kan, 1996; van Ommeren et al., 1999; Clark & Huang, 2003; Lee et al., 1994; Weinberg, 1979;
Varady, 1983). Married households are slightly less likely to move, though un-married
households with dependents (‘Head of Household’ category) are slightly more likely to move
than single households. Lower (30-50% AMI) and lower-middle (50-80% AMI) income
households have about 5% higher odds of moving than median income households; in contrast
lowest-income (<30% AMI) and higher-income (>120% AMI) are slightly less likely to move.
The finding on the lowest-income group is contrary to expectations. Perhaps some of these
households reside in deed-restricted or rent-stabilized affordable housing or can not overcome
high housing search costs, keeping their mobility relatively lower than median income
households. Changes in income, marital status, and number of dependents consistently increase a
household’s probability of moving, consistent with the findings in Chapter 2 and with theoretical
predictions that lifecycle events that lead to changes in demand for housing space consumed
(Rossi, 1955; Sabagh, van Arsdol, & Butler 1969).
Using these regression estimates, I generate a predicted probability of moving for each
household-year controlling for the independent variables in Table 42. These predicted values are
used as weights in the Location Choice and Mover Type models below.

37
Though Weinberg (1979) found that household size is positively correlated with residential mobility and Gabriel
& Painter (2008) found that number of children in the household increases move likelihood for White households
but not for Black or Latino households.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
173

Table 42. Mobility Choice Logit Regression Results

Odds
Ratio
Robust
Std. Err.
z P>z 95% Confidence
Interval
Δ in Number of
Dependents
1.17 0.00 82.25 0.00 1.16 1.17
Δ Marital Status 1.37 0.00 89.72 0.00 1.36 1.38
+25 % Δ in Income 1.22 0.00 62.68 0.00 1.21 1.23
-25% Δ in Income 1.28 0.00 81.63 0.00 1.28 1.29
Income Level (80-120% AMI is reference)
<30% of AMI 0.95 0.01 -5.25 0.00 0.94 0.97
30-50% of AMI 1.05 0.01 6.86 0.00 1.03 1.06
50-80% of AMI 1.05 0.00 11.97 0.00 1.04 1.06
>120% of AMI 0.89 0.01 -15.82 0.00 0.88 0.90
Number of Dependents (0 is reference)
1 Dependent 0.92 0.01 -12.54 0.00 0.90 0.93
2 Dependents 0.87 0.01 -14.09 0.00 0.86 0.89
3+ Dependents 0.84 0.01 -15.85 0.00 0.82 0.86
Marital Status (Single is reference)
Married 0.87 0.01 -21.98 0.00 0.86 0.88
Head of Household 1.12 0.00 32.09 0.00 1.11 1.12
Age Category (age 41-54 is reference)
age 18-25 1.68 0.02 38.39 0.00 1.63 1.72
age 26-40 1.57 0.01 59.85 0.00 1.55 1.59
age 55-70 0.76 0.00 -53.31 0.00 0.75 0.76
age 71+ 0.64 0.01 -44.38 0.00 0.63 0.65
Constant 0.31 0.01 -51.94 0.00 0.30 0.33
Year Fixed Effects: Yes
# of Observations = 76,512,326
Wald chi2(36) = 53946.58
Prob > chi2 = 0.0000
Log pseudolikelihood = -37566014
Pseudo R2 = 0.0285
Note. Dependent variable: Move or Not Move. 5,096,386 observations were dropped because they had no
age information or filer age was below 18. Std. Err. adjusted for 7,814 origin 5-digit zip code cluster
Source: Author analysis of FTB data

Location Choice Results
The Location Choice results indicate that living near rail is a strong predictor of moving to rail.
Regardless of the specification, odds ratios on the variable of interest OriginOpenRail indicate a
4.5-5.5 times higher odds of moving to neighborhood with an open L.A. Metro rail station if
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
174

your prior residence was in a different neighborhood with an open L.A. Metro rail station (Table
3). This finding stands independent of weighting by the Mobility Choice, controlling for year
fixed effects, excluding non-existing age data and household heads below eighteen years old, or
clustering standard errors at the origin 5-digit zip code. When all of these measures are taken
(Table 43, Model 6), living near rail in one period still strongly begets living near rail in the next
period.
These findings stand even when controlling for move distance. Since both the rail system and the
total population of households are denser near the central parts of Los Angeles County, I would
expect distance to be correlated with moving to rail. I find an inverse relationship between the
likelihood of moving to an open L.A. Metro rail station and move distance. Generally, the farther
a household moves, the lower odds that they move to rail, relative to households moving 0.5-2
miles. However, this effect is not particularly strong for households moving 2-5 miles and not
statistically significant (despite a very large sample size) for households moving 5-25 miles. This
may indicate that distance only slightly impacts the Location Choice of moving to rail:
households who move to rail do so routinely from distances of up to 25 miles in Los Angeles
County.
I also qualify the types of people who move to an open L.A. Metro rail station by reflecting on
the results of other covariates (Table 43, Model 6). The probability of moving toward an open
L.A. Metro rail station is inversely related to income: compared to median income households,
lowest income (<30% AMI) households have 70% higher odds of moving into rail transit-areas;
lower-income (30-50% AMI) have 55% higher odds and lower-middle (50-80% AMI) have 25%
higher odds, while higher-income (>120% AMI) have 25% lower odds. Household income
seems to play a stronger role in determining Location Choice compared to its role in the Mobility
Choice above.
Married households have slightly lower odds of moving to an open L.A. Metro rail station. This
is not surprising since married households are more likely to be homeowners, and homeowners
move less frequently than renters. There is no way to identify homeowners in the FTB data, but
the married variable may be proxying for housing tenure. The number of dependents does not
seem to influence Location Choice one way or the other. Younger adult households (age 26-40)
have 14% higher odds of moving to rail than middle-aged households (age 41-54), while older
middle-age households (age 55-70) have 14% lower odds of moving to rail and elderly
households (age 71+) have even lower odds of moving to rail. Interestingly, young and student-
aged households (age 18-25) do not seem to influence Location Choice relative to middle-aged
households.

Table 43. Location Choice Weighted Logit Regression Results, displayed as Odds Ratios
Model
(1) (2) (3) (4) (5) (6)
Variables
OriginOpenRail
5.437*** 5.037*** 5.437*** 5.037*** 5.017*** 4.599***

-534.22 -482.91 -10.43 -10.09 -10.13 -9.78
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
175

Income Category (80-120% AMI is reference)
<30% of AMI
1.724*** 1.650*** 1.724*** 1.650*** 1.758*** 1.699***

-116.64 -101.58 -15.24 -14.47 -15.83 -15.53
30-50% of AMI
1.578*** 1.529*** 1.578*** 1.529*** 1.590*** 1.548***

-93.69 -83.06 -14.89 -14.39 -15.42 -15.21
50-80% of AMI
1.256*** 1.234*** 1.256*** 1.234*** 1.263*** 1.243***

-45.42 -39.92 -11.09 -10.73 -11.67 -11.57
>120% of AMI
0.765*** 0.773*** 0.765*** 0.773*** 0.744*** 0.750***

(-45.53) (-41.55) (-6.30) (-6.65) (-7.08) (-7.59)
Number of Dependents (0 is reference)
1 Dependent
1.048*** 1.044*** 1.048 1.044 1.009 0.997

-9.88 -8.28 -1.36 -1.22 -0.27 (-0.09)
2 Dependents
1.080*** 1.084*** 1.08 1.084 1.024 1.015

-15.21 -14.65 -1.58 -1.61 -0.49 -0.3
3+ Dependents
1.173*** 1.169*** 1.173* 1.169* 1.107 1.088

-28.33 -25.54 -2.22 -2.16 -1.4 -1.15
Marital Status (Single is reference)
Married
0.729*** 0.712*** 0.729*** 0.712*** 0.790*** 0.784***

(-71.71) (-68.62) (-10.94) (-10.99) (-8.47) (-8.20)
Head of Household
0.995 0.973*** 0.995 0.973* 1.066*** 1.055***

(-1.11) (-5.15) (-0.39) (-2.08) -5.17 -4.59
Age Category (age 41-54 is reference)
age 18-25
1.023*** 1.031*** 1.023 1.031 1.021 1.022

-5.49 -7.04 -0.98 -1.33 -0.89 -0.95
age 26-40
1.118*** 1.114*** 1.118*** 1.114*** 1.142*** 1.137***

-32.72 -31.51 -8.22 -8.25 -9.57 -9.55
age 55-70
0.906*** 0.893*** 0.906*** 0.893*** 0.874*** 0.860***

(-18.66) (-21.08) (-7.18) (-8.05) (-9.34) (-10.26)
age 71+
0.603*** 0.589*** 0.603*** 0.589*** 0.593*** 0.578***

(-52.26) (-54.38) (-10.12) (-10.64) (-10.24) (-10.80)
No age
information
0.628*** 0.628*** 1.219***

(-73.04) (-8.02) -9.27
Age below 18
0.510*** 0.510*** 0.587***

(-36.66)

(-9.79)

(-7.73)

Distance Category (0.5-2 miles is reference)
2-5 miles
0.728*** 0.735*** 0.728 0.735 0.717* 0.723*

(-89.86) (-82.31) (-1.93) (-1.88) (-2.04) (-1.98)
5-10 miles
0.730*** 0.735*** 0.73 0.735 0.713 0.716

(-78.83) (-71.91) (-1.60) (-1.58) (-1.73) (-1.72)
10-25 miles
0.725*** 0.725*** 0.725 0.725 0.706 0.704

(-79.69) (-74.55) (-1.72) (-1.73) (-1.86) (-1.89)
25-100 miles
0.422*** 0.433*** 0.422*** 0.433*** 0.407*** 0.415***

(-151.81) (-140.03) (-4.67) (-4.56) (-4.88) (-4.80)
>100 miles
0.322*** 0.348*** 0.322*** 0.348*** 0.299*** 0.324***
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
176

(-173.71) (-153.85) (-6.02) (-5.66) (-6.40) (-6.06)

Weights
No Yes No Yes No Yes
Year Fixed Effects
No No No No Yes Yes
Age Categories
All N/A and
<18
excluded
All N/A and
<18
excluded
All N/A and
<18
excluded
Standard Errors
Robust Robust Clustered at 5-digit zip code
Pseudo R
2

0.071 0.062 0.071 0.062 0.083 0.075
AIC
5,278,491 1,168,728 5,278,491 1,168,728 5,212,182 1,151,953
BIC
5,278,813 1,169,020 5,278,813 1,169,020 5,212,783 1,152,520
Wald(chi2)
405403 352090 4146 3137 6548 6719
Degrees of Freedom
21 19 21 19 40 38
Weighted Log-
Likelihood
-2639223 -584344 -2639223 -584344 -2606050 -575937
Number of
Observations
17,030,828 15,578,449 17,030,828 15,578,449 17,030,828 15,578,449
Note. Dependent variable = 1 if Households moves to Neighborhood with Open L.A. Metro Rail Station
or 0 if they move anywhere else.
*<0.05 ** p<0.01 *** p<0.001
Source: Author analysis of FTB data

Location Choice Results for Alternative Transit Definitions
The above finding that living near transit begets more living near transit applies to a very narrow
definition of transit-proximity: more from/to a neighborhood with an open L.A. Metro Rail
station, which applies to 5.4% (from) / 4.0% (to) of all Los Angeles County movers in any year. I
relax the definition of transit proximity for both origin and destination neighborhoods to see
whether the results stand more generally.
I first relax the definition of transit proximity of the origin neighborhood to an L.A. Metro rail
neighborhood in any year (whether station is currently open or will be in the future). This now
covers 8.7% of all household-years (up from 5.4% above). Use of this alternative still reveals
that the odds of moving to a neighborhood with an open L.A. Metro rail station are 4 times as
high for households originating in these neighborhoods (Table 44, Model 7). I next relax the
definition of transit proximity for the destination neighborhood in a similar vein: any L.A. Metro
rail neighborhood regardless of station opening year. This increases the incidence to 6.6% of all
household-years (up from 4%). Even with this relaxed definition, movers from neighborhoods
where L.A. Metro stations are currently have 3.2 times the odds of moving to an L.A. Metro rail-
proximate neighborhood whether the station is open or not, compared to all other movers (Table
44, Model 8). I next combine these two relaxed definitions, such that both origin and destination
neighborhoods are L.A. Metro rail station neighborhoods, but where stations are not necessarily
open at the time of moving. In this case (Table 44, Model 9), movers still have over 4 times the
odds of moving to transit neighborhoods. Household characteristic and move distance patterns
mirror the Location Choice results above. In all, these temporal transit-proximity variants
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
177

indicate that moves to neighborhoods with rail transit today or in the future is strongly predicted
by living in neighborhoods ready for rail, whether they currently open or not.
I also test alternative definitions of transit, more expansive than L.A. Metro rail stations. I model
whether households moving to a TPA are more likely to have lived in a TPA. Results from
Model 10 in Table 44 indicate that current residents in TPAs have 3.3 times higher odds of
moving to TPAs. TPAs include neighborhoods within 0.5 miles of major transit route
intersections and cover on average 28% of movers’ origin neighborhoods and 25% of movers’
destination neighborhoods. Hence, movers to neighborhoods near major transit route
intersections are highly likely to live near such intersections in the past. I also model whether
households moving to an HQTA are more likely to have lived in an HTQA. Results indicate that
current HQTA residents have 4.9 times odds of moving to an HQTA compared to non-HQTA
residents (Table 44, Model 11). HQTAs include neighborhoods within 0.5 miles of transit stops
serviced at least every 15 minutes or more during peak and commute periods and cover on
average 40% of movers’ origin neighborhoods and 36% of movers’ destination neighborhoods.
As with TPAs, movers to neighborhoods with higher-frequency transit are more likely to live in
similarly transit-serviced neighborhoods in the past.

Table 44. Alternative Transit Neighborhood Definitions Location Choice Weighted Logit Regression
Results, displayed as Odds Ratios
Model (7) (8) (9) (10) (11)
Dependent Variable
Move to
Neighborhood
with Open
L.A. Metro
Rail
Move to
Neighborhood
that has or
will have L.A.
Metro Rail
Move to
Neighborhood
that has or
will have L.A.
Metro Rail
Move to
Transit
Priority
Area
(TPA)
Move to High
Quality
Transit Area
(HQTA)
Current Neighborhood
has Open L.A. Metro
Rail Station
3.181***

-7.27
Current Neighborhood
has or will have L.A.
Metro Rail Station
4.066*** 4.008***
-9.5 -11.13
Current Neighborhood
is Transit Priority Area
(TPA)
3.327***
-13.98
Current Neighborhood
is a High Quality
Transit Area (HQTA)
4.911***
-17.72
Income Category (80-120% AMI is reference)
<30% of AMI 1.694*** 1.541*** 1.514*** 1.431*** 1.438***

-14.59 -10.38 -11.57 -14.39 -15.95
30-50% of AMI 1.545*** 1.431*** 1.409*** 1.353*** 1.349***
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
178

-14.59 -10.79 -11.86 -15.35 -16.73
50-80% of AMI 1.240*** 1.207*** 1.197*** 1.186*** 1.179***

-11.32 -10.29 -11.12 -16.76 -17.19
>120% of AMI 0.750*** 0.800*** 0.808*** 0.808*** 0.874***

(-7.55) (-6.33) (-6.91) (-9.74) (-6.16)
Number of Dependents (0 is reference)
1 Dependent 1.004 0.928* 0.932** 0.969 0.962
-0.1 (-2.34) (-2.59) (-1.67) (-1.93)
2 Dependents 1.025 0.915* 0.921* 0.954 0.944*
-0.48 (-2.02) (-2.16) (-1.83) (-2.16)
3+ Dependents 1.101 0.947 0.954 0.973 0.964

-1.28 (-0.86) (-0.85) (-0.77) (-1.01)
Marital Status (Single is reference)
Married 0.790*** 0.757*** 0.766*** 0.767*** 0.754***

(-7.69) (-11.45) (-11.44) (-18.49) (-20.55)
Head of Household 1.054*** 1.052*** 1.046*** 1.069*** 1.066***

-4.14 -4.53 -4.42 -6.42 -5.94
Age Category (age 41-54 is reference)
age 18-25 1.023 1.026 1.029 1.023* 1.044***
-0.97 -1.35 -1.65 -2.05 -4.22
age 26-40 1.132*** 1.145*** 1.134*** 1.115*** 1.115***
-9.00 -10.18 -9.99 -11.24 -12.36
age 55-70 0.863*** 0.854*** 0.862*** 0.888*** 0.897***
(-9.86) (-9.45) (-9.73) (-11.16) (-9.84)
age 71+ 0.580*** 0.649*** 0.664*** 0.710*** 0.766***

(-10.58) (-8.03) (-8.15) (-12.34) (-9.63)
Distance Category (0.5-2 miles is reference)
2-5 miles 0.738 0.764* 0.775 0.805** 0.760***
(-1.89) (-1.96) (-1.83) (-2.65) (-3.81)
5-10 miles 0.731 0.735* 0.744* 0.709*** 0.695***
(-1.63) (-2.14) (-2.05) (-3.39) (-4.39)
10-25 miles 0.721 0.703* 0.713* 0.620*** 0.545***
(-1.78) (-2.33) (-2.22) (-4.39) (-5.62)
25-100 miles 0.430*** 0.406*** 0.427*** 0.413*** 0.324***

(-4.67) (-5.76) (-5.47) (-7.85) (-8.85)
>100 miles 0.331*** 0.350*** 0.361*** 0.264*** 0.225***

(-6.02) (-6.56) (-6.32) (-9.77) (-10.33)
Weights: Yes
Year Fixed Effects: Yes
Age Categories: N/A and <18 years old excluded
Standard Errors: Clustered at origin 5-digit Zip Code
Pseudo R
2
0.078 0.042 0.062 0.095 0.155
AIC 1,148,810 1,698,446 1,663,722 3,641,952 3,918,644
BIC 1,149,378 1,699,014 1,664,290 3,642,520 3,919,212
Wald(chi2) 6998.04 4356.993 5188.53 4015.972 4370.404
Degrees of Freedom 38 38 38 38 38
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
179

Weighted Log-
Likelihood
-574,366 -849,184 -831,822 -1,820,937 -1,959,283
Number of
Observations
15,578,449 15,578,449 15,578,449 15,578,449 15,578,449
*<0.05 ** p<0.01 *** p<0.001
Source: Author analysis of FTB data

Mover Types and Transit Neighborhoods
Results from the above models broadly indicate that living near transit begets more living near
transit, regardless of the definition of transit or of whether that transit was even open at the time,
in either the destination and/or origin neighborhood. The above models focus only on predictors
of the binary Location Choice decision of whether to move to transit or not to transit. However,
those analyses only focus on a small proportion of households moving from transit
neighborhoods to other transit neighborhoods. There are three other types of possible movers
(Table 45): movers from non-transit neighborhoods to other non-transit neighborhoods, movers
from non-transit neighborhoods to transit neighborhoods, and movers from transit neighborhoods
to non-transit neighborhoods. These other movers make up the bulk of all movers in Los Angeles
County, regardless of how transit is defined (Table 45). In fact, the majority of moves in any
transit definition are from non-transit neighborhoods to non-transit neighborhoods. This reveals
that this study has so far focused on a relatively narrow sub-sample of movers. In the remainder
of this paper, I explore the growth and composition of mover types county-wide and the
characteristics of these movers types by the different transit definitions. This section will shed
more light on the overall mover flow within Los Angeles County and will enable me to draw
conclusions on which types of households move toward transit-proximate neighborhoods and
which move away from them.
Table 45. Mover Types by Transit Definition
Transit Definition Type
→

Mover Type ↓
Neighborhood
with Open
L.A. Metro
Rail Station
Neighborhood
with L.A. Metro
Rail Station Now
or in the Future
Transit
Priority
Area
High-
Quality
Transit
Area
Non-transit → Non-transit 15,596,507 14,735,992 10,164,836 8,245,227
Non-transit → Transit 519,682 806,415 2,070,320 2,035,137
Transit → Non-transit 758,839 1,178,647 2,665,147 2,740,840
Transit → Transit 155,800 309,774 2,130,525 4,009,624
All Movers 17,030,828 17,030,828 17,030,828 17,030,828

Proportion of All Movers
Non-transit → Non-transit 92% 87% 60% 48%
Non-transit → Transit 3% 5% 12% 12%
Transit → Non-transit 4% 7% 16% 16%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
180

Transit → Transit 0.9% 1.8% 13% 24%

Proportion of Movers from Non-Transit Neighborhoods
Non-transit → Non-transit 97% 95% 83% 80%
Non-transit → Transit 3% 5% 17% 20%
Proportion of Movers from Transit Neighborhoods
Transit → Non-transit 83% 79% 56% 41%
Transit → Transit 17% 21% 44% 59%
Source: Author calculations of FTB data

Mover Types Over Time
Mover types are a dynamic concept since the number of movers of each type changes from year
to year. How have mover types changed over time? Are mover types sensitive to transit
proximity definition?
92% of all movers moved from non-transit to non-transit, using the open L.A. Metro rail station
definition (Table 45), though this proportion has decreased over the study time period.
Continuing with the open L.A. Metro rail definition of transit, the proportion of movers from rail
to non-rail and from non-rail to rail have grown from 2% and 1% to 6% and 5% of all movers, an
increase of 2.5 and 1.4 times over my 20-year study period (Figure 25). The proportion of rail to
rail moves also increased from 0.4% to 1.5%, possibly indicating increased preference for living
near transit for households who already live near transit or the reflecting the expansion of the
system (more rail neighborhoods total may mean more rail to rail moves).

Figure 25. Proportion of Mover Type by Year for Open L.A. Metro Rail Station Neighborhoods
Note. Non-transit to non-transit mover type not shown
Source: Author calculations on FTB data
0%
1%
2%
3%
4%
5%
6%
7%
8%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Open L.A. Metro Rail Station Neighborhood Mover Types
non-rail to rail (with station open) rail (with station open) to non-rail
rail (with station open) to rail (with station open)
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
181

Mover type proportions are more stable over time when using the relaxed definition of transit
proximity, inclusive of neighborhoods near L.A. Metro rail stations whether or not the stations
are open. An average of 87% of movers are non-rail to non-rail. The proportion of non-rail to
rail grew slightly from 4% to over 5%, while the proportion of rail to non-rail and rail to rail
remained largely flat from 1993-2012 (Figure 26). This indicates a slow in-migration of
households to transit neighborhoods regardless of whether transit service has yet begun, without
a corresponding increase in outflow from these neighborhoods.

Figure 26. Proportion of Mover Type by Year for L.A. Metro Rail Station Neighborhoods regardless of
whether station is open
Note. Non-transit to non-transit mover type not shown
Source: Author calculations on FTB data

Expanding the transit definition to TPAs and HQTAs, I find that the proportions of mover types
remain stable over the study time period (Figures 27 and 28). Note that the proportion of TPA to
TPA movers is on average 13% (Figure 27), much higher than those in Figures A and B, since
TPAs cover a larger territory of the County. Similarly, the proportion of HQTA to HQTA
movers is about 24% due to the more expansive HQTA geographical catchment (Figure 28).
0%
1%
2%
3%
4%
5%
6%
7%
8%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
L.A. Metro Rail Station Neighborhood Mover Types
non-rail to rail rail to non-rail rail to rail
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
182

Figure 27. Proportion of Mover Type by Year for TPA Neighborhoods
Note. Non-transit to non-transit mover type not shown
Source: Author calculations on FTB data

Figure 28. Proportion of Mover Type by Year for HQTA Neighborhoods
Note. Non-transit to non-transit mover type not shown
Source: Author calculations on FTB data

0%
5%
10%
15%
20%
25%
30%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Transit Priority Area (TPA) Mover Types
non-TPA to TPA TPA to non-TPA TPA to TPA
0%
5%
10%
15%
20%
25%
30%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
High Quality Transit Area (HQTA) Mover Types
non-HQTA to HQTA HQTA to non-HQTA HQTA to HQTA
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
183

Mover Type Characteristics
Having outlined the definition and proportional growth of each mover type relative to each
transit type, I next examine which household-specific characteristics are associated with which
mover type. What is the type of household who moves to / away from transit neighborhoods in
Los Angeles County?

Mover Type Model Setup
I analyze what type of characteristics describe each mover types using a multinomial logit
(MNL) model, similar in setup to Clark et al. (2006)’s model of residential choice. I define each
mover type as a potential ‘choice’ based on movers’ proximity to rail in origin and destination
locations for one of four mover types j:
1) moves from a non-transit neighborhood to a non-transit neighborhood,
2) moves from a non-transit neighborhood to a transit neighborhood,
3) moves from a transit neighborhood to a non-transit neighborhood, and
4) moves from a transit neighborhood to another transit neighborhood.
The MNL is an efficient and consistent estimator of the probability of being a mover type, has a
closed-form solution, and is straightforward in interpretation. The MNL is appropriate rather
than the conditional logit since regressors do not vary over alternatives (Cameron & Trivedi,
2005, p.500). Just like the Location Choice models above, I weight the MNL by the probability
of moving derived from the Mobility Choice. I define the probability p that mover i in year t is
one of the four mover types j as
(5) 𝑀𝑜𝑣𝑒𝑟 𝑇𝑦𝑝𝑒 𝑀𝑁𝐿 𝑚𝑜𝑑𝑒𝑙 : 𝑝 𝑖𝑗𝑡 =
𝑒 (𝑿 𝑖𝑡
′
𝛽 𝑗 +𝜇 𝑡 +𝘀 𝑧 )
∑ 𝑒 (𝑿 𝑖𝑡
′
𝛽 𝑗 +𝜇 𝑡 +𝘀 𝑧 ) 4
𝑗 =1

Adding in the weights, I specify the following weighted MNL model:
(6) 𝑄 𝑊𝑀𝑁𝐿 (𝜽 ) = ∑
𝑄 𝑖 𝐻 𝑖 𝑖 ln 𝑓 (𝑝 𝑖𝑗𝑡
|𝑿 𝑖𝑡
, 𝜽 ), where
𝑄 𝑖 𝐻 𝑖 is the probability that household i is a mover in year t.
Since the sum of the probabilities of all types are equal to 1 (i.e., ∑ 𝑝 𝑖𝑗𝑡 4
𝑗 =1
= 1), I restrict the
model so that the coefficients on Mover Type 1 are equal to 0 to ensure proper identification
(i.e., 𝜷 𝟏 = 0). Mover Type 1 is the intuitive baseline group, since it represents movers from non-
transit groups to non-transit groups and makes up the majority of moves. In the Mover Type
model, Xit is a vector of independent variables including mover i’s age, income, marital status,
and number of dependents in year t, as well as the distance they moved. As in the prior models,
µt is the year fixed effect, and εz is the error term, clustered by the 5-digit zip code in which the
household currently resides. I estimate coefficients as relative risk ratios (similar to odds ratios
for logit) for easier interpretation of MNL results. The relative risk ratio of choosing alternative j
rather than alternative 1 is
𝑃𝑟 [𝑦 𝑖 =𝑗 ]
𝑃𝑟 [𝑦 𝑖 =1]
= 𝑒 (𝑿 𝑖𝑡
′
𝛽 𝑗 +𝜇 𝑡 +𝘀 𝑧 )
, so the exponentiated coefficient 𝑒 𝛽 𝑗 gives the
proportional change in the relative risk given a 1-unit change in the regressor (Cameron &
Trivedi, 2005, p.503).
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
184

The MNL model compares a set of binary choice models between any two pairs of choices,
unaffected by characteristics of alternatives other than the pair currently under consideration. As
a result, the MNL model may impose the assumption of the independence of irrelevant
alternatives (IIA), with an error structure than is independent and identically distributed (IID)
with type I extreme value distribution (Cameron & Trivedi, p.503). Other multi-stage mobility
studies use alternative approaches to get around the IID assumption, including using a
multinomial probit (MNP) (Ioannides & Kan, 1996), a nested multinomial logit (MNML)
(Gabriel & Painter, 2008; Cervero & Duncan, 2008), or a cross-nested logit model (Yang, Zheng
& Zhu, 2013), while Lee (2014) uses the MNL model outright. The MNP model is often
imprecisely estimated in models with regressors that do not vary with the alternative (as is this
case) (Keane, 1992). The MNML model would fit well in this case, but is computationally
intractable given the sample size (17,030,828) * mover types (4) * regressors (38). Hence, I
follow Lee (2014) and use the MNL model. Weighting by the probability of moving actually
moves my model closer to a NMNL, partially allaying IIA concerns.

Mover Type Model Results
The weighted MNL results reveal interesting differences in the households characteristics of
mover types for movers into and out of open L.A. Metro rail stations. Lowest (<30% AMI),
lower (30-50% AMI) and lower-middle (50-80% AMI) income households have higher odds of
moving to open L.A. Metro rail neighborhoods from non-rail neighborhoods and even higher
odds when moving from open L.A. Metro rail neighborhoods (Table 46). In contrast, these
households are not as likely to move from rail to non-rail. Higher-income households (>120%
AMI) have lower odds of moving to rail and from rail in any of the cases.
Married households have lower odds of moving from non-rail to open L.A. Metro rail stations
compared to single households. Similar to the Location Choice model results, the number of
dependents is similar across mover types. Households aged 26-40 are more likely to be living
currently or in the future near rail, but there is little differences in their odds of moving to / from
rail. Similarly, households aged above 55 are less likely to be living currently or in the future
near rail, but there is little differences in their odds of moving to / from rail.
Results on distance moved also indicated interesting splits by mover type. Households moving
from non-rail to open L.A. Metro rail stations come equally from as far as 25 miles. In contrast,
moving from rail to non-rail generally indicates a move of 5-25 miles. For households already
living near an open L.A. Metro rail station, moving to another open L.A. Metro rail station most
often indicates a move of less than 2 miles. In fact larger distance moves are much less
associated with open rail to open rail mover types.
Mover type characteristics are quite consistent over transit definitions. Running the weighted
MNL for alternative transit definitions reveals no differences in direction of effect, and few
differences in coefficient magnitude and statistical significance (see Appendix 3-C). This
stability is remarkable given that HQTA definition cover nearly 10 times the number of movers
as the Open L.A. Metro definition.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
185

Table 46. Mover Type Weighted MNL Model Results displayed as Relative Risk Ratios for Open L.A.
Metro Rail transit definition
Location Choices Non-Rail to Open
Rail
Open Rail to Non-Rail Open Rail to Open
Rail
Income Category (80-120% AMI is reference)
<30% of AMI 1.74*** 1.34 2.10***
30-50% of AMI 1.58*** 1.33* 1.86***
50-80% of AMI 1.26*** 1.14* 1.34***
>120% of AMI 0.73*** 0.75*** 0.67**
Number of Dependents (0 is reference)
1 Dependent 1.01 0.96 0.94
2 Dependents 1.03 0.96 0.96
3+ Dependents 1.11 0.99 1.02
Marital Status (Single is reference)
Married 0.73*** 0.82* 0.86
Head of Household 1.06*** 1.13*** 1.13**
Age Category (age 41-54 is reference)
age 18-25 1.04 1.03 0.96
age 26-40 1.16*** 1.20*** 1.23***
age 55-70 0.85*** 0.83*** 0.80***
age 71+ 0.56*** 0.54*** 0.42***
Distance Category (0.5-2 miles is reference)
2-5 miles 0.92 1.31 0.42***
5-10 miles 1.04 1.49*** 0.22***
10-25 miles 1.04 1.49** 0.17***
25-100 miles 0.60** 0.89 0.01***
>100 miles 0.49*** 1.02 0.00***
Constant 0.01*** 0.02*** 0.00***
Weights: Yes
Year Fixed Effects: Yes
Age Categories: N/A and <18 years old excluded
Standard Errors: Clustered at origin 5-digit Zip Code
Pseudo R
2
: 0.0392
Wald(chi2): 368,000
Degrees of Freedom: 111
Prob>chi2: 0.000
Log Pseudo- likelihood: -1,320,007
Number of Observations: 15,578,449
Note. Baseline category: Mover Type 1: non-rail to non-rail
*<0.05 ** p<0.01 *** p<0.001
Source: Author calculations of FTB data
In addition to effects by age, it is possible that a birth year cohort effect also influences transit-
relevant mover type. Recent research from Canada has indicated that individuals born between
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
186

1980-1996 (known as Millenials) may be more likely to use transit that previous generations at
the same life stage (Newbold & Scott, 2017; Grimsrud & El-Geneidy, 2014). Additional U.S.
research has found that transit usage is increasing among youth born after 1980 (Millenials and
Generation Z) and auto ownership and usage is declining (Blumenberg et al., 2012; Brown et al.,
2016). I test whether these birth year cohort effects also affect mover type.
To test this, I substitute the age category variable with a birth year cohort variable in the
weighted MNL model of mover types (Equation 6). I use the filer’s year of birth to define birth
year cutoffs correspond to generations as conceptualized by the Pew Research Center (2015).
The birth year cutoffs are 1) pre-1928 (Greatest Generation), 2) 1929-1945 (Silent Generation),
3) 1946-1964 (Baby Boom), 4) 1965-1980 (Generation X), 5) 1980-1996 (Millenials), and 6)
after 1980 (Generation Z).
Birth year cohort results do not reveal any stark differences from age category results (Table 47).
Households born between 1965 and 1996 (Millenials and Generation X) are equally likely to
move from / to rail. Households born before 1965 are less likely to move to / from rail.
Table 47. Cohort Age Effects of Mover Type Weighted MNL Model Results displayed as Relative Risk
Ratios for Open L.A. Metro Rail transit definition
Location Choices Non-Rail to Open
Rail
Open Rail to Non-Rail Open Rail to Open
Rail
Birth Year Cohort (1965-1980 is reference)
1980-1995 1.01 0.96 0.93
1946-1964 0.91*** 0.88*** 0.88***
1928-1945 0.73*** 0.67*** 0.64***
Pre-1928 0.50*** 0.48*** 0.37***
Other Variables Included: Distance Category, Marital Status, Number of Dependents, Income Category
Weights: Yes
Year Fixed Effects: Yes
Age Categories: N/A and <18 years old excluded
Standard Errors: Clustered at origin 5-digit Zip Code
Pseudo R
2
: 0.0385
Wald(chi2): 366,429
Degrees of Freedom: 111
Prob>chi2: 0.000
Log Pseudo- likelihood: -1,320,873
Number of Observations: 15,578,449
Note. Baseline category: Mover Type 1: non-rail to non-rail. Non-age covariate results suppressed for
clarity.
*<0.05 ** p<0.01 *** p<0.001
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
187

Discussion and Conclusion
This paper set out to explore the concept of transit exposure in terms of population flows and
understand whether households sort into neighborhoods along a transit access dimension. Using
a rich longitudinal tax filer database from the California Franchise Tax Board (FTB) available
from 1993 to 2013, I tested whether new transit lines change households’ move destination
choices, whether prior transit exposure increases the likelihood of moving to transit-proximate
neighborhoods, and what kinds of households move into and out of transit-proximate
neighborhoods.
Combining mapping, descriptive statistics, and statistical analysis, this exploration has yielded
intriguing findings. First, introducing a new rail transit line does not appear to change
neighborhood to neighborhood spatial move patterns. Second, living near transit begets more
living near transit. Living near transit is an extremely strong predictor of moving near transit in
the future across transit models. Third, the majority of moves in Los Angeles County originate
outside of transit-proximate neighborhoods and end outside of transit-proximate neighborhoods,
but this is changing as the L.A. Metro rail transit system grows. Fourth, a typical household
moving into a transit-proximate neighborhood is younger, has income below the County median,
and is non-married.
Taking stock of these findings, I conclude that on a household level, the phenomenon of sorting
into neighborhoods based on transit access is present in Los Angeles County. In aggregate,
however, the introduction of more rail lines and stations has not majorly shifted neighborhood to
neighborhood population flows. That said, Los Angeles County’s overall transit exposure is
increasing as the system expands, though this expansion will be limited until the number of
housing units near transit areas increases dramatically. Population shifts also take time to full
manifest and possibly we will see more pronounced differences in move patterns in the coming
decades.
Previous exposure to transit is the strongest predictor of choosing to live near transit in my
models. Transit access is the only environment or community characteristic in my models and
may be serving as a proxy for other neighborhood characteristics correlated with transit access.
The strength of the association between previous exposure to transit and choosing to live near
transit in the future reveals potential preferences for living in transit-proximate neighborhoods,
either to use the system, consistent with Cervero & Duncan (2008) or because transit locations
are proxies for the types of neighborhoods they seek, as in Levine & Frank (2008). This finding
provides strong evidence for a “field of dreams” approach to transit infrastructure and
operations: the more you build and the better you operate, the more transit-exposed your
metropolitan area, and the higher the return on investment for the region, the transit operator, and
the residents. Giuliano (2015) underscores the need for high-frequency transit provision
especially for peak period commuting to serve existing transit riders and expand the pool of new
riders. My findings show that this need helps increase transit exposure and thus its attendant
benefits.
I also find that young, non-married, and below median income households are the likeliest to
move to a transit-proximate neighborhood. This may run counter to the hypothesis of a “back to
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
188

the city” movement by empty-nester households moving to transit-proximate neighborhoods in
the central parts of the city. There are several explanations for the behavior of these mover types.
Lower-income households use transit more frequently (Santos et al., 2009) and thus may be
attracted by its proximity in these neighborhoods. Each of these types of households are also
more likely to be renters and to look for less expensive housing. Transit-proximate
neighborhoods are more likely to have higher proportions of rentals, because of zoning and
density existing before the transit comes, or rezoning or new dense development after the
introduction of transit. Certain transit neighborhoods may also have lower rents and be more
affordable than other Los Angeles County neighborhoods (Boarnet et al., 2015). For young and
non-married households, early exposure to transit-proximate neighborhoods may increase the
propensity of moving to other transit-proximate neighborhoods once married or older.
This study faces a few data and methodological limitations, in addition to geocoding and non-
filing of taxes discussed in Chapters 1 and 2. Transportation-related data over time is only
available for the L.A. Metro rail station. Data on bus stop locations over time is unavailable.
Data on TPAs and HQTAs was only generated in 2012 in response to the implementation of SB
375. Future iterations of HQTA data exist as predictive of the transit system in 2040, but a
backward looking version does not exist. A comparison of the 2012 and 2040 HQTA maps
reveals few differences in Los Angeles County. Thus, the static neighborhood assumption for
TPAs and HQTAs may be reasonable in this case. Moreover, the analysis using static transit
neighborhood definitions still reveals interesting and relevant findings.
Methodologically, the intractability of running a nested logit on my very large sample is a
limitation. A nested logit would have been the preferred specification, given potential issues with
the independence of irrelevant alternatives assumption. As a work-around, I modeled each stage
of the Mobility and Location Choices individually and weighted the second stage by predicted
values for the first stage. Future work can improve on this by running a nested logit on a smaller
sub-sample and bootstrapping to ensure that the sub-sample does not differ significantly from the
overall sample.
This work also does not consider neighborhood characteristics outside of distance and rail-
proximity. It is possible that including other neighborhood-level data including information
about the existing characteristics of the population or the characteristics of its housing would
increase the explanatory value of my models. However, I was unable to obtain disaggregated
spatial data for each year for Los Angeles County neighborhoods. Future work should identify
other non-traditional data sources to improve the explanatory value of the models and to build a
more inclusive model of transit exposure. Despite these limitations, it is unlikely that more data
or more nuanced modeling would shift the main finding that transit exposure is the strongest
predictor of future transit exposure.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
189

Appendix
3-A. Los Angeles Metro Rail Map

Figure 29. Map of Los Angeles Metro Rail Lines open in 2013
Source: Los Angeles County Metropolitan Transit Authority
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
190

3-B. Additional Geographic Move Pattern Maps
Figure 30. A-C: Top Move Destinations (by 5 digit zip code) for Households Moving from L.A. Metro Gold Line – Boyle Heights Branch
neighborhoods, Before and After Transit service opened (2003).
Note: Each panel represents a different income stratification. Only zip codes with at least 10 moves, from 1993-2012 shown, for confidentiality
reasons.
Source: SCAG, LA County GIS, LA City Planning, Author calculations on California Franchise Tax Board data, Get, HERE, DeLorme,
MapMyIndia, OpenMapData contributors and the GIS User community. Created in ArcGIS.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
191

64,065 Total Number of Movers 16,794
1.1% % Moved fewer than 0.5 miles 1.6%
10.3% % Moved Outside of LA County 11.5%
0.3% % Suppressed for Confidentiality Reasons 2.8%

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
192

27,376 Total Number of Movers 6,938
1.2% % Moved fewer than 0.5 miles 2.0%
9.5% % Moved Outside of LA County 11.2%
1.6% % Suppressed for Confidentiality Reasons 7.5%

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
193

2,664 Total Number of Movers 1,334
2.1% % Moved fewer than 0.5 miles 5.5%
16.6% % Moved Outside of LA County 19.2%
33.3% % Suppressed for Confidentiality Reasons 44.2%

Figure 31. A-C: Top Move Destinations (by 5 digit zip code) for Households Moving from L.A. Metro Red and Purple Line neighborhoods, Before
and After Transit service opened (2003)
Note: Each panel represents a different income stratification. Only zip codes with at least 10 moves, from 1993-2012 shown, for confidentiality
reasons.
Source: SCAG, LA County GIS, LA City Planning, Author calculations on California Franchise Tax Board data, Get, HERE, DeLorme,
MapMyIndia, OpenMapData contributors and the GIS User community. Created in ArcGIS.

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
194

69,063 Total Number of Movers 313,698
0.6% % Moved fewer than 0.5 miles 1.3%
9.2% % Moved Outside of LA County 12.6%
0.1% % Suppressed for Confidentiality Reasons 0.0%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
195

31,654 Total Number of Movers 122,262
0.6% % Moved fewer than 0.5 miles 1.2%
7.6% % Moved Outside of LA County 11.2%
1.1% % Suppressed for Confidentiality Reasons 0.0%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
196

4,347 Total Number of Movers 30,142
1.7% % Moved fewer than 0.5 miles 1.6%
11.5% % Moved Outside of LA County 17.7%
19.9% % Suppressed for Confidentiality Reasons 2.5%
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
197

3-C. Weighted MNL Results for Alternative Transit Definitions

Table 48. Cohort Age Effects of Mover Type Weighted MNL Model Results displayed as Relative Risk
Ratios for L.A. Metro Rail with current and future stations transit definition
Location Choices Non-Rail to Any Rail Any Rail to Non-Rail Any Rail to Any
Rail
Income Category (80-120% AMI is reference)
<30% of AMI 1.54*** 1.30* 1.87***
30-50% of AMI 1.44*** 1.28** 1.68***
50-80% of AMI 1.21*** 1.13** 1.30***
>120% of AMI 0.80*** 0.81** 0.70**
Number of Dependents (0 is reference)
1 Dependent 0.92* 0.94 0.93
2 Dependents 0.92 0.93 0.90
3+ Dependents 0.95 0.95 0.94
Marital Status (Single is reference)
Married 0.71*** 0.81*** 0.80**
Head of Household 1.06*** 1.11*** 1.10**
Age Category (age 41-54 is reference)
age 18-25 1.06** 1.02 0.94
age 26-40 1.17*** 1.19*** 1.19***
age 55-70 0.84*** 0.84*** 0.82***
age 71+ 0.63*** 0.63*** 0.55***
Distance Category (0.5-2 miles is reference)
2-5 miles 1.05 1.20 0.41***
5-10 miles 1.08 1.32** 0.33***
10-25 miles 1.11 1.36** 0.20***
25-100 miles 0.66** 0.83 0.01***
>100 miles 0.59*** 0.99 0.00***
Constant 0.04*** 0.07*** 0.02***
Weights: Yes
Year Fixed Effects: Yes
Age Categories: N/A and <18 years old excluded
Standard Errors: Clustered at origin 5-digit Zip Code
Pseudo R
2
: 0.0262
Wald(chi2): 32,229.84
Degrees of Freedom: 111
Prob>chi2: 0.000
Log Pseudo- likelihood: -1,875,726
Number of Observations: 15,578,449
Note. Baseline category: Mover Type 1: non-rail to non-rail
*<0.05 ** p<0.01 *** p<0.001
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
198

Table 49. Mover Type Weighted MNL Model Results displayed as Relative Risk Ratios for TPA
Location Choices Non-TPA to TPA TPA to Non-TPA TPA to TPA
Income Category (80-120% AMI is reference)
<30% of AMI 1.35*** 1.18** 1.87***
30-50% of AMI 1.32*** 1.20*** 1.70***
50-80% of AMI 1.18*** 1.11*** 1.33***
>120% of AMI 0.79*** 0.83*** 0.70***
Number of Dependents (0 is reference)
1 Dependent 0.95* 1.01 1.01
2 Dependents 0.93* 0.99 0.99
3+ Dependents 0.94 0.99 1.02
Marital Status (Single is reference)
Married 0.73*** 0.83*** 0.69***
Head of Household 1.08*** 1.12*** 1.17***
Age Category (age 41-54 is reference)
age 18-25 1.07*** 0.99 0.93*
age 26-40 1.15*** 1.16*** 1.22***
age 55-70 0.88*** 0.90*** 0.83***
age 71+ 0.72*** 0.79*** 0.56***
Distance Category (0.5-2 miles is reference)
2-5 miles 1.01 1.11 0.67***
5-10 miles 1.08 1.31* 0.53***
10-25 miles 1.20 1.42** 0.31***
25-100 miles 0.82* 1.00 0.09***
>100 miles 0.65*** 1.12 0.00***
Constant 0.17*** 0.26*** 0.26***
Weights: Yes
Year Fixed Effects: Yes
Age Categories: N/A and <18 years old excluded
Standard Errors: Clustered at origin 5-digit Zip Code
Pseudo R
2
: 0.0421
Wald(chi2): 422,765
Degrees of Freedom: 111
Prob>chi2: 0.000
Log Pseudo- likelihood: -3,837,072
Number of Observations: 15,578,449
Note. Baseline category: Mover Type 1: non-rail to non-rail
*<0.05 ** p<0.01 *** p<0.001
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
199

Table 50. Mover Type Weighted MNL Model Results displayed as Relative Risk Ratios for HQTA
Location Choices Non-HQTA to
HQTA
HQTA to Non-HQTA HQTA to HQTA
Income Category (80-120% AMI is reference)
<30% of AMI 1.47*** 1.24*** 1.77***
30-50% of AMI 1.41*** 1.26*** 1.64***
50-80% of AMI 1.22*** 1.15*** 1.31***
>120% of AMI 0.81*** 0.85*** 0.81**
Number of Dependents (0 is reference)
1 Dependent 0.99 1.06 1.00
2 Dependents
0.97 1.04 0.96
3+ Dependents
1.01 1.08 0.99
Marital Status (Single is reference)
Married
0.73*** 0.84*** 0.66***
Head of Household
1.08*** 1.13*** 1.19***
Age Category (age 41-54 is reference)
age 18-25 1.10*** 0.97 0.94
age 26-40 1.16*** 1.18*** 1.25***
age 55-70 0.87*** 0.91*** 0.86***
age 71+ 0.72*** 0.86** 0.70***
Distance Category (0.5-2 miles is reference)
2-5 miles 1.08 1.22* 0.67***
5-10 miles
1.22 1.49*** 0.60***
10-25 miles
1.60*** 1.94*** 0.33***
25-100 miles
1.22 1.56*** 0.05***
>100 miles
1.11 1.91*** 0.00***
Constant
0.17*** 0.23*** 0.66*
Weights: Yes
Year Fixed Effects: Yes
Age Categories: N/A and <18 years old excluded
Standard Errors: Clustered at origin 5-digit Zip Code
Pseudo R
2
: 0.0695
Wald(chi2): 459,866
Degrees of Freedom: 111
Prob>chi2: 0.000
Log Pseudo- likelihood: -4,118,847
Number of Observations: 15,578,449
Note. Baseline category: Mover Type 1: non-rail to non-rail
*<0.05 ** p<0.01 *** p<0.001
Source: Author calculations of FTB data

Households Mobility and Neighborhood Impacts Rodnyansky (2018)
200

References

Abodo.com. (February 21, 2017) Where Renting Trumps Owning. Retrieved from
https://www.abodo.com/blog/where-renting-trumps-owning/
Ainsworth, J. W. (2002). Why does it take a village? The mediation of neighborhood effects on
educational achievement. Social Forces, 81(1), 117-152.
Air Resources Board (ARB). (n.d.). Sustainable Communities. Retrieved December 22, 2015,
from http://www.arb.ca.gov/cc/sb375/sb375.htm
Allison, P. D. (2009). Fixed effects regression models (Vol. 160). SAGE publications.
Alonso, W. (1964). Location and Land Use. Cambridge, MA: Harvard University Press.
Arrington, G. B., & Cervero, R. (2008). TCRP Report 128: Effects of TOD on Housing, Parking,
and Travel. Transportation Research Board of the National Academies, Washington,
DC, 3.
Atkinson, A. B., Piketty, T., & Saez, E. (2011). “Top incomes in the long run of history”. Journal
of economic literature, 49(1), 3-71.
Atkinson, R. (2000). Measuring gentrification and displacement in Greater London. Urban
studies, 37(1), 149-165.
Atkinson-Palombo, C. (2010). Comparing the capitalisation benefits of light-rail transit and
overlay zoning for single-family houses and condos by neighbourhood type in
metropolitan Phoenix, Arizona. Urban Studies, 47(11), 2409-2426.
Bailey, L., Mokhtarian, P. L., & Little, A. (2008). The broader connection between public
transportation, energy conservation and greenhouse gas reduction (No. TCRP Project J-
11/Task 3). Fairfax, VA: ICF International.
Baker, D. M., & Lee, B. (2017). How Does Light Rail Transit (LRT) Impact Gentrification?
Evidence from Fourteen US Urbanized Areas. Journal of Planning Education and
Research, 0739456X17713619.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
201

Banzhaf, H. S., & Walsh, R. P. (2008). Do people vote with their feet? An empirical test of
Tiebout. American Economic Review, 98(3), 843-63.
Bartholomew, K., & Ewing, R. (2011). Hedonic price effects of pedestrian-and transit-oriented
development. Journal of Planning Literature, 26(1), 18-34.
Baum-Snow, N., & Kahn, M. E. (2000). The effects of new public projects to expand urban rail
transit. Journal of Public Economics, 77(2), 241-263.
Berry, R. (1985). Islands of Renewal in Seas of Decay, The New Urban Reality.(P. Peterson,
ed). The Brookings Institution, Washington DC, 123-140.
Besser, L. M., & Dannenberg, A. L. (2005). Walking to public transit: steps to help meet
physical activity recommendations. American journal of preventive medicine, 29(4), 273-
280.
Blumenberg, E., Taylor, B.D., Smart, M., Ralph, K., Wander, M. & Brumbaugh, S. (2012).
What’s youth got to do with it? Exploring the Travel Behavior of Teens and Young
Adults. University of California Transportation Center. UCTC-FR-2012-14. Retrieved
from https://cloudfront.escholarship.org/dist/prd/content/qt9c14p6d5/qt9c14p6d5.pdf
Boarnet, M. G. and Houston, D. (2014). Case Study of Land Use Changes near Expo and Gold
Light Rail Transit Stations in Los Angeles County. Lincoln Institute of Land Policy.
Boarnet, M. G., Chakrabarti, S., Bostic, R., Rodnyansky, S., Morrison-Hawkins, B., Santiago
Bartolomei, R., and Wang, X. (2015). Housing Affordability and Transit-Oriented
Development in Los Angeles: A Challenge and an Opportunity. California Community
Foundation.
Boarnet, M.G., Bostic, R.W., Rodnyansky, S., Santiago-Bartolomei, R., & Leslie, C. (2017).
Developing Affordable Housing Guidelines Near Rail Transit in Los Angeles.
METRANS. Retrieved from https://www.metrans.org/sites/default/files/research-
project/15-13%20Final%20Report_Bostic.pdf
Boarnet, M.G., Bostic, R.W., Rodnyansky, S., Santiago-Bartolomei, R., Williams, D., Prohofsky,
A. (2017). Sustainability and Displacement: Assessing the Spatial Pattern of Residential
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
202

Moves near Rail Transit in Los Angeles County. National Center for Sustainable
Transportation. University of California - Davis.
Boarnet, M.G., Bostic, R.W., Williams, D., Santiago-Bartolomei, R., Rodnyansky, S., Eisenlohr,
A. (2017). Building Affordable Housing in Transit-Oriented Developments: Impacts on
Driving and Policy Approaches. National Center for Sustainable Transportation.
Retrieved from https://ncst.ucdavis.edu/wp-content/uploads/2015/10/NCST-TO-027-
Boarnet-Bostic-Affordable-TOD-White-Paper_FINAL.pdf
Boehm, T. P., & Ihlanfeldt, K. R. (1986). Residential mobility and neighborhood quality. Journal
of Regional Science, 26(2), 411-424.
Boutell, T. (2016). Free Zip Code Latitude and Longitude Database, 2016.
https://boutell.com/zipcodes/
Boyd, M. L., Edin, K., Clampet-Lundquist, S., & Duncan, G. J. (2010). The durability of gains
from the Gautreaux Two residential mobility program: a qualitative analysis of who stays
and who moves from low-poverty neighborhoods. Housing Policy Debate, 20(1), 119-
146.
Brightwell, E. (2013, August 28). Exploring the Planned Course of the Metro Purple Line
Extension. KCET. Retrieved from https://www.kcet.org/history-society/exploring-the-
planned-course-of-the-metro-purple-line-extension
Brown, A.E., Blumenberg, E., Taylor, B.D., Ralph, K., & Voulgaris, C.T. (2016). A taste for
transit? Analyzing Ppblic transit use trends among youth. Journal of Public
Transportation. (19) 1: 49-67.
Brown, H. J. (1975). Changes in workplace and residential locations. Journal of the American
Institute of Planners, 41(1), 32-39.
Bures, R. M. (2003). Childhood residential stability and health at midlife. American Journal of
Public Health, 93(7), 1144-1148.
California Franchise Tax Board (FTB). (2006). Tax Gap Plan: A Strategic Approach to Reducing
California’s Tax Gap. Retrieved from
https://www.ftb.ca.gov/aboutftb/taxgapstratplan.pdf
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
203

California Franchise Tax Board (FTB). (2013). California 540: 2013 Personal Income Tax
Booklet. Retrieved from https://www.ftb.ca.gov/forms/2013/13_540bk.pdf
California Franchise Tax Board (FTB). (2017). 2016 California Tax Rate Schedules. Retrieved
from https://www.ftb.ca.gov/forms/2016-california-tax-rates-and-exemptions.shtml
California Franchise Tax Board (FTB). Tax Calculator, Tables, and Rates, (n.d.) Retrieved from
https://www.ftb.ca.gov/online/Tax_Calculator/index.asp?WT.mc_id=Ind_Popular_TaxCa
lcTablesRates
California Legislative Analyst’s Office (LAO). (2015). California’s High Housing Costs: Causes
and Consequences. Sacramento, CA: Mac Taylor. Retrieved from
www.lao.ca.gov/reports/2015/finance/housing-costs/housing-costs.pdf
Cameron, A. C., & Trivedi, P. K. (2005). Microeconometrics: methods and applications.
Cambridge University Press.
Cervero, R. & Duncan, M. (2008). Residential Self Selection and Rail Commuting: A Nested
Logit Analysis. University of California Transportation Center. Working Paper.
Retrieved from http://reconnectingamerica.org/assets/Uploads/604.pdf
Cervero, R. & Gorham, R. (1995). Commuting in transit versus automobile
neighborhoods. Journal of the American planning Association (61.2): 210-225.
Cervero, R. (2007). Transit-Oriented Development’s Ridership Bonus: A Product of Self-
Selection and Public Policies. Environment and Planning A. (39): 2068-2085.
Chapple, K. (2009). Mapping susceptibility to gentrification: The early warning
toolkit. Berkeley, CA. University of California-Berkeley, Center for Community
Innovation. Retrieved from
http://communityinnovation.berkeley.edu/reports/Gentrification-Report.pdf.
Chapple, K. (2009). Overcoming Mismatch: Beyond Dispersal, Mobility, and Development
Strategies. Journal of the American Planning Association, 2006. 72:3, 322-336.
http://dx.doi.org/10.1080/01944360608976754
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
204

Chatman, D.G., Xu, R., Park, J., & Spevack, A. (2017). The Effects on Auto Use of Household
Displacement from Rail Station Areas. In Chapple, K., Loukaitou-Sideris, A., Waddell,
P., Chatman, D., & Ong, P. Developing a New Methodology for Analyzing Potential
Displacement (pp.156-180). Publication ARB No. 13-310. University of California,
Berkeley, University of California, Los Angeles, and California Air Resources Board.
Chen, C., Gong, H., & Paaswell, R. (2008). Role of the built environment on mode choice
decisions: additional evidence on the impact of density. Transportation, 35(3), 285-299.
Chen, P. C., & Wilmoth, J. M. (2004). The effects of residential mobility on ADL and IADL
limitations among the very old living in the community. The Journals of Gerontology
Series B: Psychological Sciences and Social Sciences, 59(3), S164-S172.
Chetty, R., Hendren, N., Kline, P., & Saez, E. (2014). Where is the land of Opportunity? The
Geography of Intergenerational Mobility in the United States. The Quarterly Journal of
Economics, 129(4), 1553-1623.
Clark, W. A., & Davies Withers, S. (1999). Changing jobs and changing houses: mobility
outcomes of employment transitions. Journal of Regional Science, 39(4), 653-673.
Clark, W. A., & Davies, S. (1990). Elderly mobility and mobility outcomes: Households in the
later stages of the life course. Research on aging, 12(4), 430-462.
Clark, W. A., & Huang, Y. (2003). The life course and residential mobility in British housing
markets. Environment and Planning A, 35(2), 323-339.
Clark, W., Deurloo, M., & Dieleman, F. (2006). Residential mobility and neighbourhood
outcomes. Housing Studies, 21(3), 323-342.
Clark, W.A.V, & Ledwith, V. (2006). Mobility, housing stress, and neighborhood contexts:
Evidence from Los Angeles. Environment and Planning A (38): 1077-1093.
DOI:10.1068/a37254.
Collinson, Rob. (2011). Rental Housing Affordability Dynamics: 1990-2009. Cityscape. 13:2,
pp. 71-103.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
205

Corak, M., & Heisz, A. (1999). The intergenerational earnings and income mobility of Canadian
men: Evidence from longitudinal income tax data. Journal of Human Resources, 504-533.
Coulton, Claudia, Brett Theodos, and Margery A. Turner. (2012). Residential Mobility and
Neighborhood Change: Real Neighborhoods Under the Microscope. Cityscape 14 (3):
55-89.
Cox, R., Henwood, B., Rodnyansky, S., Wenzel, S., & Rice, E. (2017). Roadmap to a unified
measure of housing insecurity. Washington Center for Equitable Growth. Washington,
DC. Retrieved from http://cdn.equitablegrowth.org/wp-
content/uploads/2017/07/11133001/07112017-WP-unified-measure-of-housing-
insecurity.pdf
Cutuli, J. J., Desjardins, C. D., Herbers, J. E., Long, J. D., Heistad, D., Chan, C. K., ... & Masten,
A. S. (2013). Academic achievement trajectories of homeless and highly mobile students:
Resilience in the context of chronic and acute risk. Child development, 84(3), 841-857.
Dawkins, C., & Moeckel, R. (2016). Transit-induced gentrification: Who will stay, and who will
go?. Housing Policy Debate, 26(4-5), 801-818.
Delmelle, E.C. & Nilsson, I. (2018). New Rail Transit Stations and the Out-Migration of Low-
Income Residents. Working paper, University of North Carolina at Charlotte.
Ding, L., Hwang, J., & Divringi, E. (2016). Gentrification and residential mobility in
Philadelphia. Regional science and urban economics, 61, 38-51.
Dong, H. (2017). Rail-transit-induced gentrification and the affordability paradox of
TOD. Journal of Transport Geography, 63, 1-10.
Duncan, M. (2011). The impact of transit-oriented development on housing prices in San Diego,
CA. Urban studies, 48(1), 101-127.
Earth Radius Latitude Calculator (n.d.). See https://rechneronline.de/earth-radius/
Ellen, I. G., & O'Regan, K. M. (2011). How low income neighborhoods change: Entry, exit, and
enhancement. Regional Science and Urban Economics, 41(2), 89-97.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
206

Ennett, S. T., Flewelling, R. L., Lindrooth, R. C., & Norton, E. C. (1997). School and
neighborhood characteristics associated with school rates of alcohol, cigarette, and
marijuana use. Journal of Health and Social Behavior, 55-71.
Evans, G. W., & Wachs, T. D. (2010). Chaos and its influence on children’s
development. Washington, DC: American Psychological Association.
Frank, L. D., & Engelke, P. O. (2001). The built environment and human activity patterns:
exploring the impacts of urban form on public health. Journal of planning
literature, 16(2), 202-218.
Fredland, D. R. (1974). Residential mobility and home purchase. Lexington, MA: DC Heath.
Freeman, L. (2005). Displacement or Succession? Residential Mobility in Gentrifying
Neighborhoods. Urban Affairs Review 40 (4): 463–491.
Freeman, L., & Braconi, F. (2004). Gentrification and displacement New York City in the
1990s. Journal of the American Planning Association, 70(1), 39-52.
Freeman, L., Cassola, A., & Cai, T. (2015). Displacement and gentrification in England and
Wales: A quasi-experimental approach. Urban Studies: 0042098015598120.
Gabriel, S. A., & Painter, G. (2008). Mobility, residential location and the American dream: The
intrametropolitan geography of minority homeownership. Real Estate Economics, 36(3),
499-531.
Geolytics, Inc. (2015). “Zip+4 2000, Zip+4 2004, Zip+4 2007, Zip+4 2009, Zip+4 2012, Zip+4
2013, Zip+4 2014 ”. See
http://www.geolytics.com/USCensus,Zip4,Data,Features,Products.asp
Goldsmith, P. R., Britton, M. L., Reese, B., & Velez, W. (2017). Will moving to a better
neighborhood help? Teenage residential mobility, change of context, and young-adult
educational attainment. Urban Affairs Review, 53(2), 305-337.
Goldstein, S. (1958). Patterns of Mobility, 1910-1950; The Norristown Study. University of
Pennsylvania Press.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
207

Goldstein, S. (1964). The extent of repeated migration: an analysis based on the Danish
population register. Journal of the American Statistical Association, 59(308), 1121-1132.
Golub, A., Guhathakurta, S., & Sollapuram, B. (2012). Spatial and temporal capitalization
effects of light rail in Phoenix: From conception, planning, and construction to
operation. Journal of Planning Education and Research, 32(4), 415-429.
Gonçalves, S., & White, H. (2005). Bootstrap standard error estimates for linear
regression. Journal of the American Statistical Association, 100(471), 970-979.
Grier, G. & Grier, E. (1978). Urban Displacement: A Reconnaissance. Bethesda, MD: Grier
Partnership.
Grimsrud, M., & El-Geneidy, A. (2014). Transit to eternal youth: Lifecycle and generational
trends in Greater Montreal public transport mode share. Transportation, 41(1), 1-19.
Grube-Cavers, A., & Patterson, Z. (2015). Urban rapid rail transit and gentrification in Canadian
urban centres: A survival analysis approach. Urban Studies, 52(1), 178-194.
Guerra, E., Cervero, R., & Tischler, D. (2011). The Half-Mile Circle: Does It Represent Transit
Station Catchments? University of California Transportation Center Report UCTC-FR-
2011-09.
Hausman, J. A., Abrevaya, J., & Scott-Morton, F. M. (1998). Misclassification of the dependent
variable in a discrete-response setting. Journal of econometrics, 87(2), 239-269.
Heilmann, K. T. (2017). Essays in Urban Economics and International Trade. University of
California, San Diego.
Hellevik, O. (2007) Linear versus logistic regression when the dependent variable is a
dichotomy. Quality & Quantity, 43(1), 59–74. http://doi.org/10.1007/s11135-007-9077-3
Hess, A. (2012, July 11). Race, Class, and the Stigma of Riding the Bus in America. The
Atlantic. Retrieved from https://www.theatlantic.com/politics/archive/2012/07/race-class-
and-the-stigma-of-riding-the-bus-in-america/428085/
Hess, D. B., & Almeida, T. M. (2007). Impact of proximity to light rail rapid transit on station-
area property values in Buffalo, New York. Urban studies, 44(5-6), 1041-1068.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
208

Higgins, C.D., & Kanaroglou, P.S. (2016). Forty years of modelling rapid transit’s land value
uplift in North America: Moving beyond the tip of the iceberg. Transport Reviews 36(5):
610-634.
Horrace, W. C., & Oaxaca, R. L. (2006). Results on the bias and inconsistency of ordinary least
squares for the linear probability model. Economics Letters, 90(3), 321-327.
Immergluck, D. (2009). Large redevelopment initiatives, housing values and gentrification: the
case of the Atlanta Beltline. Urban Studies 46(8), 1723-1745.
Ioannides, Y. M., & Kan, K. (1996). Structural estimation of residential mobility and housing
tenure choice. Journal of Regional Science, 36(3), 335-363.
Jelleyman, T., & Spencer, N. (2008). Residential mobility in childhood and health outcomes: a
systematic review. Journal of Epidemiology & Community Health, 62(7), 584-592.
Kahn, M. E. (2007). Gentrification trends in new transit‐oriented communities: Evidence from
14 cities that expanded and built rail transit Systems. Real Estate Economics, 35(2), 155-
182.
Kain, J. F. (1968). Housing segregation, negro employment, and metropolitan
decentralization. The Quarterly Journal of Economics, 82(2), 175-197.
Kang, N., & Kwak, N. (2003). A multilevel approach to civic participation: Individual length of
residence, neighborhood residential stability, and their interactive effects with media
use. Communication Research, 30(1), 80-106.
Keane, M.P. (1992). A Note on Identification in the Multinomial Probit Model. Journal of
Business and Economic Statistics (10):193-200.
Khattak, A. J., & Rodriguez, D. (2005). Travel behavior in neo-traditional neighborhood
developments: A case study in USA. Transportation Research Part A: Policy and
Practice, 39(6), 481-500.
Knaap, G. J., Ding, C., & Hopkins, L. D. (2001). Do plans matter? The effects of light rail plans
on land values in station areas. Journal of Planning Education and Research, 21(1), 32-
39.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
209

Lat/Lon Distance Excel (Spherical Law of Cosines and Haversine Formula in Excel). (2011).
Retrieved from https://webtrough.wordpress.com/2011/09/19/latlon-distance-formula-in-
excel-haversine-and-spherical-law-of-cosines/
Lawrence, J., Udell, M. & Young, T. (2011). The Income Tax Position of Persons Not Filing
Returns for Tax Year 2005. IRS Position Paper.
Lee, B. A., Oropesa, R. S., & Kanan, J. W. (1994). Neighborhood context and residential
mobility. Demography, 31(2), 249-270.
Lee, E. S. (1966). A theory of migration. Demography, 3(1), 47-57.
Lee, K. O. (2014). Why do renters stay in or leave certain neighborhoods? The role of
neighborhood characteristics, housing tenure transitions, and race. Journal of Regional
Science, 54(5), 755-787.
Leventhal, T., & Brooks-Gunn, J. (2000). The neighborhoods they live in: the effects of
neighborhood residence on child and adolescent outcomes. Psychological
bulletin, 126(2), 309.
Levine, J., & Frank, L. D. (2007). Transportation and land-use preferences and residents’
neighborhood choices: The sufficiency of compact development in the Atlanta
region. Transportation, 34(2), 255-274.
Lin, J, & Long, L. (2008). What neighborhood are you in? Empirical findings of relationships
between household travel and neighborhood characteristics. Transportation (35) 6:739.
Lin, J. (2002). Gentrification and Transit in Northwest Chicago. Journal of the Transportation
Research Forum 56: 175-191.
Long, J. S. (1997) Regression Models for Categorical and Limited Dependent Variables (1st ed.).
Sage Publications, Inc.
Los Angeles County and Metropolitan Transportation Authority (L.A. Metro). (2009). Long
Range Transportation Plan. Retrieved from:
http://media.metro.net/projects_studies/images/final-2009-LRTP.pdf
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
210

Los Angeles Department of City Planning (LADCP). (2013). Housing Element 2013-2021
(CPC-2013-1318- GPA). Retrieved from
http://planning.lacity.org/HousingInitiatives/HousingElement/Text/HousingElement_201
40321.pdf
Los Angeles Department of City Planning (LADCP). (2018). Technical clarifications to the
transit oriented communities affordable housing incentive program guidelines (CPC-
2017-1914-MSC). Los Angeles, CA. Retrieved from
https://planning.lacity.org/ordinances/docs/toc/TOCGuidelines.pdf
Malone, T., & Redfearn, C. L. (2018). Shocks & ossification: The durable hierarchy of
neighborhoods in US metropolitan areas from 1970 to 2010. Regional Science and Urban
Economics, 69, 94-121.
Manski, C.F. & Lerman, S.R. (1977). The Estimation of Choice Probabilities from Choice-Based
Samples. Econometrica (45):1977-1988.
Manville, M., Taylor, B.D., & Blumenberg, E. (2018). Falling Transit Ridership: California and
Southern California. UCLA Institute of Transportation Studies. Retrieved from
https://www.its.ucla.edu/2018/01/31/new-report-its-scholars-on-the-cause-of-californias-
falling-transit-ridership/
Marcuse, P. (1985). Gentrification, abandonment, and displacement: Connections, causes, and
policy responses in New York City. Wash. UJ Urb. & Contemp. L., 28, 195.
Martin, I. W., & Beck, K. (2018). Gentrification, Property Tax Limitation, and
Displacement. Urban Affairs Review, 54(1), 33-73.
McKinnish, T., Walsh, R., and White, K., (2010). Who Gentrifies Low-Income Neighborhoods?
Journal of Urban Economics, 67, pp.180–193.
Metzger, M. W., Fowler, P. J., Anderson, C. L., & Lindsay, C. A. (2015). Residential mobility
during adolescence: Do even “upward” moves predict dropout risk?. Social Science
Research, 53, 218-230.
Moore, E. G. (1971). Comments on the Use of Ecological Models in the Study of Residential
Mobility in the City. Economic Geography, 47(1), 73-85.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
211

Morris, T., Manley, D., & Sabel, C. E. (2018). Residential mobility: Towards progress in
mobility health research. Progress in Human Geography, 42(1), 112-133.
Morrow, G. (2013). The Homeowner Revolution: Democracy, Land Use and the Los Angeles
Slow-Growth Movement, 1965-1992. (Doctoral dissertation). University of California,
Los Angeles.
Muth, R. (1969). Cities and housing: The spatial patterns of urban residential land
use. University of Chicago, Chicago, IL.
National Institute for Advanced Studies (NIAS). (1981). Market Generated Displacement: A
Single City Case Study. Washington, DC: National Institute for Advanced Studies.
Newbold, K. B., & Scott, D. M. (2017). Driving over the life course: The automobility of
Canada’s Millennial, generation X, baby boomer and greatest generations. Travel
Behaviour and Society, 6, 57-63.
Newman, K., & Wyly, E.K. (2006). The right to stay put: gentrification and resistance to
displacement in New York City. Urban Studies: 43 (1): 23–57.
doi:10.1080/00420980500388710.
Newman, S. J., & Owen, M. S. (1982). Residential displacement: Extent, nature, and
effects. Journal of social Issues, 38(3), 135-148.
Ong, P., Zuk, M., Pech, C., & Chapple, K. (2017). Development of an off-model displacement
assessment methodology. In Chapple, K., Loukaitou-Sideris, A., Waddell, P., Chatman,
D., and P. Ong (Eds.), Developing a New Methodology for Analyzing Potential
Displacement (pp. 148-153) (Publication ARB No. 13-310). University of California,
Berkeley, University of California, Los Angeles, and California Air Resources Board.
Osterberg, T. (2000). Intergenerational Income Mobility in Sweden: What Do Tax-Data Show?.
Review of Income and Wealth, 46(4), 421-36.
Pendall, R., Gainsborough, J., Lowe, K., & Nguyen, M. (2012). Bringing equity to transit-
oriented development: Stations, systems, and regional resilience. Building Resilient
Regions: Urban and Regional Policy and its Effects, 4, 148-92.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
212

Pew Research Center. (2015). The whys and hows of generations research. Retrieved from
http://assets.pewresearch.org/wp-content/uploads/sites/5/2015/09/09-3-2015-
Generations-explainer-release.pdf
Piketty, T., & Saez, E. (2003). Income Inequality in the United States, 1913-1998. The Quarterly
Journal of Economics, 1-39
Pilgram, C. A., & West, S. E. (2017). Fading premiums: The effect of light rail on residential
property values in Minneapolis, Minnesota. Regional Science and Urban Economics.
Pollack, S., Bluestone, B., and Billingham, C. (2010). Maintaining diversity in America’s transit-
rich neighborhoods: Tools for equitable neighborhood change. Dukakis Center for Urban
and Regional Policy. Boston, MA. Retrieved from
http://nuweb9.neu.edu/dukakiscenter/wp-content/uploads/TRN_Equity_final.pdf
Pollack, S., Bluestone, B., Billingham, C. (2010). Maintaining Diversity In America’s Transit-
Rich Neighborhoods: Tools for Equitable Neighborhood Change. Dukakis Center for
Urban and Regional Policy.
Puget Sound Sage. (2012). Transit oriented development that’s healthy green and just: ensuring
transit investment in Seattle’s Rainier Valley builds communities where all families
thrive. Seattle, WA (Howard Greenwich & Margaret Wykowski).
http://www.reconnectingamerica.org/resource-center/browse-research/2012-2/transit-
oriented-development-that-s-healthy-green-and-just-ensuring-transit-investment-in-
seattle-s-rainier-valley-builds-communities-where-all-families-thrive/
Quigley, J. M., & Weinberg, D. H. (1977). Intra-urban residential mobility: a review and
synthesis. International Regional Science Review, 2(1), 41-66.
Rayle, L. (2015). Investigating the connection between Transit-Oriented Development and
displacement: Four hypotheses. Housing Policy Debate, 25(3), 531-548.
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.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
213

Rosenbaum, J. E., Reynolds, L., & DeLuca, S. (2002). How do places matter? The geography of
opportunity, self-efficacy and a look inside the black box of residential mobility. Housing
studies, 17(1), 71-82.
Rossi, P. H. 1955. Why Families Move. Beverly Hills: Sage.
Rossi, P. H., & Shlay, A. B. (1982). Residential mobility and public policy issues: “Why families
move” revisited. Journal of Social Issues, 38(3), 21-34.
Roy, A. L., McCoy, D. C., & Raver, C. C. (2014). Instability versus quality: Residential
mobility, neighborhood poverty, and children’s self-regulation. Developmental
Psychology, 50(7), 1891.
Sabagh, G., Van Arsdol Jr, M. D., & Butler, E. W. (1969). Some determinants of
intrametropolitan residential mobility: Conceptual considerations. Social Forces, 48(1),
88-98.
Saez, E., & Zucman, G. (2016). Wealth Inequality in the United States since 1913: Evidence
from Capitalized Income Tax Data. The Quarterly Journal of Economics, 131(2), 519-
578.
Sampson, R. J., & Groves, W. B. (1989). Community structure and crime: Testing social-
disorganization theory. American journal of sociology, 94(4), 774-802.
Santos, A., McGuckin, N., Nakamoto, H.Y., Gray, D, & Liss, S. (2011). Summary of travel
trends: 2009 national household travel survey (FHWA-PL-ll-022). Federal Highway
Administration. Washington, DC.
Schill, M. H., Nathan, R. P., & Persaud, H. (1983). Revitalizing America's cities: Neighborhood
reinvestment and displacement. SUNY Press.
Schmitt, S. A., Finders, J. K., & McClelland, M. M. (2015). Residential mobility, inhibitory
control, and academic achievement in preschool. Early Education and
Development, 26(2), 189-208.
Schuetz, J., Giuliano, G., & Shin, E. J. (2017). Does zoning help or hinder transit-oriented (re)
development?. Urban Studies, 0042098017700575.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
214

Schuetz, J., Giuliano, G., & Shin, E. J. (2018). Can a car-centric city become transit oriented?
Evidence from Los Angeles. Cityscape, 20(1), 167-190.
Schweitzer, L. (2014). Planning and Social Media: A Case Study of Public Transit and Stigma
on Twitter, Journal of the American Planning Association, 80:3, 218-238, DOI:
10.1080/01944363.2014.980439
Sheller, M., & Urry, J. (2006). The new mobilities paradigm. Environment and planning
A, 38(2), 207-226.
Shumaker, B. P., & Sinnott, R. W. (1984). Astronomical computing: 1. Computing under the
open sky. 2. Virtues of the haversine. Sky and Telescope, 68, 158-159.
Sims, J. R. (2016). More than gentrification: geographies of capitalist displacement in Los
Angeles 1994–1999. Urban Geography, 37(1), 26-56.
Smith, N. (1979). Toward a theory of gentrification a back to the city movement by capital, not
people. Journal of the American Planning Association, 45(4), 538-548.
Solon, G. (1992). Intergenerational income mobility in the United States. The American
Economic Review, 393-408.
South, S. J., & Crowder, K. D. (1997). Residential mobility between cities and suburbs: Race,
suburbanization, and back-to-the-city moves. Demography, 34(4), 525-538.
South, S. J., & Crowder, K. D. (1998). Avenues and barriers to residential mobility among single
mothers. Journal of Marriage and the Family, 866-877.
Southern California Association of Governments (SCAG). (2012a). 2012-2035 Regional
Transportation Plan/Sustainable Communities Strategy. Retrieved from:
http://rtpscs.scag.ca.gov/Pages/default.aspx
Southern California Association of Governments (SCAG). (2012b). Transit Priority Areas (2012)
shapefile. Retrieved from: https://gisdata-
scag.opendata.arcgis.com/datasets/21262b1b31304b9da08860e094f7bed1_0
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
215

Southern California Association of Governments (SCAG). (2012c). High Quality Transit Areas
(HQTA) 2012 – SCAG Region shapefile. Retrieved from: https://gisdata-
scag.opendata.arcgis.com/datasets/1f6204210fa9420b87bb2e6c147e85c3_0
Southern California Association of Governments (SCAG). (2015). Rail Stations shapefile.
Obtained via formal data request.
Southern California Association of Governments (SCAG). (2016). THE 2016-2040 REGIONAL
TRANSPORTATION PLAN/ SUSTAINABLE COMMUNITIES STRATEGY: A Plan
for Mobility, Accessibility, Sustainability and a High Quality of Life. Retrieved from
http://scagrtpscs.net/Documents/2016/final/f2016RTPSCS.pdf
Speare, A. (1974). Residential satisfaction as an intervening variable in residential
mobility. Demography, 11(2), 173-188.
Spears, S., Boarnet, M. G., & Houston, D. (2017). Driving reduction after the introduction of
light rail transit: Evidence from an experimental-control group evaluation of the Los
Angeles Expo Line. Urban Studies, 54(12), 2780-2799.
Spherical Law of Cosines. (n.d.). In World Heritage Encyclopedia Edition, online. Retrieved
from http://www.gutenberg.us/articles/spherical_law_of_cosines
Spillar, R. J., & Rutherford, G. S. (1990). The effects of population density and income on per
capita transit ridership in western American cities. In INSTITUTE OF
TRANSPORTATION ENGINEERS MEETING.
Spross, J. (2017, August 16). How the American bus system broke down. The Week. Retrieved
from http://theweek.com/articles/718456/how-american-bus-system-broke-down
Stack Exchange: Geographic Information Systems. (n.d.) Why is law of cosines more preferable
than haversine when calculating distance between two latitude-longitude points?.
Retrieved from http://gis.stackexchange.com/questions/4906/why-is-law-of-cosines-
more-preferable-than-haversine-when-calculating-distance-b
Taylor, B. D., & Fink, C. N. (2003). The factors influencing transit ridership: A review and
analysis of the ridership literature.
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
216

Taylor, B. D., Kim, E. J., & Gahbauer, J. E. (2009). The thin red line: A case study of political
influence on transportation planning practice. Journal of Planning Education and
Research, 29(2), 173-193.
Taylor, M. (2015). California’s High Housing Costs. Causes and Consequences. California
Legislative Analyst’s Office (LAO). Retrieved from
www.lao.ca.gov/reports/2015/finance/housing-costs/housing-costs.pdf
Tiebout, C. M. (1956). A pure theory of local expenditures. Journal of political economy, 64(5),
416-424.
U.S. Census Bureau. (2015). U.S. Mover Rate Remains Stable at About 12 Percent Since 2008,
Census Bureau Reports. Retrieved from http://www.census.gov/newsroom/press-
releases/2015/cb15-47.html
U.S. Census Bureau. (2016). Figure 2. Table A-1. Annual Geographic Mobility Rates, By Type
of Movement: 1948-2016. CPS Historical Migration/Geographic Mobility Tables.
Retrieved from https://www.census.gov/data/tables/time-series/demo/geographic-
mobility/historic.html
U.S. Internal Revenue Service (IRS). (2010). The Tax Gap: Tax Gap Estimates from 2008-2010.
Retrieved from https://www.irs.gov/uac/the-tax-gap
U.S. Internal Revenue Service (IRS). (2012). IRS Releases New Tax Gap Estimates Compliance
Rates Remain Statistically Unchanged from Previous Study. Retrieved from
https://www.irs.gov/uac/irs-releases-new-tax-gap-estimates-compliance-rates-remain-
statistically-unchanged-from-previous-study
U.S. Internal Revenue Service (IRS). (2013). Exemptions, Standard Deductions, and Filing
Information, 2013 (Publication 501). Retrieved from https://www.irs.gov/pub/irs-
prior/p501--2013.pdf
U.S. Internal Revenue Service (IRS). (2014). EITC Participation Rate by States, 2014. Retrieved
from https://www.eitc.irs.gov/EITC-Central/Participation-Rate
U.S. Internal Revenue Service (IRS). (2017). Filing Information, 2017.
https://www.irs.gov/publications/p17/ch01.html#en_US_2016_publink1000170388
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
217

United States Department of Housing and Urban Development (HUD). (n.d.). State of the Cities
Data Systems - Permits data. Retrieved March 7, 2015, from
http://socds.huduser.org/permits/
Van der Vliest, A. J., Gorter, C., Nijkamp, P., & Rietveld, P. (2002). Residential mobility and
local housing-market differences. Environment and Planning A, 34(7), 1147-1164.
Van Ommeren, J., Rietveld, P., & Nijkamp, P. (1999). Job moving, residential moving, and
commuting: a search perspective. Journal of Urban Economics, 46(2), 230-253.
Varady, D. P. (1983). Determinants of residential mobility decisions the role of government
services in relation to other factors. Journal of the American Planning Association, 49(2),
184-199.
Vigdor, J. L., Massey, D. S., & Rivlin, A. M. (2002). Does gentrification harm the poor? [with
Comments]. Brookings-Wharton papers on urban affairs, 133-182.
Weinberg, D. H. (1979). The determinants of intra-urban household mobility. Regional Science
and Urban Economics, 9(2-3), 219-246.
Winstanley, A., Thorns, D. C., & Perkins, H. C. (2002). Moving house, creating home:
Exploring residential mobility. Housing studies, 17(6), 813-832.
Yang, L., Zheng, G., & Zhu, X. (2013). Cross-nested logit model for the joint choice of
residential location, travel mode, and departure time. Habitat International, 38, 157-166.
Young, K. (2009). Equity, gentrification, and light rail (NPCT 1309). Central Corridor Equity
Coalition & University of Minnesota Center for Urban and Regional Affairs.
Minneapolis, MN. http://www.cura.umn.edu/publications/catalog/npcr-1309
Ziol‐Guest, K. M., & McKenna, C. C. (2014). Early childhood housing instability and school
readiness. Child development, 85(1), 103-113.
Zip-Codes.com. (2016). California Zip Codes, 2016. Retrieved from https://www.zip-
codes.com/state/ca.asp
Households Mobility and Neighborhood Impacts Rodnyansky (2018)
218

Zuk, M., Bierbaum, A. H., Chapple, K., Gorska, K., & Loukaitou-Sideris, A. (2017).
Gentrification, Displacement, and the Role of Public Investment. Journal of Planning
Literature, 0885412217716439.
Zuk, M., Bierbaum, A., Chapple, K., Gorska, K., Loukaitou-Sideris, A., Ong, P., Thomas, T.
(2015). Gentrification, Displacement and the Role of Public Investment: A Literature
Review. Working papers of the Federal Reserve Bank of San Francisco. San Francisco,
CA. Retrieved from http://www.frbsf.org/community-development/publications/working-
papers/2015/august/gentrification-displacement-role-of-public-investment/

Abstract (if available)

Abstract This dissertation uses tax filing data for Los Angeles County to assess whether patterns of residential mobility have changed after the introduction of a new rail transit system in the region. This research provides new ways to measure how often and where households move and to compare against prevailing neighborhood-level and county-level patterns. This research extends the empirical literature on the effects of new rail transit systems in gentrifying nearby neighborhoods and displacing prior residents, including at-risk sub-population such as lower-income households, families with children, young households, and elderly households. Using a series of empirical and descriptive methods, this dissertation explores whether new rail station openings affect neighborhood-level mobility averages (Chapter 1), individual propensity to move (Chapter 2), and move destination (Chapter 3). Findings suggest that effects of new rail transit systems on moving are nuanced and heterogeneous with respect to time and geography. The evidence also suggests that future research should focus not only on moving but on other methods of coping with rising housing costs including overcrowding, consuming less, or paying more. This dissertation also sheds light on the need to incorporate mobility statistics into policymaking and urban planning and provides a potential data source and methodology.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The impact of mobility and government rental subsidies on the welfare of households and affordability of markets

PDF

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

PDF

The built environment, tour complexity, and active travel

PDF

Healthy mobility: untangling the relationships between the built environment, travel behavior, and environmental health

PDF

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

PDF

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

PDF

Environmental justice in real estate, public services, and policy

PDF

Insights into residential mobility and pricing of rental housing: the role of gentrification, home-ownership barriers, and market concentrations in low-income household welfare

PDF

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

PDF

Crowded with potential: housing and social mobility strategies among China's educated migrants

PDF

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

PDF

Urban air pollution and environmental justice: three essays

PDF

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

PDF

Unraveling decentralization of warehousing and distribution centers: three essays

PDF

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

PDF

The role of the timing of school changes and school quality in the impact of student mobility: evidence from Clark County, Nevada

PDF

The interactions between housing and business

PDF

The development implications of China’s Belt and Road Initiative for Russia, Kazakhstan and Belarus

PDF

Essays on the economics of cities

PDF

The film industry and urban development in metropolitan Los Angeles, 1920-1975

Asset Metadata

Creator Rodnyansky, Seva (author)

Core Title Household mobility and neighborhood impacts

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Urban Planning and Development

Publication Date 07/20/2018

Defense Date 06/14/2018

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

Tag displacement,gentrification,household mobility,Los Angeles,low income,OAI-PMH Harvest,public transit,rail transit,residential mobility

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Boarnet, Marlon G (committee chair), Ailshire, Jennifer (committee member), Bostic, Raphael W (committee member), Schweitzer, Lisa (committee member)

Creator Email rodnyans@usc.edu,sevarodnyansky@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-22104

Unique identifier UC11669376

Identifier etd-Rodnyansky-6436.pdf (filename),usctheses-c89-22104 (legacy record id)

Legacy Identifier etd-Rodnyansky-6436.pdf

Dmrecord 22104

Document Type Dissertation

Format application/pdf (imt)

Rights Rodnyansky, Seva

Type texts

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

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

Repository Name University of Southern California Digital Library

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