University of Southern California Dissertations and Theses

Spatial dimensions of stratification: neighborhood change, urban inequality, and the neighborhood-school link in the U.S.

(USC Thesis Other)

Spatial dimensions of stratification: neighborhood change, urban inequality, and the neighborhood-school link in the U.S.

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Spatial Dimensions of Stratification:
Neighborhood Change, Urban Inequality, and the
Neighborhood-School Link in the U.S.

A Dissertation Presented
by
Jennifer Candipan
to
the FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(SOCIOLOGY)

August 2019

© 2019 – Jennifer Candipan
All rights reserved.

i

Abstract
This dissertation examines the changing relationship between neighborhoods and
schools in the United States in terms of their racial/ethnic composition; contextual
effects on student outcomes; and how changes in both contexts contribute to inequality.
Neighborhoods and schools are linked —historically, under a strict residence-based school
assignment system, where one lived determined where one attended school. Since the
1990s, however, the tight relationship between where parents live and where they send
their child to school has weakened. Neighborhoods and schools are both key contexts for
children ’s well-being, and understanding their links provides insight into how
inequalities in educational access and outcomes between socially disadvantaged and
advantaged groups are produced and sustained.
As educational policy changes expanded school choice options and urban changes
produced gentrification, the neighborhood-school link may have also changed, and I
examine this in three key ways. I use geospatial techniques to link schools to the local
community they serve, creating a unique dataset for every school in the U.S. I employ
statistical techniques, including demographic analysis and longitudinal causal modeling
strategies, to model mismatches in neighborhood and school composition; families ’
residential and school enrollment choices; and neighborhood effects on children ’s test
scores. I draw upon data from national sources over four decades, including
Census/American Community Survey (ACS) data on neighborhood composition,
National Center for Education Statistics (NCES) data on school composition, location,
and sector (traditional, magnet, or charter), spatial data on school attendance
boundaries, and Panel Study of Income Dynamics (PSID) data on families ’ residential
moves, school enrollment, and children ’s test scores.
ii

First, I explore how the racial/ethnic composition of local schools corresponds with
neighborhood composition over time. What happens to key neighborhood institutions
like schools as neighborhoods change along demographic lines? I find a growing
mismatch between neighborhood and school racial/ethnic composition over time —
neighborhoods are increasingly composed of more white students than the local public
school. The compositional mismatch grows most in neighborhoods experiencing
demographic change and socioeconomic ascent, particularly when there are many nearby
alternatives to the local school (e.g., charter schools). This suggests that white parents
are bypassing the local school while the neighborhood is still changing, indicating that
parents accept more diversity in their neighborhoods than they do in their children ’s
schools.
Second, I examine how neighborhood characteristics and alternatives to local schools
affect where families choose to live and enroll their children. I move beyond aggregate-
level analyses to model residential and school enrollment outcomes of families with
school-age children. I document family, school, and neighborhood factors that shape
whether or not a family enrolls their child in the local school in gentrifying and non-
gentrifying neighborhoods. I find that parents, especially recent movers, opt out of
neighborhood schools more in gentrifying neighborhoods when nearby school choice
options are available. Results demonstrate new ways of sorting into neighborhoods and
schools in the face of neighborhood and school policy changes that may uphold patterns
of school segregation and unequal access to opportunity.
Finally, I identify whether the effects of neighborhoods on children ’s outcomes have
changed substantially over time, given social and demographic changes in
neighborhoods. I use successive generational cohort data from the PSID to document
whether neighborhoods have a weaker effect on children ’s educational outcomes in 2014
than they did twenty years earlier.
iii

This dissertation makes several contributions. While most prior research focuses on
either the neighborhood or school environment, this project analyzes both contexts
simultaneously and takes seriously the spatial interdependence of neighborhoods and
schools, enhancing basic understanding of the changing relationship between
neighborhoods and schools. Additionally, this study draws on a unique longitudinal
quantitative dataset to provide one of the first multi-city, longitudinal studies
examining the evolving relationship between neighborhoods and traditional public
schools in the U.S. Broadly, findings from my dissertation contribute to our
understanding of the mechanisms upholding neighborhood and school segregation, the
consequences of educational policy and urban changes for neighborhoods and schools,
and the processes of residential and school choice, with implications for policies in the
housing, urban, education, and economic development arenas.

iv

Table of Contents

Abstract ........................................................................................................................................ i
Acknowledgements ................................................................................................................... vi
Chapter 1 – Introduction .......................................................................................................... 1
Background ................................................................................................................................. 1
Research Questions ..................................................................................................................... 8
Overview of Empirical Chapters ............................................................................................... 10
Part One. Neighborhood Change and the Neighborhood-School Gap. ................................. 10
Part Two. Linking Residential and School Enrollment Decisions in Ascendant
Neighborhoods....................................................................................................................... 12
Part Three. Neighborhood Effects over Time on Educational Outcomes. ............................ 15
Broad Contributions .................................................................................................................. 18
References ................................................................................................................................. 19
Chapter 2 – Neighborhood Change and the Neighborhood-Change Gap .......................... 24
Introduction ............................................................................................................................... 25
Data and Measures .................................................................................................................... 32
Analytic Approach .................................................................................................................... 36
Findings ..................................................................................................................................... 38
Discussion ................................................................................................................................. 46
Tables ........................................................................................................................................ 49
References ................................................................................................................................. 52
Appendices ................................................................................................................................ 57
Chapter 3 – Choosing Schools in Changing Places: Examining School Enrollment in
Gentrifying Neighborhoods ................................................................................................ 65
Introduction ............................................................................................................................... 66
Data and Measures .................................................................................................................... 75
Analysis Plan ............................................................................................................................. 83
Findings ..................................................................................................................................... 85
Discussion ................................................................................................................................. 96
Tables and Figures .................................................................................................................. 102
References ............................................................................................................................... 108
Appendices .............................................................................................................................. 118
Chapter 4 – Neighborhood Effects on Educational OUtcomes: Have They Changed Over
Time? ................................................................................................................................. 120
Introduction ............................................................................................................................. 120
Background ............................................................................................................................. 122
v

Methodological Challenges in Neighborhood Effects Research............................................. 130
Data ......................................................................................................................................... 134
Measures.................................................................................................................................. 136
Analysis Plan ........................................................................................................................... 140
Preliminary Findings: Neighborhood Effects on Student Achievement ................................. 145
Descriptive Statistics ........................................................................................................... 145
Multivariate Analyses .......................................................................................................... 148
Next Steps and Extensions ...................................................................................................... 151
Contributions ........................................................................................................................... 153
Tables and Figures .................................................................................................................. 155
References ............................................................................................................................... 160
Chapter 5 ................................................................................................................................ 167
Overall Findings and Broad Implications ................................................................................ 167

vi

Acknowledgements

I am thankful for the personal, professional, and institutional support that have made this
dissertation possible. First, I would like to thank the National Science Foundation, National
Academy of Education, and Spencer Foundation for financial support that made it possible to
complete this study.
Many valued friends, colleagues, and mentors have shepherded me through a process that
can feel long, arduous and, at times, harsh. I would first like to thank my committee members.
Gary Painter provided professional advice and wealth of experience and expertise. Manuel
Pastor offered insights into navigating the academic journey while also pushing me to reach
beyond the walls of academia with my work. Very special thanks to Jennifer Ailshire for
continuing to inspire me as a model mentor and scholar, and for regularly reminding me of my
worth. Thank you for your great wisdom and generosity and for always making me feel included.
I am incredibly grateful to have you in my corner. Finally, my heartfelt appreciation goes to Ann
Owens for the many roles you've played at various times along the way: dissertation chair,
mentor, benefactor, friend. My thinking about my own work on neighborhoods and schools has
certainly evolved based on our exchanges, and I often left our discussions bubbling over with
new ideas. To say it was fortuitous to encounter someone with as many overlapping interests as
ours is an understatement. While I enjoyed sharing such similar research interests, the times
spent indulging in our other niche interests are among my most cherished memories.
Many other faculty members at USC and beyond have supported me along the way. Special
thanks to Jen Hook for your endless generosity, advice, mentorship, and friendship (and standing
vii

desk). Thanks also to Leland Saito for your early mentorship and for continuing to take an
interest not only in my professional success, but also my well-being as a person.
Many thanks to Amber, Lisa, Melissa, Angie, and the rest of the USC Sociology staff and
student workers. Your labor is not recognized nearly enough. Thank you for holding it together
for all of us. Stachelle—I would not have made it through grad school in one piece without you.
Thank you for your sage advice, sound solutions, listening ear, and great hugs.
I was fortunate to meet some amazing friends and colleagues during this process. Shoutout to
lunch club, karaoke club, ice cream club, family diners, PSID nerd campers, Spencer folks,
Brooklyn College friends, urban and neighborhood working groups, and more. I look forward to
future reunions across the country.
Academia can feel harsh and unwelcoming at times for those of us without the right kind of
capital. Thank you to those that supported me through difficult times and lifted me up when I
needed it most. I am especially grateful for family and long-time friends for always being there
for me and reminding me to hold on to the important parts of life: your people and your
principles.
My beloved owl-faced lap cat, Nori Plum Roll, kept me company as I waded through this
long slog. She was by my side well before this latest journey began, nearly making it to the end.
The final stretch of grad school was made rougher without you around to pounce lovingly on my
work. Thank you for your much-needed snuggles. I miss you dearly.
Finally, thank you to my parents for your many sacrifices and to the rest of my family for
your endless support and encouragement. I share this with you.
1

Chapter 1
Introduction

Background
This dissertation explores the changing relationship between neighborhoods and traditional
public schools in the United States. In earlier decades, the nearly all children attended their
neighborhood school. Recent statistics show, however, that roughly one quarter of families now
send their children to non-local schools (National Center for Education Statistics 2014). Recent
research suggests that school choice and private school options have loosened this historically
tight relationship. To the extent that the neighborhood-school relationship has weakened over
time, what are the consequences for children’s educational outcomes? Are schools becoming
more segregated than their neighborhoods? Do parents increasingly choose neighborhoods and
opt out of local schools? This dissertation aims to answer these broad questions.
Despite mainstream narratives about how residential decisions are linked to schools,
academic scholarship has not fully undertaken this line of research. Only recently have a small
handful of scholars begun investigating the complex links between neighborhood and school
preference (Lareau and Goyette 2014). Drawing upon the literature on urban sociology,
stratification, and the sociology of education, the interrelated papers comprising my dissertation
examine the relationship between neighborhoods and schools over time in terms of racial/ethnic
composition; contextual effects on student outcomes; and how changes in both contexts
contribute to inequality, with a particular focus on the neighborhood-school link in gentrifying
neighborhoods. The neighborhood-school link weakened as both policy changes expanded
2

school choice and largescale urban changes produced gentrification. Using multiple quantitative
data sources, and with a focus on neighborhoods experiencing economic change, I examine the
implications of the changing neighborhood-school link in three ways: (1) mismatches in
neighborhood and school composition; (2) families’ residential and school enrollment choices;
and (3) neighborhood effects on children’s test scores.
After several decades of scholarship on gentrification, surprisingly little quantitative research
has documented the relationship between gentrification and schools in the U.S.. Because schools
are thought to be important neighborhood institutions, understanding how neighborhood change
affects schools has important implications for theory and policy.

Theoretical Grounding and Relevant Literature
Gentrification represents a residential sorting process that is currently reshaping the spatial
landscape of many cities. Gentrification offers a counternarrative to how neighborhoods become
segregated, instead describing an alternative view of residential sorting that leads, at least
temporarily, to integration along racial and economic lines. Gentrification is understood at its
most basic level as the process by which large numbers of wealthy, high-SES residents move
into a low-income neighborhood and replace longtime residents of lower socio-economic
standing (Glass 1964; Lees 2003; Zukin 1995).
White and higher-income families are moving to and gentrifying certain neighborhoods
across the U.S., but little is known about whether this neighborhood revitalization will lead to a
greater presence of white students in public schools. Research is needed to understand how
neighborhood institutions, such as schools, are also lifted up or left behind as neighborhoods
revitalize. Despite a large body of gentrification research, we still know very little about how the
3

changing racial composition and socioeconomic status of neighborhoods is reflected in the racial
and socioeconomic profiles of one the most important neighborhood institutions: traditional
public schools.
One factor that may weaken the compositional link between neighborhoods and schools is
the availability of alternative schooling options. While charter, magnet, and private schools and
other choice schools that serve students from many neighborhoods have the potential to break the
neighborhood-school link for families of all socioeconomic groups, it tends to break the link for
higher-socioeconomic status (SES) families that are able to activate “choice” options (Renzulli &
Evans 2005; Saporito 2003; Saporito & Sohoni 2006). School choice policy may offer opt-outs
to middle-class parents residing in lower-SES urban school zones who hold reservations about
sending their child to its neighborhood school based on its racial composition or poverty rate
(Johnson 2015; Kimelberg and Billingham 2013; Oberti 2007). Other ethnographic research
finds that charter schools offer middle class families alternatives, particularly in neighborhoods
experiencing neighborhood upgrading (Keels et al. 2013; Makris 2015).
How individuals select neighborhoods and schools aggregates to shape spatial patterns of
inequality, and these decisions regarding neighborhood and school selection may partially
explain why cities have been durably segregated along racial and economic lines. Previous work
has demonstrated how race drives residential and school preferences. More white families,
relative to nonwhite families, prefer racially homogenous schools and neighborhoods
(Billingham and Hunt 2016; Saporito and Sohoni, 2006; Zubrinsky and Bobo 1996). At the same
time, earlier research has found that whites tend to hold negative stereotypes about nonwhite
groups, especially blacks (Charles 2003). While "white flight" has been identified in previous
eras as a driver of racial segregation, "white avoidance" (Ellen 2000) may also explain the
4

persistence of racial segregation in neighborhoods and schools (Sikkink and Emerson 2008). The
concept of white avoidance describes whites’ reluctance to move into majority black
neighborhoods. While white families with children may avoid black and minority neighborhoods
because of racial prejudices (Krysan 2002), they may also avoid these neighborhoods because
they perceive them to experience greater disinvestment, higher crime, and poorer institutional
quality, as racial proxy theories suggest (Harris 1999). Prior research observes this pattern also
playing out in schools as white middle-class parents’ schooling decisions tend, on average, to be
influenced by the presence of nonwhites (Billingham and Hunt 2016; Fairlie and Resch 2002;
Saporito 2009).
Academic scholarship has only recently begun examining how family decisions about
neighborhoods take school choices into account (Holme 2002; Lareau and Goyette 2014; Owens
2016). The handful of qualitative research examining residential decisions of middle-class
families in gentrifying neighborhoods has been mixed and limited to a single district or school.
While some find that middle-class families choose local schools because they value diversity and
other urban qualities (Billingham and Kimelberg 2013; Kimelberg 2014; Posey-Maddox 2014),
others discover underlying tensions between gentrifier parents’ desire for their children to be
exposed to racial and socioeconomic diversity and their actual behaviors when confronting social
mix at school (Kimelberg and Billingham 2013; Posey-Maddox et al 2014; Roda and Wells
2013; Stillman 2012).
Understanding the forces behind school and neighborhood change is important because both
contexts affect kids. Exposure to disadvantaged neighborhoods is negatively associated with
educational achievement and a host of quality-of-life indicators (Levanthal and Brooks-Gunn
2000; Sampson et al 2002; Chetty et al 2014). Exposure to school segregation is similarly
5

negatively associated with numerous social and economic indicators of well-being, producing
both short- and long-term effects (Entwisle and Alexander 1992; Johnson 2011; Rumberger and
Palardy 2005; Reardon 2015).
In the neighborhood effects literature, institutional resource theory views schools as key
neighborhood institutions and potential mediators of neighborhood effects (Galster 2012; Jencks
and Mayer 1990; Sharkey and Faber 2014). In past decades, when neighborhoods were more
tightly linked, the composition of a child’s neighborhood correlated more closely with the
composition of a child’s school. If the neighborhood-school link has weakened over time, and
schools are no longer key neighborhood institutions, there may be new implications in terms of
changing neighborhood effects on children’s educational outcomes.

Expected Contributions
This project makes several substantive contributions. First, this project contributes to a small
but emergent literature examining the changing relationship between neighborhoods and schools
over time. While previous research has documented the disparities between school segregation
and neighborhood segregation, nearly all studies examine this gap at a single point in time and
within a limited number of cities or districts. While important, these studies have largely ignored
how these processes emerge and play out over time in the context of urban and neighborhood
change. Addressing this gap, this dissertation examines how the changing socioeconomic status
of neighborhoods is reflected in the racial and socioeconomic profiles of schools. Given how
demographic changes are reconfiguring urban areas across the U.S., this research has important
implications for theory, as well as for urban, housing and educational policies.
6

Second, this dissertation examines the ways in which families sort into neighborhoods and
schools to understand how, if at all, neighborhood change may be driven by factors related to
how families make decisions about schools. Gentrification, as a particular type of neighborhood
change, offers a unique opportunity to observe the relationship between race, class, and
residential and schooling outcomes in racially and socioeconomically diversifying areas.
Surprisingly little empirical research—and almost no quantitative work— has documented the
relationship between gentrification and schools. My dissertation represents one of the first
quantitative studies exploring this relationship, and one of few national-level quantitative studies
linking residential and school enrollment outcomes in neighborhoods experiencing gentrification.
Third, while much research has examined how neighborhood context affects student
outcomes, the neighborhood effects literature has overlooked how these outcomes may differ
between different periods. This current project puts neighborhood effects in a full temporal
framework by modeling (1) cumulative exposure to neighborhood contexts and (2) changes over
time. My dissertation aims to identify whether neighborhood effects have changed over time as a
first step toward identifying whether any differences may be related to the changing relationship
between neighborhoods and schools. If neighborhood and school contexts matter, do they matter
for the same reasons today as compared to previous periods? To that point, my work also extends
the field by examining how micro processes, such as residential and school decisions, may be
shaped by macro contexts (e.g. demographic change; school choice expansion; rising income
inequality).
Fourth, I use innovative spatial and statistical techniques to assemble a novel, national-level
longitudinal dataset of school attendance boundaries. I combine multiple data sources of varying
geographic scales, thus allowing me to answer previously understudied questions about the
7

neighborhood-school link over time. Answering questions about how school-aged children in
gentrifying neighborhoods sort into schools also requires combining multiple data sources on
children, families, and neighborhoods with information on students’ assigned and enrolled
schools. Very little research has combined data in such a way, and those that do have been
limited to a single district or city. My national data matches school-age children to family,
school, neighborhood, and metropolitan contexts, thus allowing me to extend the geographic
scope and substantive reach in my work on residential and school linkages.
Fifth, this project examines neighborhoods and schools jointly and takes spatial and temporal
dynamics seriously, which few studies do, and analyzes dynamics of change in neighborhood
and school processes. I investigate how changes in one context affect changes in the other. This
has implications for policymakers— policies developed for schools may also impact
neighborhoods and residential contexts, and vice versa. Such is the case with school policies
(such as district choice programs) that affect neighborhood processes (i.e. residential sorting). To
that end, my dissertation will inform directly on policies that consider jointly the residential
contexts of households and the school environments that children attend—policies such as
economic development initiatives aiming to draw families back into urban cores and public
schools, district policies pursuing racial and socioeconomic integration, district policies about
choice implementation, affordable housing policy, and decisions about where to site schools.
On a theoretical level, I engage with and extend stratification theories of social closure and
residential/school sorting. My research sheds light on new inequality-producing mechanisms that
disrupt efforts to equalize the playing field while upholding the durable structure of opportunity
and inequality. While its internal architecture may change, the overall structure of inequality
remains stubbornly intact. As individuals adjust to shifts in these internal designs, the most
8

resourced groups often have the means to crowd out others from opportunities, thereby becoming
architects upholding the overall structure of inequality. We can observe how social closure and
systematic opportunity hoarding re-emerge in new ways by examining higher income parents’
anxious efforts to gain advantages for their kids. Such is the case when higher income parents
invest in a new kind of package deal by moving into gentrifying neighborhoods, but avoiding
their racially and socioeconomically diverse schools.

Research Questions
This dissertation is organized into three distinct parts consisting of three standalone empirical
chapters, described below.
Paper 1. Link between gentrification and local school composition.
Part One examines trends in racial and socioeconomic neighborhood change and its relationship
to schools, paying particular attention to how school and neighborhood composition correspond
in neighborhoods experiencing gentrification. For all analyses, I use a broad definition of
gentrification based on neighborhood ascent, which refers to neighborhood SES improvement as
measured by indicators commonly used in the quantitative literature (Owens 2012; Owens and
Candipan 2018). In this chapter, I build on recent work that finds a decoupling relationship
between neighborhood and school racial and SES composition from 2000 to 2010 (Hemphill and
Mader 2016). I am interested in understanding how school racial and SES composition changes
as neighborhoods change, particularly in schools serving gentrifying neighborhoods. I ask:
RQ1. Does charter, magnet and private expansion decouple neighborhood and school racial
composition?
RQ2. Is neighborhood-school racial composition decoupling the most in ascending
neighborhoods vs. non-ascending neighborhoods?
9

RQ3. Does charter expansion decouple neighborhood-school composition more in ascending
neighborhoods?
Paper 2. Link between neighborhoods and enrollment outcomes and school alternatives.
Part Two moves beyond aggregate-level analyses to model residential and school enrollment
outcomes of families with school-aged children. First, I examine elementary school enrollment
patterns among families in gentrifying and non-gentrifying tracts and predict the likelihood of
sending one’s child to a neighborhood school. I ask:
RQ4. Does a neighborhood’s school choice context predict whether parents opt out of
neighborhood schools?
RQ5. Are parents in gentrifying neighborhoods more likely to opt out of neighborhood
schools than parents in non-gentrifying neighborhoods?
RQ6. Are recent movers more likely to opt out of neighborhood schools, particularly in
gentrifying neighborhoods where newcomers and existing residents may differ both
demographically and in terms of their knowledge of local schools?
RQ7. Do parents who opt out enroll their children in schools with different racial/ethnic
composition than the neighborhood school?
Paper 3. Neighborhood effects over time on educational outcomes.
If schools and neighborhoods are becoming dissimilar in terms of their compositions, but
both are promoting inequality in some way, what are the implications for children’s educational
outcomes? Part Three examines how neighborhood effects on educational outcomes have
changed over time, then ponders whether this is because schools are increasingly not a
neighborhood institution. I ask:
RQ8. How have neighborhood effects on students’ academic achievement (as measured by
test scores) changed over time?
RQ9. How have neighborhood effects on students’ persistence (as measured by high school
graduation) changed over time?
10

Overview of Empirical Chapters
Part One. Neighborhood Change and the Enduring Neighborhood-School Gap.
The first part of my dissertation is motivated by a simple question: What happens to key
neighborhood institutions, such as schools, as neighborhoods change along demographic and
socioeconomic lines? To answer this question, I examine how school and neighborhood
composition correspond to one another over time. I then turn to potential explanations for this
growing mismatch, and my findings reveal the role of both school and neighborhood
explanations.
Chapter 2 (Paper 1) examines how the expansion of school choice is decoupling the
relationship between neighborhood and school racial composition and observes whether this
decoupling is greater in neighborhoods experiencing socioeconomic ascent. For decades, school
assignment in the U.S. has been based on where you live. Because of this tight link between
where you live and where you go to school, the most advantaged neighborhoods often have the
highest-resourced schools while least advantaged neighborhoods have lower-resourced schools.
This reciprocal relationship also means that segregated neighborhoods are often served by
segregated schools. Since the 1990s, this tight link between neighborhoods and neighborhood
schools has weakened. School choice, which severs the link between residential and schooling
decisions, has expanded in recent decades. Today more than a quarter of students attend a school
that is not the assigned neighborhood school, and that proportion is much higher in urban areas.
The expansion of school choice represents a changing educational market. Does it change how
parents choose schools and neighborhoods? The answer to this question has implications for the
composition of neighborhoods and neighborhood schools.
11

Chapter 2 combines neighborhood data from the decennial census and American Community
Survey, and school data from NCES Common Core of Data, with a unique longitudinal set of
school attendance boundaries for 46 of the largest and most diverse urban districts in 2000 and
2010. Drawing on this unique longitudinal dataset, I find a modest, but growing mismatch
between neighborhood and school racial/ethnic composition over time—neighborhoods are
increasingly composed of more white students than the local public school. The overall growth
in these neighborhood-school mismatches, however, masks a great deal of heterogeneity between
different types of neighborhoods. The compositional mismatch grows the most in neighborhoods
experiencing demographic change and socioeconomic ascent, particularly as the number of
nearby non-neighborhood schools increases, suggesting that higher-income white parents are
bypassing the local school while the neighborhood is still changing. Private school choice
decouples the neighborhood-school link in all neighborhoods. Taken together, results suggest
that neighborhood integration may still result in segregated schools, and that the demographic
diversity produced via neighborhood socioeconomic ascent or gentrification may still maintain
stratified educational experiences among children.
Findings from this chapter contribute more broadly to scholarship on how race and class-
based residential and school sorting upholds longstanding systems of inequality by way of
neighborhood and school stratification. Results further serve as a reminder that housing and
school policy are linked, and that efforts to integrate neighborhoods as a means of addressing
school segregation may not always work.

12

Part Two. Linking Residential and School Enrollment Decisions in Ascendant
Neighborhoods.
Recent research suggests that school choice and private school options have loosened the
historically tight relationship between neighborhoods and schools, particularly in
socioeconomically ascendant areas. Chapter 3 (Paper 2) of my dissertation investigates links
between residential and school enrollment outcomes to understand potential mechanisms driving
the mismatch between neighborhood and school composition. Do parents increasingly choose
neighborhoods and opt out of local schools? How do schooling decisions in gentrifying
neighborhoods compare to those in non-gentrifying neighborhoods? Gentrification offers a
unique opportunity to observe how the changing racial and socioeconomic profiles of
neighborhoods are reflected in key neighborhood institutions, such as schools. As a process,
gentrification complicates the usual narrative regarding residential sorting and neighborhood
segregation in the U.S., instead describing a process in which white and higher-SES households
move into lower-SES, minority neighborhoods. Observing how parents in gentrifying
neighborhoods choose schools may provide key insights into mechanisms that uphold school
segregation even while their surrounding neighborhoods diversify.
Despite mainstream narratives about how residential decisions are linked to schools,
academic scholarship has not fully undertaken this line of research. This paper examines the
ways in which families are sorted into neighborhoods and schools to understand how, if at all,
neighborhood change may be driven by factors related to how families make decisions about
schools. One way to understand how closely neighborhoods and schools are linked is to observe
whether families that have moved into demographically changing neighborhoods enroll their
children in neighborhood schools. I thus focus particularly on enrollment outcomes in
13

gentrifying neighborhoods where the inflow of white and higher-SES parents to lower-SES and
minority neighborhoods presents an interesting means of observing whether commonly
understood narratives of race- and income-based school preferences are upheld even when
families choose diverse neighborhoods.
This chapter builds on findings from the first, moving beyond aggregate-level analyses to
model residential and school enrollment outcomes of families with school-age children. In this
chapter, I examine how neighborhood characteristics and alternatives to local schools affect
where families choose to live and enroll their children. Specifically, I use logistic regression to
examine school enrollment patterns by neighborhood type, focusing on whether enrollment
outcomes among families in gentrifying tracts differ from patterns among families in
socioeconomically declining, low/mid-SES or stably upper-SES tracts. Importantly, I examine
whether the availability of nearby school choice options facilitates opt-outs differently in
neighborhoods undergoing gentrification. Combining restricted data from the Panel Study of
Income Dynamics with data from Census/ACS, the National Center for Educational Statistics
(NCES), and the 2014 School Attendance Boundary Survey (SABS14), I match school-age
children to their assigned and enrolled school to observe the likelihood that households recently
moving into socioeconomically ascendant neighborhoods will send their child to the
neighborhood school. Next, I observe which characteristics of the family, school and
neighborhood shape that decision. Finally, I examine how the availability of nearby choice and
private schools is associated with the odds of attending one’s local school, and how the adoption
of choice alternatives varies by neighborhoods experience various SES trajectories.
Findings suggest that school choice operates differently by neighborhood type. I find that
parents, especially recent movers, opt out of neighborhood schools more in gentrifying
14

neighborhoods when nearby school choice options are available. Nearby private school options
breaks the link between neighborhoods and schools for all neighborhood types. The overall
tentative conclusion is that a greater availability of non-neighborhood school options increases
odds of non-neighborhood school enrollment, particularly in gentrifying neighborhoods. These
findings complement results from aggregate-level analyses from Chapter 2 where I find that the
neighborhood-school compositional gap (i.e. the difference between neighborhood and school
racial composition) has increased the most in socioeconomically ascendant neighborhoods.
Findings are robust to alternative definitions and categorizations of gentrification.
Parents consider schools when making residential decisions, and since neighborhoods have
historically been segregated along economic and racial/ethnic lines, schools have also
experienced long histories of segregation. When neighborhoods begin to diversify, do schools
also follow suit? This study aims to understand how new ways of sorting into neighborhoods and
schools may produce new forms of inequality and marginalization in the context of urban
demographic change. Findings suggest that when there are choice options to opt-out of the
neighborhood school, in-migrating families to gentrifying neighborhoods will often activate
these options. Thus, even as neighborhoods begin to integrate (at least temporarily), schools may
remain segregated. These findings have implications for thinking about how urban and school
policy is linked, and how schools may be left behind even as neighborhoods are lifted up.
Broader implications of Parts One and Two
Chapters 2 and 3 extend research on residential mobility and neighborhood choice by taking
into account how changing neighborhood and school contexts intersect to shape households’
residential choices. Moreover, these studies speak more broadly to stratification research by
15

demonstrating how individual-level processes, such as decisions about where to live and attend
school, aggregate to create wider patterns of spatial inequality.
Although findings from the first two sections of this dissertation focus largely on gentrifying
and ascendant neighborhoods, these studies aim more broadly to understand the process of racial
transition and neighborhood change. These studies raise new questions about how race and class-
based residential and school sorting upholds existing racial and class hierarchies by way of
neighborhood and school stratification, shedding light on new inequality-producing mechanisms
that disrupt efforts to equalize the playing field while upholding the durable structure of
opportunity and inequality. Findings have implications for housing and school policies, including
affordable housing, neighborhood revitalization, school choice, neighborhood and school
integration, economic development, student assignment policies and many other urban and
education policies.
Part Three. Neighborhood Effects over Time on Educational Outcomes.
If schools and neighborhoods are becoming dissimilar in terms of their compositions, but
both are promoting inequality in some way, what are the implications for children’s educational
outcomes? Chapter 4 (Paper 3) first identifies whether the association between children’s
exposure to residential contexts on educational outcomes have changed substantially over time,
given social and demographic changes in neighborhoods. Then, building off findings from the
first two empirical chapters of this dissertation, I lay out a plan for future analyses to explore
whether this is because schools are increasingly not a neighborhood institution. While the first
two sections of my dissertation focus on how the relationship between housing and education
markets has weakened over time, the third section lays out a plan to explore potential
consequences of this decoupled relationship between neighborhoods and schools on children’s
16

educational outcomes. One reason neighborhoods are hypothesized to affect children’s
educational outcomes is because neighborhoods determine where a child attends school. If this is
decreasingly the case, do neighborhoods have a weaker effect on children’s outcomes in 2014
than they did twenty years earlier?
While much research has examined how neighborhood context affects student outcomes, the
neighborhood effects literature has ignored how these outcomes may differ between different
eras. In studying period and cohort effects, the idea is that differences in neighborhood effects by
period and age cohort reflect changes in larger societal contexts in recent decades. This injects a
temporal aspect to the study of neighborhoods and schools to show the extent to which
neighborhood effects have changed over time. Evidence of changes in neighborhood effects over
time could suggest that schools, which have been theorized as a key institution that mediates
neighborhood effects, are no longer as tightly linked to the areas they serve.
In Chapter 4, I review the literature on neighborhood effects and describe how past work has
incorporated temporal dimensions into their studies. I then lay out an analysis plan to investigate
whether exposure to disadvantaged residential contexts is stronger or weaker in 2014 than it was
in earlier decades. After presenting preliminary results modeling how the effect of cumulative
exposure to neighborhood disadvantage on academic achievement has changed over time, I then
describe plans for subsequent analyses that will test whether the decoupled relationship between
neighborhoods and schools is mediating these changes.
To answer questions about how neighborhood effects on children’s educational outcomes
have changed over time, I link administrative data on neighborhood characteristics to three
decades of data on households’ residential histories from the Panel Study of Income Dynamics
(PSID), a nationally representative longitudinal survey of households. The restricted PSID
17

Geomatch files allow me to observe households’ residential history across survey waves, and
enable me to link data on neighborhood features from the census to households’ residential
outcomes.
My analysis plan focuses on two types of educational outcomes: (1) one measure of
academic persistence, as measured by the odds of high school graduation; and (2) one measure
of academic achievement, as measured by test scores on the Woodcock-Johnson Revised Tests of
Achievement (WJ-R). To measure change over time, I pool data on two successive generational
cohorts of children from the PSID Childhood Development Supplements I & II (CDS-I and
CDS-II; first waves in 1997 and 2014, respectively). While the “intergenerational” design of the
PSID-CDS makes for a clean cross-cohort comparison, these cohorts are also ideal because they
represent an era before and after substantial school choice expansion (pre- and post-1997) and
coincide with a period during which the neighborhood-school relationship further decoupled
(Candipan 2019).
This chapter employs analytic models that incorporate the indirect effects of neighborhood
disadvantage on educational outcomes while also adjusting for possible confounding (due to
non-random residential selection into different neighborhoods across time) by time-varying
covariates. First, I use inverse probability of treatment weights (IPTW) on a marginal structural
model (MSM). Second, I compare results from MSM-IPTW with models using Regression with
Residuals (RWR) on a constrained structural nested mean model (SNMM), following recent
work on neighborhood effects (Sampson 2008; Wodtke et al. 2011; Wodtke 2018).
Substantively, this research will provide evidence on how the effect of exposure to
neighborhood disadvantage on children’s educational outcomes has changed over time.
Theoretically, this work casts a temporal framework on the neighborhoods effects literature and
18

speaks more broadly to questions about spatial and temporal dimensions of inequality. Future
planned analyses will contribute to our understanding of neighborhood effects and the
importance of schools as a neighborhood institution. Preliminary results from early models
suggest that the negative effects of childhood exposure to neighborhood disadvantage has
weakened slightly over time among white students, but has remained stubbornly persistent for
black students exposed to the most disadvantaged residential environments.

Broad Contributions
Broadly, these papers emerging from my dissertation contribute to our understanding of the
mechanisms upholding neighborhood and school segregation, the consequences of educational
policy and urban changes for neighborhoods and schools, and the processes of residential and
school choice, with implications for policies in the housing, urban, education, and economic
development arenas. My overall dissertation project also contributes to our understanding of how
contextual differences between periods (e.g. school choice expansion) shape individual processes,
such as residential choices, that aggregate into structural patterns of inequality.

19

References
Billingham, C. M. and M. O. Hunt. 2016. “School Racial Composition and Parental Choice:
New Evidence on the Preferences of White Parents in the United States.” Sociology of
Education 89(2):99–117.
Billingham, Chase M. and Shelley McDonough Kimelberg. 2013. “Middle-Class Parents, Urban
Schooling, and the Shift from Consumption to Production of Urban Space.” Sociological
Forum 28(1): 85–108.
Charles, Camille Zubrinsky. 2003. “The Dynamics of Racial Residential Segregation.” Annual
Review of Sociology 29(1): 167–207.
Ellen, Ingrid Gould. 2000. Sharing America’s Neighborhoods : The Prospects for Stable Racial
Integration. Harvard University Press.
Entwisle, DR., & Alexander, KL. 1992. "Summer Setback: Race, Poverty, School Composition
and Math Achievement in the First Two Years of School." American Sociological Review
57: 72-84.
Galster, George C. 2012. “The Mechanism(s) of Neighbourhood Effects: Theory, Evidence, and
Policy Implications.” Pp. 23–53 in Neighbourhood Effects Research: New Perspectives,
edited by M. Van Ham, D. Manley, N. Bailey, L. Simpson, and D. Maclennan.
Dordrecht, Neth.: Springer
Glass, R. 1964. Introduction: aspects of change. In London: Aspects of Change, ed. Centre for
Urban Studies, London: MacKibbon and Kee, xiii–xlii.
Harris, David R. 1999. “Property Values Drop When Blacks Move In, Because…Racial and
Socioeconomic Determinants of Neighborhood Desirability?” American Sociological
Review 64(3): 461– 479.
20

Hemphill, C., & Mader, N. 2016. Segregated schools in integrated neighborhoods: the city's
schools are even more divided than our housing. New School, Center for New York City
Affairs.
Jencks, Christopher and Susan E. Mayer. 1990. “The Social Consequences of Growing Up in a
Poor Neighborhood.” Pp. 111–86 in Inner-City Poverty in the United States, edited by L.
E. Lynn and M. G. H. McGreary. Washington, DC: National Academy Press.
Johnson, R. C. (2011). Long-run impacts of school desegregation & school quality on adult
attainments (No. w16664). National Bureau of Economic Research.
Keels, Micere, Julia Burdick-Will, and Sara Keene. 2013. “The Effects of Gentrification on
Neighborhood Public Schools.” City & Community 12(3): 238–59.
Kimelberg, S. M. and C. M. Billingham. 2013. “Attitudes Toward Diversity and the School
Choice Process: Middle-Class Parents in a Segregated Urban Public School District.”
Urban Education 48(2): 198–231.
Krysan, Maria. 2002. “Whites Who Say They’d Flee: Who are They and Why Would They
Leave?” Demography 39(4): 675–96.
Lareau, A & Goyette, K. eds. 2014. Choosing Schools, Choosing Neighborhoods: Residential
Segregation and the Search for a Good Schools. New York: Russell Sage Press.
Lees. 2003. “Super-gentrification: The Case of Brooklyn Heights, New York City.” Urban
Studies. Vol. 40(12): 2487-2509
Leventhal, T. and J. Brooks-Gunn. 2000. “The Neighborhoods They Live in: The Effects of
Neighborhood Residence on Child and Adolescent Outcomes.” Psychological Bulletin
126(2):309–37.
21

Makris, Molly Vollman. 2015. Public Housing and School Choice in a Gentrified City. New
York: Palgrave Macmillan.
National Center for Education Statistics. 2014. “State Support for School Choice and Other
Options, Tables 4.2-4.4.” Retrieved (http://nces.ed.gov/programs/statereform/sss.asp.).
Oberti, Marco and Amy Jacobs. 2007. “Social and School Differentiation in Urban Space:
Inequalities and Local Configurations.” Environment and Planning A 39(1):208–27.
Owens, A. 2012. “Neighborhoods on the Rise: A Typology of Neighborhoods Experiencing
Socioeconomic Ascent.” City & Community 11 (4), 345-369.
Owens, Ann. 2016. “Inequality in Children’s Contexts: Income Segregation of Households With
and Without Children”. American Sociological Review 81 (3):549-574
Owens, Ann and Jennifer Candipan. 2018. Racial/ethnic transition and hierarchy among
ascending neighborhoods. Urban Affairs Review, 1078087418770810.
Posey-Maddox, L. (2014). When Middle-Class Parents Choose Urban Schools: Class, Race, and
the Challenge of Equity in Public Education. Chicago, IL: University of Chicago Press.
Posey-Maddox, Linn, Shelley McDonough Kimelberg, and Maia Cucchiara. 2014. “Middle-
Class Parents and Urban Public Schools: Current Research and Future Directions.”
Sociology Compass 8(4):446–56.
Reardon, Sean F. 2015. “School Segregation and Racial Academic Achievement Gaps.” (CEPA
Working Paper No. 15-12). Retrieved (Stanford Center for Education Policy Analysis).
Renzulli, L., & Evans, L. 2005. School choice, charter schools, and white flight. Social
Problems, 52(3): 398–418.
22

Roda, Allison and Amy Stuart Wells. 2013. “School Choice Policies and Racial Segregation:
Where White Parents’ Good Intentions, Anxiety, and Privilege Collide.” American
Journal of Education 119(2):261–93.
Rumberger, R. W. and G. J. Palardy. 2005. “Test Scores, Dropout Rates, and Transfer Rates as
Alternative Indicators of High School Performance.” American Educational Research
Journal 42(1):3–42.
Sampson, R. J. 2008. “Moving to inequality: Neighborhood effects and experiments meet social
structure.” American journal of sociology 114(1), 189-231.
Sampson, Robert J., Jeffrey D. Morenoff, and Thomas Gannon-Rowley. 2002. “Assessing
‘Neighborhood Effects’: Social Processes and New Directions in Research.” Annual
Review of Sociology 28(1):443–78.
Saporito, Salvatore. 2003. “Private choices, public consequences: Magnet school choice and
segregation by race and poverty.” Social Problems, 50(2), 181–203.
Saporito, Salvatore. 2009. “School Choice in Black and White: Private School Enrollment
Among Racial Groups, 1990–2000.” Peabody Journal of Education 84(2):172–90.
Saporito, S. and Danesh Sohoni. 2006. Mapping educational inequality: Concentrations of
poverty among poor and minority students in public schools. Sociology of Education,
79(2), 81–105.
Saporito, S. and Danesh Sohoni. 2006. “Coloring Outside the Lines: Racial Segregation in Public
Schools and Their Attendance Boundaries.” Sociology of Education 79(2):81–105.
Saporito, Salvatore, Jana M. Chavers, Laura C. Nixon, and Megan R. McQuiddy. 2007. “From
Here to There: Methods of Allocating Data between Census Geography and Socially
Meaningful Areas.” Social Science Research 36(3):897–920.
23

Sharkey, Patrick and Jacob W. Faber. 2014. “Where, When, Why, and For Whom Do
Residential Contexts Matter? Moving Away from the Dichotomous Understanding of
Neighborhood Effects.” Annual Review of Sociology 40(1):559–79.
Sikkink, David and Michael O. Emerson. 2008. “School Choice and Racial Segregation in US
Schools: The Role of Parents’ Education.” Ethnic and Racial Studies 31(2):267–93.
Stillman, J. 2012. Gentrification and Schools: The Process of Integration When Whites Reverse
Flight. New York: Palgrave MacMillan.
Wodtke, G. T. 2018. “Regression-based adjustment for time-varying confounders.” Sociological
Methods & Research, 0049124118769087.
Wodtke, G. T., Harding, D. J., & Elwert, F. 2011. Neighborhood effects in temporal perspective:
The impact of long-term exposure to concentrated disadvantage on high school
graduation. American Sociological Review, 76(5), 713-736.
Zubrinksy, C. Lawrence Bobo, L. 1996. “Prismatic Metropolis: Race and Residential
Segregation in the City of the Angels.” Social Science Research 25(4): 335-374.
Zukin, Sharon. 1987. “Gentrification: Culture and Capital in the Urban Core.” Annual Review of
Sociology 13(1):129–47.

24

Chapter 2
Neighborhood Change and the Neighborhood-School Gap

Abstract

Few studies examine how school and neighborhood composition in the U.S. correspond over
time, particularly in a context of neighborhood change. As neighborhoods diversify along racial
and economic lines, do public schools also diversify or grow increasingly dissimilar from their
surrounding areas? Drawing on novel data linking neighborhoods and schools in the U.S. in 2000
and 2010, I document: how racial composition corresponds over time between traditional public
schools and the neighborhoods they serve; how the compositional gap changes when greater
school choice is available; and how the compositional gap varies between neighborhoods
experiencing various trajectories of socioeconomic change. I find an increasing mismatch in the
white composition of public schools and their surrounding neighborhoods, specifically that
schools enroll fewer white students than the composition of the neighborhood. The
compositional mismatch grows the most in neighborhoods experiencing socioeconomic ascent,
particularly as the number of nearby non-neighborhood schools increases.

25

Introduction

Historically, nearly all children in the U.S. attended their neighborhood school. Residence-
based school assignment systems created a strong link between a child’s home residence and
where she attended school. Given this tight relationship, a school’s demographic makeup often
closely resembled that of the neighborhood it served, and its economic and social resources were
intimately tied to its neighborhood. Since the 1990s, however, the historically tight link between
where families live and send their children to school has weakened (NCES 2014). This
neighborhood-school decoupling arises from the dramatic expansion of school choice, which
provides parents opportunities to bypass their zoned neighborhood school. In 2014, nearly a
quarter of all students attended a school other than their assigned school. The proportion of
families opting out of the neighborhood school is even greater in urban areas where more than a
third of all students attend non-assigned schools (Grady and Bielick 2010).
Some proponents of school choice argue that expanding non-neighborhood options can
“liberate” lower-SES and nonwhite students from traditional public schools in lower-resourced
neighborhoods. Critics, however, worry that the increasing provision of school choice provides
white and higher-SES parents greater opportunities to bypass schools with a higher proportion of
minority or low-income students, thus upholding school segregation, as recent work in the U.S.
(Renzulli & Evans 2005; Saporito 2003; Saporito & Sohoni 2006), UK (Benson et al. 2015;
Hamnett et al. 2013), and Europe (Noreisch 2007; Oberti 2007) suggests. As whites opt out of
neighborhood public schools with a higher presence of nonwhite and poor students, the racial
compositional match between neighborhoods and schools loosens—schools become less white
than their surrounding neighborhood.
26

The expansion of school choice also coincides with a period of white migration back to urban
cores (Ehrenhalt 2012). These trends intersect in key ways. School choice decouples decisions
about housing from neighborhood public schools, perhaps facilitating families’ moves to more
diverse neighborhoods. Enrollment of their children in more diverse schools, however, may not
follow. Past research finds that white families hold racially biased preferences, avoiding schools
with higher nonwhite or low-income students (Billingham and Hunt 2016). Therefore, the in-
migration of these families into racially and socioeconomically diverse urban neighborhoods
may not produce integrated schools if families choose charter, magnet or private schools over
their assigned neighborhood schools (Hemphill and Mader 2016). Given whites’ school
preferences, we might expect schools to become even more dissimilar from their neighborhoods
in diverse areas experiencing an influx of white families, such as socioeconomically ascendant
neighborhoods where the neighborhood public schools may still be largely nonwhite, lower-
resourced or lower-performing.
In this study, I examine how neighborhood and school composition of school-age children in
the U.S. correspond in socioeconomically changing neighborhoods. As the increasing provision
of school choice options weakens the housing-school nexus, will schools in socioeconomically
changing areas become more dissimilar to the neighborhoods they serve? As neighborhoods
diversify along racial and economic lines, will urban public schools also diversify or grow
increasingly dissimilar from their surrounding areas? Drawing upon multiple administrative
datasets on neighborhood and school characteristics, including spatial data on school attendance
boundaries, I construct a novel longitudinal dataset that links schools to the local community
from which they draw in 2000 and 2010. I find an increasing mismatch in the white composition
of public schools and their surrounding neighborhoods: public schools enroll fewer white
27

students than the composition of the neighborhood. The compositional mismatch grows the most
in neighborhoods experiencing demographic change and socioeconomic ascent, particularly as
the number of nearby non-neighborhood schools increases.
Background
The Reciprocal Relationship between Neighborhoods and Schools
Because neighborhoods have been durably segregated along racial and economic lines,
schools have also been segregated, since residential-based school assignments have historically
determined where children attend school. The most segregated and disadvantaged neighborhoods
are often populated with disadvantaged institutions, such as schools (Wilson 1987). Conversely,
the most affluent neighborhoods tend to have the most advantaged schools. For higher-income
families that can afford to buy into neighborhoods with higher-performing schools, residential
decisions in affluent neighborhoods become their own form of school choice (Lareau and
Goyette 2014).
However, since the 1990s, there have been great demographic shifts in neighborhoods and
schools. Immigration and demographic change since the 1990s have produced an increasingly
diverse school-age population in both neighborhoods and schools. Meanwhile, recent metro
migration patterns are reshaping the demographic and socioeconomic makeup of cities. White
and higher-SES families are returning to diverse urban cores and transforming the composition
of neighborhoods once characterized by disinvestment and white flight (Ehrenhalt 2012). An
increasing number of neighborhoods are undergoing socioeconomic ascent, which refers broadly
to neighborhood SES improvement over time, and in some cases gentrification (one particular
type of neighborhood ascent featured prominently in scholarly and public discourses) (Owens
2012). The ascent/gentrification narrative disrupts traditional discourses on racial residential
28

patterns—that whites prefer predominantly white neighborhoods and avoid heavily minority
(particularly black) neighborhoods (Charles 2003)—and represents a residential sorting process
that leads, at least temporarily, to integration along racial and economic lines (Brown-Saracino
2016). As white and higher-SES families move into diverse neighborhoods, how are these
demographic shifts translated into corresponding changes in school composition?
1

School Choice Expansion and School Preferences
One factor that weakens the compositional link between neighborhoods and schools is the
availability of school choice options that allow families to enroll their children outside of the
traditional public school in their assigned school attendance boundary. This includes magnet,
private, and charter schools, as well as student assignment policies enacted at the district level
(e.g. open enrollment; transfer programs; etc.). Private schools have long provided an alternative
to neighborhood schools, though high costs of attending largely restrict this opportunity to the
most advantaged families with greater economic resources (Murnane and Reardon 2017).
2

Magnet schools are public schools of choice, founded in the 1970s, with explicit aims of
integrating racially segregated schools. Charter schools, which are also publicly funded but
operate independently of public school districts, have expanded substantially in recent years and
now feature in public discourse as one of the most prominent market-based alternatives to non-
neighborhood schools (Berends 2015).
The uneven distribution of neighborhood school quality and choice options across districts
can either enable or limit participation in school choice. While school choice has the potential to

1
This study does not aim to make normative claims about integration, but rather identifies mechanisms that might change the
population. Moreover, while the costs of segregation are well documented, schools’ internal institutional norms vary, and some
work has shown how diverse student bodies do not always guarantee equal opportunities (Lewis-McCoy 2014; New and Merry
2014).
2
Some private schools, especially Catholic schools, serve low-income students but the population is increasingly overrepresented
with white and higher-SES students (Mickelson 2008).
29

break the neighborhood-school link for families of all socioeconomic groups, it tends to break
the link more often for white and higher-SES families that are able to activate choice options
(Bifulco et al. 2009; Renzulli & Evans 2005; Wells et al. 1999). While “white flight” has been
identified in previous eras as a driver of racial residential segregation, “white avoidance”—
whites’ reluctance to move into majority Black neighborhoods or enroll in majority Black
schools— may also explain the persistence of racial segregation in both neighborhoods and
schools (Ellen 2000). Previous research finds that a greater presence of nonwhite (particularly
black) students reduces the likelihood of white middle-class parents sending their children to
their neighborhood school (Fairlie and Resch 2002; Sikkink and Emerson 2008), even when
school test scores are equal (Billingham and Hunt 2016). While explicit racial bias could explain
white parents’ decisions to avoid nonwhite schools, their avoidance is also consistent with racial
proxy theory (Harris 1999; Krysan 2002), which asserts that whites use race as a proxy for
neighborhood quality. Past work suggests that middle-class white parents’ perceptions of school
quality drops as minority (particularly black) racial composition increases (Goyette et al. 2012;
Holme 2002). Furthermore, parents tend to rely on word-of-mouth for input regarding school
quality, and because parental networks are also segregated by race and SES, this results in
socially constructed understandings of “good schools” that are highly correlated with race more
so than test scores or other evaluative metrics (Holme 2002; Roda and Wells 2013).
Moreover, the local neighborhood context shapes the likelihood that parents activate choice.
Prior work finds that white middle-class parents are more likely to send their children to private
schools if they live in a neighborhood with more non-white neighbors (Fairlie and Resch 2002;
Saporito 2009). Despite magnet schools’ integrative aims, Saporito (2003) finds that white
students are more likely to apply to and enroll in magnet schools when the share of nonwhites in
30

their neighborhood increases. Whether parents activate charter school options may depend on
where these schools locate. While charter schools boast a large minority representation overall
(largely because many mission-based charter schools locate in durably disadvantaged urban
cores with disproportionate numbers of nonwhite and poor students), market-based charter
schools tend to locate in gentrifying areas and attract a more middle-class and less diverse
demographic than regular neighborhood schools (Burdick-Will et al. 2013; Candipan and Brazil
2018).
School choice options, especially charter schools, might be particularly appealing to parents
moving to historically disadvantaged, racially diverse urban neighborhoods, such as gentrifying
neighborhoods (Keels et al. 2013; Makris 2015). To the extent that white and higher-SES
families choose non-neighborhood schooling alternatives to avoid high-minority or low-income
student populations, we would observe persistent segregation in traditional public schools even if
the neighborhoods they serve become more racially and socioeconomically diverse. If, however,
white and middle-class families are drawn to cities because they value diversity and other urban
qualities in neighborhoods (Brown-Saracino 2016), these parents may choose more diverse
neighborhood schools, as recent qualitative work on gentrification and schools in the U.S. finds,
though these studies are limited to a single district or school (Posey-Maddox 2014). Still, other
work identifies underlying tensions between gentrifier parents’ expressed desire for their
children to be exposed to diversity and their actual behaviors when confronting social mix at
school (Kimelberg and Billingham 2013; Roda and Wells 2013). Research in UK and European
contexts uncovers similar tensions as middle-class families appraise school quality in gentrifying
neighborhoods based on a school's share of minority and lower-SES students, often seeking
market-based school choice alternatives, thereby reinforcing school segregation (Boterman 2012;
31

Noreisch 2007; Raveaud and van Zanten 2007). There is also recent evidence that neighborhood
ascent may be associated with greater probability of a charter school opening (Davis and Oakley
2013; Burdick-Will et al. 2013), while other work suggests increasing charter and choice options
may actually cause the socioeconomic ascent of low-income minority neighborhoods in urban
districts (Pearman and Swain 2017). Thus, substantial charter school expansion in recent years in
the U.S. may contribute to greater dissimilarity between school and neighborhood composition
because it provides middle-class white parents more opportunities to opt out.
The Changing Neighborhood-School Link
While previous research has documented disparities between school and neighborhood
composition, these studies examine this gap at a single point in time and within a limited number
of cities or districts. In this study, I document how racial composition in traditional public
elementary schools for 46 school districts in the U.S. corresponds with racial composition of the
neighborhoods they serve, and how this changes over time. I explore potential factors for this
growing mismatch, and my findings reveal the role of both school and neighborhood
explanations. The key school explanation I investigate is the increasing availability of school
choice. I hypothesize that growth in the neighborhood-school compositional gap will be more
pronounced in neighborhoods with a greater number of nearby magnet, charter and private
schools. The key neighborhood factor I investigate is neighborhood socioeconomic change, and I
hypothesize that growth in neighborhood-school compositional mismatches will be greatest in
areas experiencing socioeconomic ascent, where higher-educated and higher-income white
parents may bypass the local school while the neighborhood is still changing.

32

Data and Measures

I combine key demographic and socioeconomic neighborhood data from the census and American
Community Survey (ACS) with school data from the National Center for Educational Statistics’ (NCES)
Common Core of Data (CCD) and Private School Universe Survey (PSS) for 2000 and 2010. I create a
novel longitudinal dataset that links neighborhood and elementary school characteristics for 46 large and
diverse urban school districts in the U.S. In my sample, each school is: (1) uniquely associated with one
school attendance boundary (neighborhood); and (2) appears at both waves. The resulting dataset
includes 4255 4
th
grade school-neighborhood combinations in 2000 and 2010.
3

Unit of Analysis: School Attendance Boundaries (“Neighborhoods”)
I gather school attendance boundary information for two academic years: 1999-00 and 2009-
10. For 1999-00, I rely on available elementary school attendance boundary data collected by
Saporito and colleagues. The sample consists of large urban school districts that enroll a higher
share of black, Hispanic, and lower-SES students than the composition of their MSAs.
4
For
2009-10, I rely on School Attendance Boundary Information System (SABINS) data.
5
I use
geospatial processing techniques to normalize tabular census data and census geography to
school attendance boundaries (SABs), which become my operationalized “neighborhood” for all
empirical analyses (see online appendix). Using SABs as neighborhoods is the appropriate unit
for comparing neighborhood and school composition because it reflects the area from which a
school could potentially enroll students.

3
See appendix for characteristics of districts included in my sample.
4
Data obtained through a special-use contract with Salvatore Saporito.
5
The College of William and Mary and the Minnesota Population Center. Version 1.0. Minneapolis, MN: University of
Minnesota 2011. See sabinsdata.org
33

Dependent Variable: Neighborhood-School Gap (NS gap)
Informed by past research, I hypothesize that growing racial compositional mismatches between
neighborhoods and schools are driven by white families opting out of neighborhood schools. My
dependent variable thus measures how non-Hispanic white
6
racial composition corresponds between
schools and their surrounding neighborhoods. The full count decennial census reports race-by-age
counts at the block level for 2000 and 2010. To estimate neighborhood racial composition, I calculate
the white proportion of the 5-9-year-old population (the age of kindergarten to 4
th
grade students) in
each SAB from geographically apportioned census block data. For elementary school composition, I use
CCD data to estimate proportion white in kindergarten to 4
th
grades. I then estimate my dependent
variable, the neighborhood-school gap (NS gap), the difference in the share of white elementary school-
age children in neighborhoods and schools in a single year (2000 or 2010). The NS gap is scaled -100 to
100, with positive values indicating greater white racial composition in neighborhoods than schools
(e.g. NS gap of 4 indicates that a neighborhood’s white population share is 4 percentage points greater
than its school’s white population share).
Key Predictors
My key schooling-related predictors are proximity measures of school choice in 2000 and
2010. For each block within a given school attendance boundary (SAB), I estimate separately the
number of 4
th
grade charter, magnet and private schools that falls within a two-mile radius of its
centroid, then take the average for all blocks associated with each SAB. I use a two-mile radius
because past research indicates parents consider this a reasonable distance when assessing school
options (Denice and Gross 2016). The resulting school proximity measures are the number of

6
For the remainder of this paper, white refers to non-Hispanic white.
34

nearby non-neighborhood school options (i.e. magnet; charter; private schools) to the average
family’s residence.
My key neighborhood-related predictor is a typology of neighborhood SES trajectories. To
define neighborhood SES trajectories, I first construct SES scores for each neighborhood. I draw
on tract-level data from the long-form 1990 Census data and 2008-12 ACS data on five
indicators of neighborhood SES: median home value, median rent, median household income,
percent 25 years and older with at least a college degree, and percent 16 years and older in
managerial, professional or technical occupations (Owens 2012). I then use factor analysis, a
method used to reduce a number of correlated variables into a set of linearly uncorrelated
underlying dimensions, to construct standardized SES factor scores for each time point. I use
these scores to assign an SES percentile rank (0-100, with higher ranks representing tracts with
the highest SES) for each tract relative to others within the same metropolitan statistical area
(MSA, using 2003 OMB definition) with higher ranks representing tracts with relatively higher
SES. I measure SES over a two-decade period, starting in 1990, in order to capture a
neighborhood’s SES trajectory prior to the baseline year of my study, 2000, and identify
neighborhoods already observably changing to potential residents.
After calculating tract-level SES ranks, I use population weights to reapportion tract SES
ranks to the level of SABs. I categorize SABs as ascending, declining, upper-SES, or stable
based on their changes in SES rank over time. SABs beginning in the bottom four SES quintiles
in 1990 and increasing by at least ten percentage points in SES rank from 1990 to 2010 are
categorized as ascendant, while SABs that decline by ten percentage points or more are
identified as declining. Upper-SES SABs begin and end in the upper quintile in SES rank. Stable
35

SABs rank in the lowest four quintiles in 1990, and do not increase or decline in rank by more
than ten points. Neighborhood types are mutually exclusive categories and exhaustive.
Covariates
My analyses examine changes in school options and neighborhood demographics that might
affect the NS gap. The NS gap could also change over time because SABs themselves have been
redrawn in a way that alters their white racial composition. Districts may redraw SABs to
account for population shifts in neighborhoods, as may be the case in socioeconomically
ascendant attendance areas. Changes to catchment area boundaries may cause compositional
changes, either mechanically or by inducing mobility. Recent work finds that when boundaries
are redrawn to promote racial diversity, white families relocate to catchment areas with greater
white population share (Weinstein 2016).
To account for the potential effect of boundary change, I include a 3-category indicator
denoting whether the SAB surrounding a school was redrawn between 2000 and 2010 in such a
way that either increased or decreased the overall white racial composition in the SAB by at least
one percent (reference category is “no change”). To construct this measure, I overlay SAB
shapefiles from 2010 onto SAB shapefiles for the same schools in 2000. I match 2010 block
population to 2000 school attendance zones to create a hypothetical racial composition if a SAB
had not changed from 2000 to 2010. The difference between actual and hypothetical white racial
composition indicates the degree to which SAB proportion white changes as a result of school
boundaries being redrawn.
Models also control for various population and socioeconomic characteristics hypothetically
associated with how neighborhood and school composition correspond, based on prior work. At
the SAB level, I control for population density, measured as persons per square mile, land square
36

miles, and school-age population (both in thousands). I also account for school characteristics that
may shape white and middle-class families’ decisions about school enrollment. School poverty is
measured as the proportion of students in a school enrolled in the federal Free Lunch (FL) Program
in the prior year. Families whose incomes fall below 130 percent of the poverty line are eligible
for FL.
7
I also account for black school racial composition in the prior year, which past research
finds matters for whites’ explicitly race-based preferences or may act as a proxy for white parents’
perceptions of school quality and climate. School poverty and black racial composition are both
lagged one year to capture school characteristics prior to enrollment, and each are originally scaled
from 0-100. To ease interpretation, these four controls are expressed as deviations from their
respective sample means. Finally, I account for baseline neighborhood white racial composition
by constructing a time-constant binary indicator, initial neighborhood majority white, which
captures whether the white share of the under-18 population is greater than 75 percent in 2000.
Analytic Approach
I begin with descriptive analyses that identify changes in neighborhood-school racial
compositional mismatches over time and across ascendant, declining, upper-SES and stable
neighborhood types. I then move to longitudinal regression models, regressing the NS gap on
key predictors and covariates, which allow me to examine both school and neighborhood factors
associated with growth in the NS gap from 2000 to 2010.

School Choice
For mismatches in racial composition between neighborhoods and local schools to exist,
there must be alternatives to the neighborhood school. Thus, I use a fixed effect framework to

7
While an imperfect measure of school poverty, FL is often the only available indicator to parents who may rely on these data to
make decisions.
37

examine how an increase in the provision of school choice is associated with changes in the NS
gap between 2000 and 2010. All models include fixed effects for neighborhoods (SABs) which
account for unobservable time-invariant factors that may differ between school attendance
boundaries, and ensure that I only compare within-neighborhood change. Since schools are
uniquely associated with one SAB in this sample, the fixed effects also effectively account for
any omitted time-invariant school factors. I estimate the equation:

𝑁𝑆𝐺𝑎𝑝 𝑛𝑡
= 𝐶 ℎ𝑜𝑖𝑐𝑒 𝑛𝑡
+ 𝐵𝑜𝑢𝑛𝑑𝑎𝑟𝑦 𝑛𝑡
+ 𝑌𝑒𝑎𝑟 𝑡 10
+ 𝑀𝑎𝑗𝑊 𝑡 00
∗ 𝑌𝑒𝑎𝑟 𝑡 10
+ 𝑋 𝑛𝑡
+ 𝛾 𝑛

My outcome variable is the neighborhood-school gap, NSGapnt, measured as the difference
between neighborhood and school white racial composition (elementary-age) of neighborhood n in
yeart (2000 or 2010). My key predictor, Choicent, is a vector of time-varying school choice
proximity measures that separately calculate proximity of: 1) magnet; 2) charter; and 3) private
elementary schools to neighborhoodn in yeart. Coefficients for school choice report whether
growing choice options are associated with greater mismatches between neighborhood and school
racial composition. I control for whether school attendance boundaries were redrawn in a way that
changed the proportion white from 2000 to 2010 (Boundarynt). Xnt is a vector of time-varying
covariates (described earlier), while MajWnt is a time-invariant measure indicating whether a
neighborhood is majority white at baseline, which I interact with Year10 to conform to the fixed
effects framework. The coefficient for this interaction indicates how initial majority white status
contributes to growth in the NS gap over time. All models include neighborhood fixed effects, yn,
and clustered standard errors (by district).

38

Neighborhood SES Trajectory
I hypothesize that the NS gap will grow the most in ascendant neighborhoods where white
and higher-SES parents are moving into historically disadvantaged areas that may be served by
lower-performing schools. The following models test whether change in the NS gap is higher in
ascendant neighborhoods compared to other neighborhood types. I first estimate whether change
in neighborhood SES rank is associated with growth in the NS gap using the full model:
𝑁𝑆𝐺𝑎𝑝 𝑛𝑡
= 𝑆𝐸𝑆𝑅𝑎𝑛𝑘 𝑛𝑡
+ 𝐶 ℎ𝑜𝑖𝑐𝑒 𝑛𝑡
+ 𝐵𝑜𝑢𝑛𝑑𝑎𝑟𝑦 𝑛𝑡
+ 𝑌𝑒𝑎𝑟 𝑡 10
+
𝑀𝑎𝑗𝑊 𝑡 00
∗ 𝑌𝑒𝑎𝑟 𝑡 10
+ 𝑋 𝑛𝑡
+ 𝛾 𝑛

Then I move from simple rank to distinguishing between neighborhood SES trajectory types
using the full model:
𝑁𝑆𝐺𝑎𝑝 𝑛𝑡
= 𝐶 ℎ𝑜𝑖𝑐𝑒 𝑛𝑡
+ 𝐵𝑜𝑢𝑛𝑑𝑎𝑟𝑦 𝑛𝑡
+ 𝑌𝑒𝑎𝑟 𝑡 10
+ 𝑇𝑦𝑝𝑒 𝑛 ∗ 𝑌𝑒𝑎𝑟 𝑡 10
+
𝑀𝑎𝑗𝑊 𝑡 00
∗ 𝑌𝑒𝑎𝑟 𝑡 10
+ 𝑋 𝑛𝑡
+ 𝛾 𝑛

where Typen is a four-category variable denoting whether a neighborhood is ascending
(reference group), declining, upper-SES or stable. Since neighborhood type is time-invariant, I
interact this measure with a binary indicator for the year 2010 (Yeart10), and the coefficient for
this interaction indicates how growth in the NS gap differs between neighborhood types.
Findings
Descriptive Analyses: Changes in the NS Gap from 2000 to 2010

Table 1 presents the average NS gap in 2000 and 2010 (top rows) and average proportion
white in neighborhoods and schools (bottom rows). Among all neighborhoods, the average share
of elementary-age white residents in 2000 is 34.2 percent, reflecting my sample of large, urban
districts. Across all schools, the average proportion of white K-4 students declines from 30.9 to
22.9 percent from 2000 to 2010.
39

On average, the NS gap increases from 2000 to 2010 by about one percentage point: from
about 3 to about 4 percentage points (Appendix Figure 2.1). While there is heterogeneity across
the 46 districts in my sample, the NS gap is positive in 2010 in nearly all districts, indicating
greater shares of white children in neighborhoods compared to schools (Appendix Figure 2.2).
The modest average increase obscures a great deal of heterogeneity between neighborhood SES
types. As Table 2.1 shows, the NS gap in 2000 is smallest in declining and stable neighborhoods,
indicating a tighter neighborhood-school link, while it grows least (.51) in stable neighborhoods.
The NS gap is highest in upper-SES neighborhoods in 2000, but it narrows substantially (3.3
points) from 2000 to 2010. The NS gap grows the most (by over 3 points) and is largest in 2010
in ascendant neighborhoods (Appendix Figure 2.3).
Changes to the NS gap can occur due to changes in the neighborhood or school composition.
Table 2.1 shows that, in declining neighborhoods, the white share of elementary-age children
decreases substantially in both neighborhoods and schools from 2000 to 2010, but the decline is
greater in schools than neighborhoods, resulting in an increase in the NS gap of 1.8 percentage
points. By contrast, the white neighborhood share declined only slightly in ascendant
neighborhoods between 2000 and 2010, the least of any neighborhood type, while the proportion
of white students in elementary schools declined more substantially as the higher-SES white
population in the neighborhood grew. Thus, while the NS gap grew in both ascendant and
declining neighborhoods, the reason underlying this change is substantively different. These
descriptive trends provide preliminary support that the growth in the NS gap may be driven
partly by residential and school sorting patterns in ascendant neighborhoods.
Table 2.1 also presents descriptive statistics for key covariates and controls. My study
period—2000 to 2010—coincides with an expansion of school choice, reflected in the increase in
40

the number of magnet and, particularly, charter schools located within two miles of a given
neighborhood. Charter expansion is especially pronounced in ascendant neighborhoods where
growth is more than ninefold (from roughly .26 to 2.40). Growth in magnet availability is much
more modest—it increases from an average of .38 magnets in 2000 to .61 in 2010 overall—and
the modal outcome of nearby magnets in both years is zero. On the other hand, while private
schools outnumber magnets and charters, the number of nearby private schools decreases slightly
between 2000 and 2010 across all neighborhood types. About 27 percent of all zones are redrawn
between 2000 and 2010 in such a way that increases (13.3 percent) or decreases (14 percent) the
overall share of white residents by at least one percent. Notably, redrawing boundaries in upper-
SES neighborhoods increases the share of white school-age children substantially (2.5 percent).

Predicting the NS Gap

School choice
For racial compositional mismatches to occur between neighborhoods and schools, there
must be non-neighborhood school alternatives available that allow students to bypass their zoned
neighborhood school. I first examine how growth in the number of nearby magnet, charter and
private schools is associated with growth in the NS gap. I hypothesize that the prevalence of
nearby non-neighborhood schools will increase the compositional mismatch between white
school-age children in neighborhoods and schools.
Table 2.2, Model 1 regresses the NS gap on neighborhood and school demographic and
economic controls—population density, land area, neighborhood school-age population, school
poverty, black school racial composition, and initial majority white neighborhood composition.
Because continuous control variables are centered at their sample means, the conditional effect
of Year10 represents growth in the NS gap for the average non-majority white neighborhood. For
41

the average non-majority white neighborhood, the neighborhood-school gap is about 3.7 in 2000,
and it increases by 1.7 points in 2010, all factors considered. This growth in the NS gap,
however, obscures a great deal of heterogeneity across neighborhoods, which I revisit in later
models.
The magnitude and direction of the coefficients for control variables are consistent across
Models 1-3. Increases in school poverty and school black racial composition are both associated
with increases in the NS gap, consistent with research that finds that white and higher-SES
parents avoid black and high-poverty schools. Growth in the NS gap is roughly 4 points lower in
neighborhoods that were majority white in 2000, suggesting that schools more closely resemble
their neighborhoods in more racially isolated white neighborhoods.
Model 2 adds my indicator for school rezoning, which denotes whether a school attendance
boundary has been redrawn in ways that increase or decrease its share of white residents. All
factors considered, the coefficients for boundary change indicate that the NS gap narrowed in
neighborhoods where school zones were redrawn in ways that decreased the overall share of
white residents (-2.1) but increased when zones were redrawn in ways that increased white
presence (4.0). There are many ways by which rezoning may drive compositional mismatches
between neighborhoods and schools. If school rezoning assigns white families to relatively
nonwhite neighborhoods to address racial imbalance, this may ultimately motivate white parents
to seek non-neighborhood school options. While beyond the scope of this paper, the complex
question of how redrawing school attendance boundaries contributes to growing neighborhood-
school compositional mismatches warrants further exploration.
Model 3 introduces my key covariates—school choice proximity measures —to examine
how the proliferation of nearby non-neighborhood options is associated with growth in the NS
42

gap. Does school choice break the link between neighborhoods and schools? As hypothesized,
nearby private and charter options are both statistically significant and positively associated with
the NS gap, net of model controls. One additional nearby private school is associated with a .41
point increase in the NS gap, while one additional charter school is associated with a .24
increase. Magnet school availability, though positively associated with the NS gap as
hypothesized, is not statistically significant. Magnets are more specialized options with a
relatively smaller presence, having expanded substantially less than charters, so they may not
drive mismatches like other choice options.
Results provide support for the hypothesis that greater availability of nearby charter and
private schools contributes to the NS gap, creating more demographically dissimilar
neighborhoods and schools. When the availability of charter and private choice increases, local
traditional schools become less white than the neighborhoods they serve, implying that when
white parents can opt out of neighborhood schools more easily, they do, driving growing
mismatches between the share of white children in neighborhoods and schools.
Neighborhood SES Change
Next I examine whether there are differences in these dynamics across neighborhoods
experiencing various trajectories of socioeconomic change. Prior work finds that higher-SES and
white parents are moving into socioeconomically improving neighborhoods that are more
diverse, but may not be integrating their schools (Bischoff and Tach 2018; Candipan 2019;
Hemphill and Mader 2016). Thus, there is reason to expect that the influx of white and higher-
SES parents into socioeconomically improving neighborhoods provides the setting to observe
greater growth in the NS gap when school choice options are available.
43

Does the neighborhood-school compositional link weaken as neighborhood SES increases?
Model 4 examines how changes in a neighborhood’s SES rank (relative to other neighborhoods
within the same metro) contributes to growth in the NS gap. As hypothesized, an increase in
neighborhood SES rank is statistically significantly associated with an increase in the NS gap.
When a neighborhood’s SES percentile rank increases by 5 points, the NS gap increases an
additional 3.4 points. Here, the coefficient for charter proximity remains in the hypothesized
direction, but loses statistical significance. It could be that charter school adoption varies by
neighborhood type, which I explore in subsequent models.
These results provide preliminary support for the hypothesis that schools are becoming more
dissimilar to the neighborhoods they serve when neighborhood SES increases. Neighborhoods
experiencing substantial socioeconomic improvement are generally those into which higher-SES
and white parents are moving, and results here suggest that these parents are increasingly
avoiding their neighborhood schools.
Although Model 4 shows that the NS gap grows as neighborhood SES increases, we might be
concerned that the effect of change in SES rank is larger or smaller at the upper and lower
bounds of the rank distribution. We might also expect patterns of choice adoption to play out
differently in ascendant, declining, upper-SES and stable neighborhoods. Parents residing in
declining neighborhoods may have fewer resources and more constraints that prohibit choice
adoption. Stable neighborhoods do not experience substantial demographic or socioeconomic
change, so one may expect less growth in neighborhood-school compositional mismatches.
While private school choice has long been an option adopted by higher-SES parents, higher-SES
neighborhoods generally have the most advantaged neighborhood schools, so we might find
44

parents in these neighborhoods choosing zoned schools over public choice options like charter
schools.
Table 2.3 thus examines whether growth in the NS gap is uniform across four types of
neighborhoods experiencing various SES trajectories: ascendant; declining; upper-SES and
stable. Model 1 examines growth in the NS gap in these neighborhood types, net of
neighborhood and school controls. Since ascent is my reference category for neighborhood type,
the conditional effect for Year10 estimates the decadal growth in the NS gap for an average
(initially) non-majority white ascendant neighborhood whose school boundaries were not
redrawn (coef=4.3; p<.001). While ascendant, stable and declining non-majority white
neighborhoods all experience growth in the NS gap from 2000 to 2010, the negative coefficients
for these neighborhood types indicate that the NS gap increase is substantially larger in
ascendant neighborhoods, all factors considered.
Model 2 combines school choice proximity and neighborhood type measures. As with school
choice models from Table 2.2, the coefficient for private schools is positive and significant,
suggesting that greater availability of nearby private school options increases the NS gap across
all neighborhoods. Conversely, the coefficient for charters is non-significant after accounting for
neighborhood types. Recall, however, that the charter coefficient from Table 2.2 (Model 4) was
also positive, but lost statistical significance when accounting for neighborhood SES, suggesting
that the charter main effect (Table 2.2, Model 3) may be masking differences between
neighborhood types in terms of how charter availability contributes to the NS gap.
Charter adoption may play out differently in neighborhoods experiencing socioeconomic
change, which I explore in Model 3. I introduce an interaction between charter proximity and
neighborhood type to test whether the expansion of charter availability widens or narrows the NS
45

gap more in ascendant neighborhoods with growing charter options. In this model, the main
effect for charter proximity is positive and significant, indicating that growth in nearby charters
is associated with higher NS gaps in ascendant neighborhoods. Interestingly, the interaction
between charter proximity and neighborhood type results in negative coefficients for declining,
upper-SES and stable neighborhoods, indicating that increasing charter availability decouples
neighborhood and school composition the most in ascendant neighborhoods (further illustrated in
Appendix Figure 2.4). In ascendant neighborhoods, one additional nearby charter school is
associated with a .37 point increase in the NS gap. Combining the charter main and interaction
terms (with neighborhood type) in Model 3 shows that the NS gap grows only minimally in
stable neighborhoods per each additional charter school (i.e. (.37-.30) =.07 points). In upper-SES
neighborhoods, one additional charter school corresponds to a .31 point decrease in the NS gap
((.37-.68) = -.31). Charter growth decreases the NS gap in declining neighborhoods, though the
effect is only trend-level significant. For upper-SES neighborhoods, white families from more
advantaged neighborhoods may choose zoned schools over charter schools, which past research
finds. Stable neighborhoods are not experiencing substantial demographic or SES change, so
greater school choice occurs in a different context than in ascendant neighborhoods, where
school choice provides newcomers options to avoid the local school. Overall, results by
neighborhood type reveal nuances that are obscured when looking strictly at average growth in
the NS gap.
Together, findings suggest that most schools became increasingly less white than their
neighborhoods from 2000 to 2010, particularly in socioeconomically ascendant neighborhoods.
School choice is the key factor that creates NS composition gaps, and the interaction models
indicate that when choice is more readily available in ascendant neighborhoods, the NS gap is
46

even larger. White parents may opt out of schools that are historically lower-performing or
predominantly non-white, and a greater provision of school choice provides these parents the
opportunity to search for more desirable schools outside of their assigned school zone.
Discussion
White and higher-income families are moving in and diversifying the composition of certain
neighborhoods across the U.S., but little is known about whether these neighborhood
demographic changes will lead to a greater presence of white students in public schools. As
neighborhoods change, do key neighborhoods institutions such as schools also change? Using a
novel longitudinal dataset that links schools to the local community from which they draw in
2000 and 2010, this study of 46 urban districts finds that when neighborhoods improve along
socioeconomic lines, schools become increasingly dissimilar to the neighborhoods they serve—
neighborhoods become whiter than their local assigned schools. Policies such as school choice
further break the link between neighborhood and school composition, instead allowing those
with greater advantages to pull further ahead.
Results indicate that the in-migration of white and higher-SES families into ascendant
neighborhoods may produce demographic change in neighborhoods while simultaneously
upholding segregation in schools. While white families returning to cities may express tastes for
diversity in some arenas, they more often refuse to send their children to majority-minority
schools. While my data do not allow me to observe the motives underlying white parents’
choices, findings align with prior work that suggests racist bias could be a motivating factor.
This has implications for policies that aim to integrate neighborhoods as a way of addressing
school segregation, since my findings suggest that neighborhood demographic change does not
necessarily produce school integration. The sort of de facto diversity produced via neighborhood
47

ascent does not produce the same demographic changes in schools, but instead helps maintain
racially and socioeconomically stratified educational experiences among children. One concern
is that ascendant neighborhoods will continue along a trajectory towards being stably upper-SES,
with white families re-entering zoned neighborhood schools once demographics tip to their
liking. That upper-SES neighborhoods begin with higher compositional mismatches in 2000, but
see a decline over time (Table 1) suggests that some may be previously ascendant
neighborhoods.
This study has limitations that future work should address. First, that the NS gap grew over
time net of school choice options and rezoning indicates that white parents are opting out of
assigned schools in additional ways. Some district programs that allow students to enroll in non-
neighborhood schools within and outside one’s district (e.g. inter- and intra-district transfer
programs; open enrollment) have also expanded dramatically since the 1990s (Brunner et al.
2013), and one study finds that these programs served more students in 2012 than all magnet,
charter and voucher students combined (Lavery and Carlson 2012). Voucher and ability
programs (e.g. Gifted and Talented) also provide opt-outs. Comprehensive school-level data for
such programs over time are sparse, so my analyses are unable to explicitly examine these
additional schooling alternatives. Second, while results suggest that higher-SES white parents’
racially biased decisions are driving the growth in compositional mismatches in ascendant
neighborhoods, my data do not allow me to directly examine the individual choices that parents
make about neighborhoods and schools. Future research should move beyond aggregate-level
analyses to explicitly test individual-level factors that drive choices about neighborhoods and
schools. Next, I find that school rezoning contributes to the NS gap when it alters the racial
composition of the neighborhood. How rezoning influences enrollment and residential decisions
48

is beyond the scope of this paper, and future research should explore the complex and varied
associations between school rezoning, segregation, and neighborhood and school composition.
Finally, while I focus on neighborhood SES, there are many ways by which neighborhoods are
stratified, including along racial/ethnic lines. Future work should explore demographic processes
accompanying school choice expansion in neighborhoods undergoing other types of change,
such as those that transition to and from various types of mixed racial/ethnic configurations.
Overall, these findings have implications for theories of neighborhood and school
stratification, and raise new questions about how residential and school sorting upholds existing
racial hierarchies in an era of increasing choice and neighborhood change. While much work has
documented how rising segregation among families with school-age children between school
districts perpetuates one level of inequality, this study documents how inequality is also
produced within districts.

49

Tables

Table 2.1. Neighborhood and School Characteristics (Overall and By Neighborhood Type)

Overall Ascendant Declining Upper-SES Stable

N=4255 N=617 N=987 N=276 N=2375
Measure Mean SD Mean SD Mean SD Mean SD Mean SD
Neighborhood-School Gap

2000

NS Gap (%White 5-9) 3.36 10.05 4.54 11.42 1.45 8.59 8.99 15.54 3.20 9.11

2010

NS Gap (%White 5-9) 4.33 9.66 7.90 12.82 3.22 7.54 5.66 12.57 3.71 8.84

Change in NS Gap, 2000-2010 0.97 8.44 3.36 9.69 1.78 7.23 -3.33 11.07 0.51 7.96

School Choice

2000

Number of Magnets < 2m 0.38 1.24 0.27 0.77 0.16 0.70 0.28 0.94 0.51 1.50

Number of Charters < 2m 0.27 0.75 0.26 0.77 0.19 0.55 0.12 0.35 0.32 0.84

Number of Privates < 2m 6.59 9.77 9.76 13.67 4.06 6.19 5.97 10.15 6.89 9.50

2010

Number of Magnets < 2m 0.61 1.25 0.48 1.05 0.43 0.99 0.38 0.74 0.74 1.41

Number of Charters < 2m 1.46 2.25 2.40 3.05 0.64 1.06 0.74 1.41 1.63 2.32

Number of Privates < 2m 5.75 9.31 8.99 13.24 3.61 5.53 5.82 9.89 5.79 8.96

Neighborhood SES

SES Rank in 1990 (0-100) 42.3 27.8 30.9 20.8 57.4 19.9 89.9 5.0 33.5 25.2

SES Rank in 2010 (0-100) 40.9 26.6 53.0 22.0 39.3 20.3 89.0 5.1 32.8 24.3

School Controls

% School Poverty (lag -1yr) 55.8 28.7 60.9 28.3 49.0 25.4 16.0 16.5 62.0 26.7

% School Black (lag -1yr) 28.9 31.5 29.0 30.9 28.1 28.8 10.6 13.7 31.3 33.4
School Boundary Change

% Rezoned (decreased %white) 14.0

19.0

16.6

11.2

11.9

% Rezoned (increased %white) 13.3 15.4 15.3 21.0 11.0

Change in %White from Rezoning -0.22

-0.45

-0.34

2.45

-0.50

Neighborhood. (SAB) Controls

School-age Population (5-17) 1599 855 1362 855 1711 798 1570 734 1617 879

Pop. Density (in thousands) 148 221 204 289 108 173 123 285 154 205
Land Square Miles 1.9 2.9 2.0 3.7 2.0 2.3 3.5 4.2 1.6 2.7

% Majority White in 2000 13.4

13.9

9.8

51.1

10.4

Racial Composition

2000

SAB % White (5-9) 34.2 30.2 33.2 31.4 38.3 25.8 73.9 13.8 28.1 29.4

School % White (KG-4) 30.9 30.5 28.7 31.6 37.0 28.1 65.0 22.9 25.0 28.9

2010

SAB % White (5-9) 27.1 26.6 31.7 26.9 26.5 22.5 66.5 15.1 21.7 25.0

School % White (KG-4) 22.9 26.2 23.8 27.3 23.3 23.2 60.9 20.4 18.0 24.0
Notes: Neighborhoods are defined as school attendance boundaries (SABs); Population density refers to total persons per
square mile. Neighborhoods types are mutually exclusive and exhaustive.

50

Table 2.2. Fixed Effects Regression Results Predicting the Neighborhood-School Gap Conditional on School Choice
and Neighborhood SES

Model 1 Model 2 Model 3 Model 4
Decadal Change

Year = 2000 (ref)

Year = 2010 1.681*** 1.461*** 1.493** 1.744***

(0.391) (0.387) (0.479) (0.477)
School Choice

Number of Magnet <2 Miles 0.199 0.196

(0.292) (0.296)

Number of Private <2 Miles 0.408** 0.390**

(0.134) (0.128)

Number of Charter <2 Miles 0.235* 0.118

(0.108) (0.107)
Neighborhood SES

SES Relative Rank Change

0.067*

(0.023)
School Rezoning

No Boundary Change from 2000 to 2010 (ref) (ref) (ref) (ref)

Boundary Change (decrease % white) -2.136** -2.221** -2.312**
(0.688) (0.687) (0.697)

Boundary Change (increase % white) 4.036*** 3.941*** 3.832***

(0.611) (0.594) (0.579)
School Controls

Lagged (-1 Yr) School Poverty 0.112*** 0.112*** 0.111*** 0.123***

(0.024) (0.024) (0.023) (0.025)

Lagged (-1 Yr) School % Black 0.182*** 0.190*** 0.189*** 0.186***

(0.027) (0.025) (0.025) (0.026)
Neighborhood Controls

Density (in thousands) -0.001 -0.000 -0.001 -0.001

(0.003) (0.003) (0.003) (0.003)
Land Square Miles -0.117 -0.113 -0.107 -0.082
(0.081) (0.086) (0.086) (0.088)

Population 5-17 (in thousands) -0.359 -0.373 -0.442 -0.501

(0.383) (0.370) (0.381) (0.376)

Majority White in 2000 -4.091*** -4.136*** -4.160*** -4.263***

(0.790) (0.788) (0.808) (0.820)

Constant 3.655*** 3.664*** 0.840 1.541

(0.196) (0.195) (.994) (1.003)

N 8510 8510 8510 8510
R2 0.838 0.845 0.847 0.849

Notes: Longitudinal regression with SAB fixed effects; Neighborhood-School Gap refers to the difference in
percent non-Hispanic white (5-9 yrs) in neighborhoods vs. schools (grades KG-4); Continuous control variables
centered at sample means; All values for boundary change are set to zero in 2000; All models include robust
standard errors (clustered by district).

* p<0.05, ** p<0.01, *** p<0.001

51

Table 2.3. Fixed Effects Regression Results Predicting the Neighborhood-School Gap Conditional on
School Choice and Neighborhood SES Type

Model 1 Model 2 Model 3
Decadal Change

Year = 2000 (ref) (ref)

Year = 2010 4.305*** 4.311*** 3.810***

(0.682) (0.772) (0.895)

Neighborhood SES

Ascendant x Year 2010 (ref) (ref) (ref)

Declining x Year 2010 -2.921*** -2.821*** -2.074*

(0.778) (0.802) (0.962)

Upper-SES x Year 2010 -5.055*** -5.062*** -4.270**

(1.274) (1.277) (1.407)

Stable x Year 2010 -3.309*** -3.096*** -2.510**

(0.670) (0.669) (0.803)
School Choice

Number of Magnet <2 Miles 0.181 0.180

(0.283) (0.282)
Number of Private <2 Miles 0.381** 0.371**
(0.125) (0.123)

Number of Charter <2 Miles 0.129 0.368*

(0.100) (0.147)

Charter x Ascendant

(ref)

Charter x Declining

-0.753+

(0.421)

Charter x Upper-SES

-0.679*

(0.272)

Charter x Stable

-0.303*

(0.152)
School Rezoning

No Boundary Change from 2000 to 2010 (ref) (ref) (ref)
Boundary Change (decrease % white) -2.361*** -2.457*** -2.472***
(0.675) (0.679) (0.676)

Boundary Change (increase %white) 3.976*** 3.850*** 3.815***

(0.590) (0.571) (0.572)
School Controls

Lagged (-1 Yr) School Poverty 0.114*** 0.111*** 0.111***

(0.024) (0.023) (0.023)

Lagged (-1 Yr) School % Black 0.189*** 0.187*** 0.189***

(0.026) (0.026) (0.027)
Neighborhood Controls

Density (in thousands) -0.001 -0.001 -0.002

(0.003) (0.003) (0.003)
Land Square Miles -0.073 -0.064 -0.074
(0.092) (0.092) (0.092)

Population 5-17 (in thousands) -0.372 -0.446 -0.419

(0.368) (0.377) (0.378)

Majority White in 2000 x Year10 -3.741*** -3.798*** -3.826***

(0.799) (0.812) (0.809)

Constant 3.676*** 1.071 1.159

(0.192) (0.940) (0.931)

N 8510 8510 8510
R2 0.849 0.851 0.851

Notes: Longitudinal regression with SAB fixed effects; Neighborhood-School Gap refers to the difference in percent
non-Hispanic white (5-9 yrs.) in neighborhoods vs. schools (grades KG-4); Continuous control variables centered at
sample means. All values for boundary change are set to zero in 2000. All models include robust standard errors
clustered by district.

* p<0.05, ** p<0.01, *** p<0.001

52

References
Benson, M., Bridge, G. and Wilson, D. 2015. "School Choice in London and Paris – A
Comparison of Middle-class Strategies." Social Policy & Administration 49(1):24-43.
Berends, M. (2015) “What We Know After Two Decades of Charter Schools.” Annual Review of
Sociology 41:159-180.
Bifulco, R., Ladd, H., and Ross, S. 2009. “Public school choice and integration evidence from
Durham, North Carolina.” Social Science Research 38(1):71-85.
Billingham, C.M. and M.O. Hunt. 2016. “School Racial Composition and Parental Choice: New
Evidence on the Preferences of White Parents in the United States.” Sociology of
Education 89(2):99–117.
Bischoff, K. and Tach, L. 2018. “The Racial Composition of Neighborhoods and Local Schools:
The Role of Diversity, Inequality, and School Choice.” City & Community 17(3):675-
701.
Boterman, W. 2012. "Dealing with Diversity: Middle-class Family Households and the Issue of
‘Black’ and ‘White’ Schools in Amsterdam." Urban Studies p.1-18.
Brown-Saracino, J. 2016. “An Agenda for the Next Decade of Gentrification Scholarship.” City
& Community 15(3):220-225.
Brunner, E., Sung-Woo Cho, J. and Reback, R. 2012. “Mobility, housing markets and
schools: Estimating effects of inter-district choice programs.” Journal of Public
Economics 96(7/8):604-614.
Burdick-Will, J., Keels, M. and Schuble, T. 2013. “Closing and Opening Schools: The
Association between Neighborhood Characteristics and Location of New Educational
Opportunities in a Large Urban District.” Journal of Urban Affairs 35(1):59-80.
53

Candipan, J. 2019. “Choosing Schools in Changing Places: Examining School Enrollment in
Gentrifying Neighborhoods.” Paper presented at the American Educational Research
Association annual conference.
Candipan, J. and Brazil, N. 2018. “Charter School Expansion and Neighborhood Socioeconomic
Ascent.” Paper presented at the Population Association of American annual meeting.
Charles, C. 2003. “The Dynamics of Racial Residential Segregation.” Annual Review of
Sociology 29(1): 167–207.
Denice, P. and Gross, B. 2016. “Choice, Preferences, and Constraints: Evidence from Public
School Applications in Denver.” Sociology of Education 89(4): 300-320.
Ehrenhalt, A. 2012. The Great Inversion and the Future of the American City. NY: Vintage.
Ellen, I.G. 2000. Sharing America’s Neighborhoods : The Prospects for Stable Racial
Integration. Harvard University Press.
Fairlie, R.W. and Resch, A. 2002. “Is There ‘White Flight’ into Private Schools? Evidence from
the National Educational Longitudinal Survey.” Rev of Econ and Stat. 84(1): 21–33.
Goyette, K. Farrie, D. and Freely, J. 2012. “This School's Gone Downhill: Racial Change and
Perceived School Quality Among Whites.” Social Problems 59(2): 155-176.
Grady, S. and Bielick, S. 2010. Trends in the Use of School Choice: 1993 to 2007. Statistical
Analysis Report NCES 2010-004.
Hamnett, C., Butler, T., and Ramsden, M. 2013. "‘I Wanted my Child to go to a More Mixed
School’: Schooling and Ethnic Mix in East London." Environment and Planning A 45:
553–574.
54

Harris, D.R. 1999. “Property Values Drop When Blacks Move In, Because…Racial and
Socioeconomic Determinants of Neighborhood Desirability?” American Sociological
Review 64(3): 461–479.
Hemphill, C. and Mader, N. 2015. “Segregated Schools in Integrated Neighborhoods: The City's
Schools are Even More Divided than Our Housing.” Center for New York City Affairs.
Holme, J.J. 2002. “Buying Homes, Buying Schools: School Choice and the Social Construction
of School Quality.” Harvard Educational Review 72(2): 177–205.
Keels, M., Burdick-Will, J. and Keene, S. 2013. “The Effects of Gentrification on Neighborhood
Public Schools.” City & Community 12(3): 238–59.
Kimelberg, S. M. and Billingham, C.M. 2013. “Attitudes Toward Diversity and the School
Choice Process: Middle-Class Parents in a Segregated Urban Public School District.”
Urban Education 48(2): 198–231.
Krysan, M. 2002. “Whites Who Say They’d Flee: Who are They and Why Would They Leave?”
Demography 39(4): 675–96.
Lareau, A & Goyette, K. eds. 2014. Choosing Schools, Choosing Neighborhoods: Residential
Segregation and the Search for a Good Schools. NY: Russell Sage.
Lavery, L. and Carlson, D. 2015. “Dynamic Participation in Interdistrict Open Enrollment.”
Educational Policy 29(5): 746-779.
Lewis, L. 2014. Inequality in the Promised Land. Stanford, CA: Stanford University Press.
Makris, M.V. 2015. Public Housing and School Choice in a Gentrified City. NY: Palgrave.
Murnane, R. and Reardon, S. 2018. “Long-Term Trends in Private School Enrollments by
Family Income.” AERA Open 4(1): 1-24.
55

National Center for Education Statistics. 2014. “State Support for School Choice and Other
Options, Tables 4.2-4.4.” Retrieved (http://nces.ed.gov/programs/statereform/sss.asp.).
New, W. and Merry, M.S. 2014. “Is Diversity Necessary for Educational Justice?” Educational
Theory 64(4):205-225.
Noreisch, K. 2007. "School Catchment Area Evasion: The Case of Berlin, Germany." Journal of
Education Policy 22:69–90.
Oberti, M. 2007. "Social and school differentiation in urban space: inequalities and local
configurations." Environment and Planning A 39:208-227.
Owens, A. 2012. “Neighborhoods on the Rise: A Typology of Neighborhoods Experiencing
Socioeconomic Ascent.” City & Community 11(4):345-369.
Posey-Maddox, L. 2014. When Middle-Class Parents Choose Urban Schools: Class, Race, and
the Challenge of Equity in Public Education. Chicago, IL: University of Chicago Press.
Pearman, F. and Swain, W. (2017) “School Choice, Gentrification, and the Variable Significance
of Racial Stratification in Urban Neighborhoods.” Sociology of Education p.1-23
Renzulli, L., and Evans, L. 2005. “School Choice, Charter Schools, and White Flight.” Social
Problems 52(3): 398–418.
Roda, A. and Wells, A.S. 2013. “School Choice Policies and Racial Segregation: Where White
Parents’ Good Intentions, Anxiety, and Privilege Collide.” American Journal of Education
119(2): 261–93.
Saporito, S. 2003. “Private choices, public consequences: Magnet school choice and segregation
by race and poverty.” Social Problems 50(2):181–203.
Saporito, S. 2009. “School Choice in Black and White: Private School Enrollment Among Racial
Groups, 1990–2000.” Peabody Journal of Education 84(2): 172–90.
56

Saporito, S. and Sohoni, D. 2006. “Coloring Outside the Lines: Racial Segregation in Public
Schools and Their Attendance Boundaries.” Sociology of Education 79(2): 81–105.
Sikkink, D. and Emerson, M.O. 2008. “School Choice and Racial Segregation in US Schools:
The Role of Parents’ Education.” Ethnic and Racial Studies 31(2): 267–93.
Weinstein, J. 2016. The Impact of School Racial Compositions on Neighborhood Racial
Compositions: Evidence from School Redistricting. Economic Inquiry 54(3): 1356-1382.
Wells, A.S., Lopez, A., Scott, J., Holme, J.J., 1999. “Charter schools as postmodern paradox:
rethinking social stratification in an age of deregulated school choice.” Harvard Educ.
Rev. 69(2): 172–204.
Wilson, W.J. 1987. The Truly Disadvantaged: The Inner City, the Underclass and Public Policy.
Chicago: University of Chicago Press.

57

Appendices

Appendix Table 2.1A. District Means in the Neighborhood-School Gap, 2000 to 2010.
District Name
NS Gap (%White 5-9)
2000 2010 ∆ 2000-2010
Albuquerque -3.74 4.98 8.72
Anne Arundel 2.43 1.78 -0.65
Arlington ISD 0.23 2.93 2.70
Austin ISD 1.47 6.63 5.16
Baltimore City 4.28 5.43 1.16
Baltimore County 2.10 3.02 0.92
Broward 2.08 5.64 3.56
Clark County -3.13 -1.74 1.40
Dade 9.51 6.94 -2.57
Dallas ISD 4.53 6.95 2.42
Denver County 4.67 3.98 -0.69
Detroit 0.21 1.24 1.04
Fairfax County -1.99 0.26 2.25
Fort Bend ISD 0.09 0.18 0.09
Fort Worth ISD 2.28 4.12 1.83
Gwinnett County -3.34 3.64 6.98
Hillsborough 0.00 1.09 1.10
Houston ISD 5.07 5.49 0.41
Indianapolis 1.15 3.33 2.18
Jefferson County (CO) -1.75 -0.58 1.18
Jordan District (UT) -6.40 -4.65 1.74
Long Beach 11.17 5.68 -5.49
Los Angeles 6.56 5.84 -0.72
Milwaukee 11.06 4.72 -6.34
Minneapolis 16.13 4.18 -11.94
New York City 6.08 5.94 -0.14
North East ISD -1.12 -0.18 0.93
Northside ISD -5.13 0.57 5.70
Oakland 7.86 8.09 0.23
Orange (FL) 2.42 3.24 0.82
Palm Beach 4.91 8.34 3.43
Philadelphia 7.12 6.63 -0.49
Phoenix -2.09 2.92 5.01
Pinellas 4.60 -2.38 -6.99
Portland -2.28 5.61 7.89
Prince George's County 1.69 1.93 0.25
Prince William -4.19 0.93 5.12
San Antonio ISD 1.23 2.52 1.29
San Diego 6.62 8.05 1.43
San Juan Unified -2.08 0.19 2.27
Santa Ana Unified 2.21 1.60 -0.62
St. Paul 12.48 11.65 -0.83
Tucson -0.20 2.11 2.32
Virginia Beach -1.86 -2.05 -0.19
Wake County -0.69 4.51 5.20
Ysleta ISD -1.03 0.08 1.11
Notes: Phoenix-area elementary schools districts are comprised of several small independent districts,
which I consolidated into a single district.
58

Appendix Table 2.2A. Mean Neighborhood Racial Composition in School Districts vs.
Metropolitan (MSA) Regions by Year

Notes: Values represent the share (scaled 0-1) of the total population for each racial/ethnic group residing in
school districts (SD) and census-defined metropolitan statistical areas (MSA) (2003 OMB definitions). Dist. ID
is the local education agency identification number (LEAID) for the school district, as provided by the National
Center for Education Statistics (NCES). White, black and Asian shares pertain to non-Hispanic populations
(NH). Shares for other racial/ethnic groups are omitted; thus, row totals for each year do not sum to one.
Phoenix-area elementary schools districts are comprised of several small independent districts, which I
consolidated into a single district.

59

Appendix Figure 2.1A. Overall Change in the NS Gap (% White 5-9 y/0), 2000-2010.

60

Appendix Figure 2.2A. Mean Change in the Neighborhood-School Gap (%White 5-9) by
District

N=46 School Districts

61

Appendix Figure 2.3A. Change in the NS Gap (%White 5-9 y/o) by Neighborhood Type,
2000-2010.

62

Appendix Figure 2.4A. Change in Neighborhood-School Gap from 2000 to 2010 by Charter
Growth and Neighborhood Type

Notes: Neighborhood Types are mutually exclusive and exhaustive. This figure illustrates change in the
neighborhood-school compositional gap (NS gap) from 2000 to 2010 at various levels of charter growth
(i.e. as neighborhoods go from zero charters in 2000 to N charters in 2010). The NS gap grows the most
in ascendant neighborhoods from 2000 to 2010 (nearly 4 points) with compositional mismatches between
neighborhoods and schools increasing at each level of charter growth. While the NS gap widens more
modestly in stable neighborhoods, growth in nearby charter options only minimally contributes to the NS
gap. Conversely, the NS gap narrows from 2000 to 2010 in upper-SES neighborhoods with increasing
charter presence resulting in a greater white student presence in schools than the composition of the
neighborhoods they serve. Although an increase in the number of charter schools from 2000 to 2010
narrows the NS gap in declining neighborhoods, the effect of charter growth on compositional mismatches
is significantly different between declining and ascendant neighborhoods only at the trend level (p<.10),
and not significantly different at conventional levels of statistical significance.

63

Technical Appendix A. Population-Weighted Geographic Reapportionment

Although school attendance boundary information in the U.S. is available publicly for recent
academic years, the sparse availability of these boundary data for earlier years has limited
research on longitudinal changes in racial composition in schools’ neighborhoods. In this study, I
rely on previously collected school attendance boundary data for 47 urban school districts in the
U.S. from Saporito and colleagues for 1999-2000, which I combine with 2009-2010 data from
the School Attendance Boundary Information System (SABINS) project.
8

One challenge is that researchers studying neighborhoods and schools must deal with
geospatial and tabular census data (e.g., blocks, block groups and tracts) that do not share the
same spatial boundaries as school attendance zones. Thus, for this study, I must reapportion
neighborhood data at the census block and tract level to school attendance boundaries. The
geographical boundaries of census blocks and tracts are collected from the National Historical
Geographic Information System (NHGIS).
9
Following Saporito et al. (2007),
10
I use population-
weighted geographic reapportionment in ArcGIS, as this method is shown to produce the most
accurate results, to create neighborhood (SAB)-level measures of racial composition and SES, as
illustrated below. For example, we may want to use population weights to geographically
reapportion a tract-level measure, such as the proportion of residents with a college degree, to the
SAB level. This process of reapportionment is expressed as:

8
The College of William and Mary and the Minnesota Population Center. Version 1.0. Minneapolis, MN: University of
Minnesota 2011. See sabinsdata.org
9
Minnesota Population Center, National Historical Geographic Information System: Version 2.0. Minneapolis: University of
Minnesota, 2011.
10
Saporito, S., Chavers, J., Nixon L., and McQuiddy, M. 2007. “From Here to There: Methods of Allocating Data between
Census Geography and Socially Meaningful Areas.” Social Science Research 36(3): 897–920.
𝐵𝐴
𝑟 =
∑ (𝑃𝑜𝑝 𝑟𝑡
)(𝐵𝐴
𝑟𝑡
)
𝑛 𝑟𝑡 =1
∑ (𝐵𝐴
𝑟𝑡
)
𝑛 𝑟𝑡 =1

64

where BAr is the proportion of residents 25-years and older with a college degree living in
school attendance zone r, Poprt is the total population 25-years and older living within the area of
overlap of school attendance zone r and tract t, and BArt is the proportion of residents with a
college degree 25-years and older within that intersect. This procedure produces a measure for
the school attendance zone that is a population-weighted average of the constituent census tract
measure.
Reapportioning block data, the smallest unit of geography, to both tracts and SABs is fairly
straightforward. All blocks nest fully inside census tracts, and 95 percent of blocks nest fully
inside SABs. For the five percent bisected between two SABs, I assign that block to the SAB in
which the majority of the block lies so that all blocks nest fully within a given SAB, thus
facilitating aggregation of block counts to the school attendance boundary level.
Reapportioning tract-level data to SABs involves additional steps. First, I identify which
blocks nest fully inside of each tract. Then, I identify which tracts fall (fully or partially) inside
each SAB, calculating the population of the tract represented in each SAB (i.e. creating tract-
SAB partitions). Finally, I measure the population share of each tract-SAB partition represented
in each SAB, and these become my population-weights that I use to reapportion tabular census
tract data to the level of school attendance boundaries. See Saporito et al. 2007 for additional
details regarding data reapportionment between different geographic units.

65

Chapter 3
Choosing Schools in Changing Places:
Examining School Enrollment in Gentrifying Neighborhoods

Abstract

School choice expansion in recent decades has weakened the strong link between
neighborhoods and schools created under a strict residence-based school assignment system,
decoupling residential and school enrollment decisions for some families. Recent work suggests
the neighborhood-school link is weakening the most in neighborhoods experiencing
socioeconomic ascent. Using a novel combination of individual, school, and neighborhood data
that links children to both assigned and enrolled schools, this study examines family, school and
neighborhood factors that shape whether parents enroll in the assigned local school. I find that
parents are more likely to opt out of neighborhood schools in gentrifying neighborhoods,
compared to non-gentrifying neighborhoods, when nearby choice options are available. Recent
movers to gentrifying neighborhoods bypass local schools more compared to parents that have
lived in the neighborhood longer. Results suggest the changing racial and SES profile of
gentrifying neighborhoods is not reflected in key neighborhood institutions like schools.

66

Introduction
Historically, nearly all children in the U.S. attended their neighborhood school via a
residence-based school assignment system. Because of this school assignment system, residential
sorting by race and income not only upheld segregation in neighborhoods, but also produced
segregated schools. Since the 1990s, however, the proliferation of school choice options—i.e.
any arrangement allowing children to receive schooling outside of their assigned neighborhood
school (e.g. magnet, charter, and private schools; open enrollment and transfer programs;
homeschooling; etc.)—loosened the tight relationship between residential location and school
assignment. Roughly one quarter of families now enroll their children in a school other than their
assigned neighborhood school (National Center for Education Statistics 2014), and the share of
students opting out for non-neighborhood schools is much greater in urban areas (Grady and
Bielick 2010).
School choice expansion occurred during a period when white and upper-SES households
began migrating back to diverse neighborhoods in urban cores (Wells 2015). Under a residence-
based school assignment system, white and higher-SES households’ mobility into historically
lower-socioeconomic status (SES) and minority urban neighborhoods would contribute to
changes in both neighborhood and school composition. A few recent aggregate-level studies
connecting these trends, however, find that neighborhood and school composition become the
most dissimilar in demographically and socioeconomically changing areas, particularly these
gentrifying neighborhoods (Bischoff and Tach 2018; Candipan 2019; Mader et al. 2018). Such
demographic divergences in gentrifying neighborhoods represent the aggregate of many
household-level decisions about where to live and where to enroll children in school—whether to
activate school choice or enroll in a local neighborhood school. White and higher-SES families

67

are moving into historically lower-SES and minority urban neighborhoods, thus contributing to
demographic and SES change, but little quantitative work has examined whether these changes
extend to important neighborhood institutions, such as traditional public neighborhood schools.
This study takes these insights as starting points to investigate the factors that contribute to
opting out of the neighborhood school, focusing particularly on whether opt-out is more likely in
in gentrifying areas. Observing how parents in gentrifying neighborhoods choose schools may
provide key insights into mechanisms that uphold school segregation even while their
surrounding neighborhoods diversify. Using a novel combination of national-level data from the
Panel Study of Income Dynamics (PSID), School Attendance Boundary Survey (SABS),
Census/American Community Survey (ACS), and National Center for Educational Statistics
(NCES), I identify whether children attend their assigned (neighborhood) school, and examine
the family, school and neighborhood factors that shape school enrollment patterns.
I focus on gentrifying neighborhoods to observe whether commonly understood narratives of
race- and income-based school preferences are upheld even when families choose diverse
neighborhoods. Gentrification complicates the usual narrative regarding residential sorting and
neighborhood segregation in the U.S., instead reflecting a process in which white and higher-
SES households move into lower-SES, minority neighborhoods. Some have proposed integrating
neighborhoods as a means of tackling urban school segregation (Cucchiara and Horvat 2009),
and neighborhood revitalization has at times been presented by some as a way of resolving urban
school segregation (Lipman 2008; Smrekar 2009). However, in logistic regression analyses, I
find that parents in gentrifying neighborhoods are more likely than parents in socioeconomically
stable or declining neighborhoods to opt out of neighborhood schools when there are nearby
school choice options. Recent movers are more likely to opt out than longtime residents in

68

gentrifying neighborhoods, and when they do, they tend to enroll their children in non-
neighborhood schools that serve higher proportions of white students and lower proportions of
black students than their assigned neighborhood school. Therefore, the changing racial and SES
profile of gentrifying neighborhoods is not reflected in key neighborhood institutions like
schools.
The current study explicitly links residential and school processes, which relatively few
studies do (Lareau and Goyette 2014; Owens 2016). By bringing gentrification under its analytic
lens, this study connects processes of neighborhood change with the literature on school and
residential sorting—literatures often examined separately in past work (Lareau and Goyette
2014). Moreover, little past quantitative research investigates micro-level processes underlying
residential and school enrollment outcomes, and those that do are limited to a single-city or
geographic region. The current study broadens the geographic scope using national-level data.
Finally, while previous work has modeled individual household-level school enrollment
outcomes using nationally representative data (Butler et al. 2013), my study is, to the best of my
knowledge, the first national-level quantitative study that considers both actual school enrollment
and assigned school data to model individual schooling enrollment outcomes.
Neighborhood and School Choices in the U.S.
Residence-based school assignment systems, the primary method of assigning students to
schools, created a strong link between where a child lived and attended school. Access to high
quality schools under a residence-based school assignment system is rationed through housing
markets, with white and higher-SES families using neighborhood choice as a means of accessing
the most advantaged and highest performing schools (Goyette 2008; Lavery and Carlson 2014),
thus resulting in stratification across communities and schools by race and socioeconomic status.

69

How individuals select neighborhoods aggregates to shape spatial patterns of inequality, and
since school assignment for decades has been so tightly linked to residence in the U.S., decisions
regarding neighborhood selection also explain why schools have been durably segregated along
racial and economic lines. School choice breaks the link between neighborhoods and schools. As
the number of public choice options like magnet and (especially) charter schools in the U.S.
expanded in recent decades, providing alternatives to attending one’s zoned school, the
historically tight relationship between where a child lived and attended school loosened
(Mickelson et al. 2008). Private schooling, a longstanding form of choice, continued to provide
non-neighborhood options to families with the economic means to attend, though the share of
students served by private schools has declined slightly over the past two decades (Murnane and
Reardon 2018). Moreover, rising participation in “soft” forms of school choice—e.g. inter- and
intra-district transfer programs, ability-based (e.g. gifted & talented) and school voucher
programs, and homeschooling, etc.— also contributed to this decoupling between housing and
educational markets (Brunner et al. 2012; Loeb et al. 2011).
In theory, the expansion of non-neighborhood school choice options has the potential to
liberate students from segregated schools by severing the neighborhood-school link under a strict
residence-based assignment system (Davis and Bauman 2013). Whether parents opt-out of
neighborhood schools, however, depends on preferences and constraints (often mutually
reinforced), both of which vary by race and SES, and middle-class families are often in better
positions to act on their school and neighborhood preferences (Bifulco, Ladd and Ross 2009;
Goyette 2008; Kimelberg and Billingham 2013; Makris 2018). In particular, white middle-class
parents’ schooling decisions tend to be influenced by the presence of nonwhites, with white
parents more likely to flee or avoid schools with higher shares of nonwhite students for magnet,

70

charter, and private schools (Billingham and Hunt 2016; Fairlie and Resch 2002; Hamnett et al.
2013; Oberti 2007; Renzulli and Evans 2005; Saporito 2009; Saporito and Sohoni 2006).
Whether due to racial prejudice or other factors for which race proxies (Krysan 2002), past work
finds that middle-class white parents’ perceptions of school quality drops as the minority
(particularly black) racial composition increases (Goyette et al. 2012; Holme 2002). Moreover,
their perception of school quality is also shaped by word-of-mouth input from parental networks
that are also segregated by race and SES, resulting in socially constructed understandings of
“good schools” that are highly correlated with race and SES above and beyond test scores or
other evaluative metrics (Holme 2002; Roda and Wells 2013).
The ability to opt-out also presupposes existing choice options. How choice is geographically
distributed varies widely by district, and whether a family opts out of their neighborhood school
depends on how they perceive their options (Blagg and Chingos 2017; Denice and Gross 2016).
For example, low-income and predominantly nonwhite neighborhoods typically have lower-
resourced schools, so some parents will seek public choice options to find better schooling
options. Public school choice options serve a larger share of low-SES and minority students in
part because they locate more often in low-income and minority neighborhoods (Lubienski and
Gulosino 2007; Burdick-Will et al. 2013). That share, however, could potentially be even higher
if low-income parents residing in these neighborhoods had fewer constraints, relative to their
higher-earning peers, that shape their preferences and limit their perceptions of plausible non-
neighborhood choice options, such as insufficient transportation, complex enrollment
procedures, a lack of information from less connected social networks, among other factors
(Blagg and Chingos 2017; Rhodes and DeLuca 2014; Neild 2005; Schneider et al. 1997). While
private schools are distributed throughout most urban areas—the vast majority of students in

71

urban areas live near at least one private option (70 percent within two miles; 92 percent within
five miles) (Blagg and Chingos 2017)—the ability to enroll in private school is economically
prohibitive for many low-income families.
Examining Dynamics in Gentrifying Neighborhoods
Families, especially white middle-class families, often take schools into account when
choosing where to live (Lareau 2014). Past research shows that white families prefer to live in
white neighborhoods, in part because they perceive nonwhite neighborhoods to have lower-
quality institutions like schools (Harris 1999). Gentrification, however, represents a different
residential sorting process. The original definition describes gentrification as the phenomenon of
large numbers of higher-SES residents move into low-SES neighborhoods (Glass 1964), but a
large body of work has documented the racialized structure of neighborhood change (Brown-
Saracino 2017), with racial transition often occurring alongside class transitions during
gentrification in the U.S. (Owens and Candipan 2018).
Despite a large body of gentrification scholarship, almost no quantitative work has examined
how schools, a key neighborhood institution, change alongside gentrifying neighborhoods.
Qualitative studies examining the social mix of demographically and socioeconomically
transitioning neighborhoods identify differences between higher-SES newcomers and longtime
residents in terms of neighborhood norms, use of public space, and connections to longstanding
churches, businesses, and local organizations, all of which lead to segregated social spaces
within mixed-income neighborhoods (Freeman 2006; Hyra 2015; Tach 2009; Tissot 2011). As
gentrifying neighborhoods change along racial and economic lines, will in-moving newcomer
parents participate in or avoid their neighborhood schools? Given both whites’ preferences for
white schools and low-SES families’ constraints, we might expect non-neighborhood choice

72

enrollment patterns to differ both between gentrifying neighborhoods and other types of
neighborhoods and between newcomers and existing residents within gentrifying neighborhoods,
since incoming families are typically higher-SES and white (Owens and Candipan 2018).
Examining how children in gentrifying neighborhoods sort into schools may provide insight into
mechanisms that maintain school segregation even while the neighborhoods they serve diversify.
Gentrification thus provides the conditions to observe whether the influx of white and higher-
SES residents that produces neighborhood diversity (at least temporarily) translates into a
corresponding increase of white students in schools.
Families that move into gentrifying neighborhoods may be served by schools that are still in
the process of changing and thus composed of a relatively greater presence of poor or nonwhite
students or that have been historically disadvantaged in some way. Some parents may feel more
comfortable residing in gentrifying neighborhoods precisely because there are non-neighborhood
school options available. Indeed, one recent study even finds that gentrification is more likely to
occur in minority neighborhoods with non-neighborhood school options (Pearman and Swain
2018). Public choice options like magnet schools, originally created to attract higher-SES
families, offer special programs and curriculums (e.g. language immersion and arts programs)
that may appeal to middle-class parents seeking non-neighborhood alternatives (Jordan and
Gallagher 2015). Charter schools, with curricular and programmatic flexibility that may appeal
particularly to middle-class parents, have expanded substantially in the past decade, more than
doubling as a share of all public schools (from 3.1 in 2004 to 6.6 in 2014 (NCES 2016)). They
now feature in public discourse as one of the most prominent public choice alternatives to non-
neighborhood schools (Berends 2015). Recent ethnographic work finds that middle-class
families view enrollment in charter schools, especially prestige charters (Brown and Makris

73

2018), as attractive options to bypass undesirable neighborhood schools. This is particularly true
in neighborhoods experiencing gentrification or neighborhood socioeconomic ascent (Keels et al.
2013; Kimelberg and Billingham 2012; Makris 2015) where, since the early 2000s, an increasing
share of market-oriented charter schools have opened (Burdick-Will et al. 2013; Davis and
Oakley 2013; Lubienski and Gulosino 2007; Brown and Makris 2018). One study using survey
data finds that white and middle-class families perceived charter schools as higher quality
schools even when they were lower performing relative to neighborhood schools (Weiher and
Teiden 2002). Public school choice may therefore grant gentrifier parents who crave a certain
threshold of diversity in their neighborhoods the opportunity to remain in diverse neighborhoods
without choosing diverse neighborhood schools. Middle-class and white parents’ broad
perceptions of local school quality, coupled with race- and income-based school preferences,
may translate into their selective school sorting out of assigned neighborhood schools,
particularly when those schools are located in gentrifying low-income or minority neighborhoods
(Mader et al. 2018).
On the other hand, gentrifier families, particularly early-wave gentrifers (Brown-Saracino
2009), may be drawn to cities because they value diversity and other urban qualities in
neighborhoods (Billingham and Kimelberg 2013; Kimelberg 2014), suggesting that parents may
choose diverse local neighborhood schools that are racially and socioeconomically diverse. A
handful of qualitative studies suggests that while gentrifier parents may be willing to choose
diverse urban neighborhood schools, their schooling decisions are largely influenced by the
enrollment choices of other gentrifier parents (Billingham and Kimelberg 2013; Stillman 2012).
When they do choose diverse urban schools, they frequently conceive contingency plans that
allow them to opt-out if they observe undesirable qualities in their assigned school (Stillman

74

2012). Qualitative scholarship has found that even when neighborhood schools in gentrifying
neighborhoods become racially and socioeconomically diverse due to incoming white and
higher-SES students, there are problems of displacement and disempowerment of existing
families or frictions that cause gentrifier parents to eventually exit the local school (Cucchiara
2013; Stillman 2012). Together, these qualitative studies speak to underlying tensions between
gentrifier parents’ desire to expose their children to diversity and their actual behaviors when
confronting integrated social settings at school (Boterman 2013; Butler 2003; Posey-Maddox et
al. 2014; Roda and Wells 2013; Stillman 2012; Wells et al. 1999).
The Present Study
Collectively, these insights raise questions about how, in an era of expanding school choice,
parents make enrollment decisions in neighborhoods experiencing demographic and
socioeconomic change. This study examines whether parents opt out of the neighborhood school
by addressing the following empirical questions:
RQ1. Does a neighborhood’s school choice context predict whether parents opt out of
neighborhood schools?
RQ2. Are parents in gentrifying neighborhoods more likely to opt out of neighborhood
schools than parents in non-gentrifying neighborhoods?
RQ3. Are recent movers more likely to opt out of neighborhood schools, particularly in
gentrifying neighborhoods where newcomers and existing residents may differ both
demographically and in terms of their knowledge of local schools?
RQ4. Do parents who opt out enroll their children in schools with different racial/ethnic
composition than the neighborhood school?
This study aims to make several substantive and empirical contributions, focusing on the
family, school, and neighborhood factors that shape school enrollment decisions. Together, the
four questions explore whether parents in gentrifying neighborhoods embrace or avoid
neighborhood schools that may still serve a relatively larger share of minority and low-income

75

students. Importantly, this study combines multiple data sources in novel ways to address
previously unanswered questions with quantitative data. I link students to their neighborhood
and school contexts, assigned and enrolled schools, and residential histories to gain traction on
unexplored aspects of neighborhood and school processes. Few past quantitative research
investigates micro-level processes underlying residential and school enrollment outcomes, and
those that do are limited to a single-city or geographic region. The current study broadens the
geographic scope using national-level data.
Data and Measures
My analyses require information on where children live and when they moved to that
neighborhood, as well as where they attend school and whether that school is their local assigned
neighborhood school. Restricted-use data from the PSID main interview and 2014 PSID-Child
Development Supplement (CDS-II) include identifiers of children’s residential location and
enrolled school, making these data uniquely suited to answer questions that connect parents’
decisions about where to live and send their children to school. The PSID is a nationally
representative longitudinal survey of households that began in 1968 with annual follow-up
interviews until 1996, and biennial interviews starting in 1997. The CDS-II was administered to
sample children under age 18. PSID includes census tract identifiers and NCES school
identifiers, which I use to identify characteristics of children’s neighborhoods from the
Decennial Census and the ACS and characteristics of students’ schools from NCES. I take
advantage of PSID’s longitudinal design to track when and where families move to construct
measures of residential tenure, which allows me to examine differences in school enrollment
patterns between newcomers and long-time residents in gentrifying neighborhoods.

76

Identifying Children’s Neighborhoods and Schools
My analyses require that I know the school to which children are assigned based on their
residential location and the school that each child actually attends in 2014. PSID restricted-use
Geospatial Match (Geomatch) files provide census block identifiers for the home residence of
each PSID household in each survey wave. To determine a child’s assigned school, I created a
census block-school crosswalk for nearly all blocks in the U.S by spatially joining georeferenced
block-level data from the 2010 Census and attendance zone shapefiles from the 2014 SABS.
11

SABS level-specific shapefiles identify the school assigned to each block. Essentially all blocks
nest fully inside school attendance boundaries.
12
I merge the crosswalk with the Geomatch file to
identify the school assigned to each child’s block. Next, to determine a child’s enrolled school, I
rely on the CDS-II restricted-use file which includes school identifiers for each school-age
child’s enrolled school. My dependent variable is a measure of whether or not a student attends
her assigned school in 2014—whether the assigned and enrolled schools match.
I also use the Geomatch files to identify children’s neighborhoods at each survey wave.
Following a large body of neighborhood-focused quantitative research, I use census tract as my
proxy for neighborhood. Since each census block nests fully and uniquely within one census
tract, I am able to assign families to their respective census tracts based on their PSID-provided
block identifiers.

11
The 2013-14 SABS provides shapefiles designating primary and secondary attendance zones along with NCES
school identifiers. I link school IDs from these shapefiles to CCD tabular data to determine grades served by each
school.
12
Just over 95 percent of blocks in the U.S. nest fully inside school attendance boundaries. For the five percent that
do not fall entirely within a school’s catchment area in my crosswalk, I assign blocks to the SAB in which the
majority of its population resides, following Saporito et al. (2007). In my PSID sample, there were no blocks that
were split by multiple school attendance boundaries, but there were a small number of cases in which a child’s
residence was served by multiple SABs (~ five percent). In these cases, I considered attendance at either of these
schools as enrolling in a child’s assigned neighborhood school.

77

The resulting dataset is restricted to elementary- and secondary-level children
13
residing in a
census-defined metropolitan statistical area (MSA, using 2003 OMB definitions), and that have
complete data on enrolled school (via PSID) and assigned school (via SABS) (n=1094).
14
I limit
to children living in metropolitan areas because this is where gentrification occurs and where
school choice has expanded the most; 186 of the 380 MSAs in the U.S. (excluding Puerto Rico)
are represented in my sample (roughly 49 percent). After matching children to both schools and
neighborhoods, I am then able to link to neighborhood and school characteristics that may shape
children’s enrollment outcomes. Collectively, these data will allow me to explore factors that
predict whether parents enroll their children in neighborhood schools or bypass their residentially
assigned school for non-neighborhood options.
Dependent Variable: Opt-Out from Neighborhood School
My dependent variable is a dichotomous measure indicating enrollment in a non-assigned
neighborhood school option. In addition to well-known choice options such as magnet, charter and
private schools, my measure of opt-out also captures parents that bypass local schools in favor of
homeschooling, participate in district programs such as intra-district transfer or Gifted & Talented
programs, or any other non-neighborhood school arrangement (including parents that move to a

13
NCES does not have a standard definition of elementary levels with regard to grade span. While most define
elementary levels as KG-5, districts vary in how they categorize elementary grades (with some considering 6
th
-8
th

grades as elementary). Secondary levels are generally defined as grades 9-12. SABS includes grade span
information for each catchment zone, so I am able to confirm whether a child in my sample is served by elementary
or secondary schools as defined by their district.
14
The limiting factor for my sample is whether I could identify children’s assigned school(s) with the 2013-14
SABS. The response rate for all traditional public schools in SABS (including schools located in non-metro areas
and open enrollment areas) was 85.6 percent (Phan 2015). In my sample, roughly 80 percent of children matched to
an assigned school. One reason the match rate is lower than the SABS response rate could be related to how I treat
open enrollment districts. SABS includes shapefiles for complete open enrollment zones that span entire districts,
which I remove from my sample since my analyses are concerned with understanding whether residents participate
in assigned local schools. This means complete open enrollment districts (i.e. districts without any assigned schools)
are not included in my sample. Open enrollment, nonetheless, represents an additional school choice option that
could shape a family’s decision about where to live that I do not observe directly in this study.

78

new school attendance boundary but continue to send their children to the school served by their
former neighborhood). All of these represent options that decouple gentrifying neighborhoods
from neighborhood schools.
Key Covariates
School Choice Context
For parents to opt out of the assigned neighborhood school, there first must be non-
neighborhood school options available. My key schooling-related predictor is a proximity measure
of school choice, which captures the number of nearby charter, magnet and private schools in
2014.
15
To construct this measure, I rely on georeferenced data on school location and sector from
the Common Core of Data (CCD) (for charter and magnet schools) and the Private Universe of
School Survey (PSS) (for private schools) and use spatial techniques to calculate the number of
each type of school that falls inside a two-mile radius of the centroid of each child’s census tract.
I use a two-mile radius following past work, as this represents a reasonable distance for parents
when considering potential school options (Denice and Gross 2016; Glazerman and Dotter 2017).
16

The resulting school proximity measures are the number of grade-specific nearby non-
neighborhood public school choice (i.e. magnets and charters) and private options (private
schools), as well as the total number of choice options, to each family’s residence.
17
These

15
Some schools did not report magnet or charter status in the CCD. For these cases, I used a word scraping
algorithm to identify whether these were charter and magnet schools. I also constructed alternative measures of
choice proximity, including (1) top coding the number of choice options and (2) combining public and private
choice into one measure. The substantive findings of my analyses do not change.
16
Substantive patterns hold in sensitivity analyses testing alternative radii of one and three miles.
17
Many districts (including some of the largest districts) have programs, such as intra-district transfer programs,
which allow parents to choose non-neighborhood schools (Grady and Bielick 2010). Children that bypass local
schools via transfer programs still appear in my sample as opt-out students. Comprehensive, detailed national data
on these programs are not available and thus not included in the construction of my school choice measure, though
having this information would improve my measure of a neighborhood’s school choice context by capturing a richer
set of non-local options that parents may perceive as available.

79

measures capture the choice context of a given neighborhood. It should be noted, however, that
parents’ perceptions of choice options are often disconnected from the reality of being able to
activate those options (Makris 2018).
Residential Tenure
Next, I focus on how long families have lived in the neighborhood. We might expect to
observe different behaviors with respect to schooling decisions between those that have recently
moved into a neighborhood compared to families that have resided in the neighborhood for
longer, particularly in gentrifying neighborhoods where there is greater divergence in terms of
the SES profiles of newcomers and longtime residents. It could also be the case that existing
residents are more acclimated to the neighborhood school and less influenced by reputation
(Wells et al. 2018). For my analyses, I construct a binary measure of residential tenure,
newcomer, which designates whether a family moved to its current school attendance boundary
within the last two years.
18
I create this measure using the block identifiers included in the
Geomatch files which allow me to construct residential histories for families and observe
whether a household makes a residential move into a new school attendance boundary in each
study wave.
Neighborhood SES Trajectory
My key neighborhood-related predictor is a typology of neighborhood SES trajectories: 1)
gentrifying; 2) stable low/middle socioeconomic status; 3) declining; and 4) upper-SES.

18
Results hold in sensitivity analyses that define residential newcomers as those that moved within the last four
years. In additional analyses, I explored a 3-category definition of residential tenure to test whether school
enrollment outcomes vary between first- and second-wave gentrifiers. In these models, I do find some differences
between parents who moved to their new school attendance boundary 2-6 years before to those who moved within
the past 2 years; results align with the qualitative work on first- and second-wave gentrifiers (results available upon
request). I retain my binary measure of newcomer in part because the timing of gentrification is something that is
difficult to capture using census data, and further disaggregating the newcomer measure into first- and second-wave
in-movers may not capture whether these two groups are truly different.

80

To define neighborhood SES trajectories, I draw on tract-level data from the long-form 1990
Census and 2008-12 five-year aggregate ACS on five indicators of neighborhood SES that
capture housing and socioeconomic characteristics: median home value, median rent, median
household income, percent of residents 25 years and older with at least a college degree, and
percent of residents 16 years and older in a managerial, professional or technical occupation
(Owens 2012; Owens and Candipan 2018). Because census tract geographies change over time,
I normalize tracts to 2010 boundaries, using the Longitudinal Tract Database (LTDB) (Logan,
Xu, and Stults 2017). I then use factor analysis, a method used to reduce several correlated
variables into a set of linearly uncorrelated underlying dimensions, to construct neighborhood
SES factor scores. I calculate SES factor scores using the entire universe of tracts for each MSA
represented in my sample. I measure SES over a two-decade period, starting in 1990, in order to
capture a neighborhood’s SES trajectory prior to the year of the dependent variable (2014). Each
tract is assigned a standardized SES factor score with a mean of 0 and standard deviation of 1 in
1990 and 2008-2012 (hereafter, 2010), and I use these scores to assign an SES percentile rank for
each tract relative to others within the same MSA. Relative SES rank is scaled from 0 to 100,
with higher ranks representing tracts with the highest SES.
After constructing relative SES ranks in both years, I categorize tracts as gentrifying, stable
low/mid-SES, declining, or upper-SES based on changes in their relative SES rank from 1990 to
2010. Following prior work using ten-percentage point cutoffs to define neighborhood change
(Ellen and O’Regan 2011; McKinnish et al. 2011; Owens and Candipan 2018), I categorize tracts
that begin in the bottom four SES rank quintiles in 1990 and increase in relative SES rank by at
least ten percentage points as gentrifying (n=224 students living in these neighborhoods).
Importantly for my analyses, gentrifying tracts are also defined as having above-average growth

81

rates in white population from 1990-2010 relative to their MSA.
19
Declining tracts are those that
decline by ten percentage points or more (n=247). All remaining tracts that begin in the bottom
four SES quintiles in 1990 are considered stable low/mid-SES (n=450). Finally, upper SES tracts
are those that begin in the upper quintile in terms of SES rank (n=173), and end in the top two
quintiles. Neighborhood types are mutually exclusive and exhaustive.
20
The regional distribution
is mostly proportional across neighborhood types, though there are fewer neighborhoods from
the East in my sample (9.9 percent) and more from the South (45.0 percent). A higher proportion
of all low/mid-SES neighborhoods is located in the South region (58.3 percent)
Control variables
Motivated by past research, my analyses control for a battery of family characteristics,
individual (child) attributes, and school factors that may shape whether a child attends his/her
assigned neighborhood school. For family characteristics, number of children is a three-category
variable capturing whether a family has one, two, or three or more children under the age of 18
during the 2013 survey wave. Models also include a continuous measure capturing the age of
youngest child in a family.
21
Models also control for family income, a continuous measure
constructed by averaging family income (divided by 10,000) in the 2013 and 2015 survey
waves.
22
Marital status is a binary measure indicating whether a child’s primary caregiver is

19
Eighty-two percent of all tracts that ascend by at least ten points in relative SES rank also experienced above-
average growth in non-Hispanic white population. Models excluding the growth rate criterion from definitions of
gentrification or excluding ascendant cases (n=41) with below-median white growth rate both result in substantively
similar results.
20
In sensitivity analyses, patterns hold using alternative specifications of gentrification and neighborhood type. This
includes neighborhood type definitions that rely on 9- and 11-point thresholds for categorizing SES change. I also
tested models using non-exhaustive definitions of neighborhood SES type which restricted stable Low/Mid-SES
neighborhoods to those that changed five percentage points or fewer, and patterns hold across all models.
21
Some researchers use oldest school-age child to measure how child age shapes parents’ residential and schooling
decisions. In sensitivity analyses, models using oldest child produce substantial similar results.
22
The PSID-provided family income measures are comparable across survey waves and reflect family income for
the year prior to the main interview (i.e. family income in 2012 for interview year 2013).

82

married (in 2013), and homeownership indicates whether a household owns its home. All models
also account for child-level factors. Grade is a 2-category variable capturing whether a child is
in: 1) elementary (i.e. grades K-8); or 2) high school (grades 9-12) during the 2013-14 academic
school year.
23
Aside from operational differences and varying choice admissions processes,
elementary levels tend to have smaller catchment areas than secondary levels, though secondary
levels typically have more choice options. A few qualitative studies suggest that gentrifier
parents of elementary-level children may choose public schools, but reevaluate this decision as
they progress through higher grades as these decisions are often contingent on access to selective
enrollment schools (Billingham and Kimelberg 2013; Kimelberg and Billingham 2013).
Motivated by existing work on school preferences that finds white parents particularly sensitive
to school racial composition, I control for child race with a binary indicator denoting whether or
not a child is non-Hispanic white. Informed by past work showing that parents might be
particularly sensitive to minority (particularly black) school racial composition (Billingham and
Hunt 2016), I measure school racial/ethnic school composition as percent black, derived from
the 2012-2013 Common Core of Data (CCD). I lag this measure by one year to capture the
racial/ethnic context of a child’s assigned school prior to making school enrollment decisions.
Finally, I control for metro population density to ensure that my fixed radius approach to
measuring a neighborhood’s choice context is not driven by dense neighborhoods with
substantially more non-neighborhood schooling options.

23
I use “primary” and “secondary” level shapefiles from the School Attendance Boundary Survey to match children
to their assigned neighborhood school. I then confirm that the assigned school matching a child’s home residence
serves the grade of the child.

83

Analysis Plan
I perform logistic regression with robust standard errors (clustered by household) to predict
the odds that a child opts out of the assigned neighborhood school based on household,
neighborhood, and school characteristics. I begin with models examining how the availability of
school choice shapes non-neighborhood school enrollment patterns. I estimate the equation:

logit (OptOut) = α + β
𝑐 𝑃𝑢𝑏𝐶 ℎ𝑜𝑖𝑐𝑒 + β
𝑝 𝑃𝑟𝑖𝑣𝑎𝑡𝑒 + β
𝑘 𝑋 𝑘 + ε (1)

where my dependent variable, OptOut, is a dichotomous indicator identifying whether a child
enrolls in any type of school other than her neighborhood school (1=opt out; 0=attends
neighborhood school). My key covariates of interest, PubChoice and Private, capture the number
of nearby magnet, charter, and private schools to a child’s home residence. Here, the coefficients
indicate whether greater availability of nearby choice options is associated with the odds of
opting out of the neighborhood school. All models include X, a vector of family, individual and
(assigned) school covariates (described above).
Next, I investigate how residential tenure shapes school enrollment outcomes, specifically
whether recent in-movers to neighborhoods are more likely to opt out. This is expressed as:
Logit (OptOut) = α + β
𝑐 𝑃𝑢𝑏𝐶 ℎ𝑜𝑖𝑐𝑒 + β
𝑝 𝑃𝑟𝑖𝑣𝑎𝑡𝑒 + β
𝑛 𝑁𝑒𝑤𝑐𝑜𝑚𝑒𝑟

+ β
𝑘 𝑋 𝑘 + ε (2)
Here, my second key predictor, Newcomer, is a binary measure designating a recent move (in
last two years) to the neighborhood. The coefficient for this term reveals whether recent in-
movers to all neighborhoods are more likely than non-newcomers to bypass the assigned
neighborhood school.

84

In the next set of models, I consider how school enrollment outcomes may differ across
neighborhoods with varying neighborhood SES trajectories. I estimate the full model:
logit (OptOut) = α + β
𝑐 𝑃𝑢𝑏𝐶 ℎ𝑜𝑖𝑐𝑒 + β
𝑡 𝑇𝑦𝑝𝑒 + β
𝑟 𝑁𝑒𝑤𝑐𝑜𝑚𝑒𝑟 + β
𝑟𝑡
𝑁 𝑒𝑤𝑐𝑜𝑚𝑒𝑟 ∗ 𝑇𝑦𝑝𝑒
+ β
𝑐𝑡
𝑃𝑢𝑏𝐶 ℎ𝑜𝑖𝑐𝑒 ∗ 𝑇𝑦𝑝𝑒
+ β
𝑝 𝑃𝑟𝑖𝑣𝑎𝑡𝑒 + β
𝑘 𝑋 𝑘 + ε (3)
These analyses are particularly interested in observing how gentrification plays into the
decision to send a child to the neighborhood school, and I do this in two ways. First, I include
Type, which categorizes neighborhoods as gentrifying (reference group), stable low/mid-SES,
declining, or upper-SES. The coefficient for neighborhood type indicates whether school
enrollment patterns differ across neighborhoods with varying SES trajectories, controlling for the
same individual and family background and school characteristics as in earlier models. Note that
the coefficient for 𝑇𝑦𝑝𝑒 indicates the odds that a family opts out of the assigned neighborhood
school conditional on residing in a particular neighborhood SES type. They do not predict
whether a family moves into gentrifying or other types of neighborhoods. As such, the aim of
these models is to document where choice is more likely to be activated. To take the findings of
the previous analyses into account, I include an interaction term between public choice
availability and neighborhood type, 𝑃𝑢𝑏𝑙𝑖𝑐𝐶 ℎ𝑜𝑖𝑐𝑒 ∗ 𝑇𝑦𝑝𝑒 ,

which indicates whether parents are
more likely to opt out in gentrifying neighborhoods when there is more public choice. (I control
for private school proximity but focus on the role of public choice, which expanded significantly
during this period. I discuss this further below.) Second, I interact my measure for newcomer
status with neighborhood type, Newcomer*Type, to test whether newcomers are more likely to
opt-out of their neighborhood school compared to non-newcomers, and how these differential
odds of opt-out by residential tenure differs between gentrifying and stable low/mid-SES,
declining and upper-SES neighborhoods.

85

Finally, building from this set of regression analyses, and continuing the focus on
newcomers, I close with descriptive analyses observing the racial composition of assigned and
enrolled schools for newcomers that opt out of the neighborhood school. After establishing
where newcomers are more likely to opt out in regression models, these descriptive analyses then
aim to document particular features of schools (i.e. school racial composition (particularly white
and black student composition)) may potentially factor into in-movers’ decisions to bypass
neighborhood schools.
Findings
Table 3.1 displays descriptive statistics for my dependent measure, key independent
variables, and controls. The overall opt-out rate (i.e. non-neighborhood school enrollment) in my
analytic sample is just over 44 percent, which is slightly higher than the national average for
urban districts. Gentrifying neighborhoods in my sample have the lowest opt-out rate (40.6
percent), but gentrifying neighborhoods also have the fewest number of nearby choice options
(1.3), on average. Thirty-eight percent of gentrifying neighborhoods have any nearby choice
options compared to 2/3 of stable low/mid-SES, declining and upper-SES neighborhoods. Stable
low/mid-SES neighborhoods have the greatest average number of both nearby public and private
choice options (1.4 schools of either type). Among only neighborhoods that do have choice
options, however, both stable low/mid-SES and gentrifying neighborhoods have the most public
and private choice options, on average (public and combined public/private choice means are 3.3
and 4.4 for low/mid-SES; 3.2 and 3.5 for gentrifying). Gentrifying and upper-SES
neighborhoods also have the greatest proportion of white residents (71 and 76 percent in 2010)
while declining neighborhoods have the greatest proportion of nonwhite residents (61 percent in
2010). While the proportion of white residents declined among all neighborhood types from

86

1990 to 2010 (reflected national demographic changes), it declined the least in gentrifying
neighborhoods (where, by my definition, neighborhood white population growth exceeded the
median growth rate of their respective MSAs from 1990 to 2010).
In my sample, PSID children residing in gentrifying and upper-SES neighborhoods also tend
to be white (72.8 and 67.6 percent), while children in declining neighborhoods are mostly
nonwhite (64 percent). Just over two-fifths of children are recent in-movers (“newcomers”) to
their current neighborhood. Notably, among newcomers to gentrifying neighborhoods, about 64
percent are white, the largest share of all neighborhood SES types. On the other hand, only 32
percent of newcomers to declining neighborhoods are white, representing the lowest share
among neighborhood types.
24

Predicting Neighborhood School Opt-Out
I begin with regression analyses examining family, school, and neighborhood factors that
shape school enrollment outcomes for families with school-aged children. Table 3.2 displays
results from logistic regression models predicting the odds of enrollment in a non-neighborhood
school. Model 1 presents the baseline model predicting the odds of opting out of the
neighborhood school, given family and child factors and school racial composition. A greater
number of kids (p<.05) is associated with lower odds of enrollment in non-neighborhood
schools—families with three or more children are less likely to bypass their assigned
neighborhood school. Family income has a positive effect (p<.05) on the odds of opting out

24
Note that families in the CDS 2014 rarely reside in the same neighborhoods. The non-nested structure means that
“newcomers” in this analysis are not examined in direct comparison to non-newcomers residing in the same
neighborhoods. Additionally, newcomers and longer-tenured residents may reside in different types of ascendant
neighborhoods (i.e. newcomers in my sample tend to move into ascendant neighborhoods with lower SES than the
SES of ascendant neighborhoods in which non-newcomers reside). These particularities of the PSID sample should
not substantively alter results from my analyses since I am interested more generally in the enrollment outcomes of
newcomers.

87

indicating that class plays a role in decoupling neighborhoods and local schools—higher income
families are more likely to enroll in non-local schools. For each additional ten thousand dollars in
family income, the odds of enrolling in a non-neighborhood school increase by just over three
percent.
Model 2 adds measures of school choice proximity, my main school-level predictors of non-
neighborhood school enrollment, which capture the public and private choice context of a
neighborhood. We would expect an increase in the number of nearby available school choice
options to increase the odds of opting out of one’s assigned neighborhood school. Consistent
with expectation, greater availability of nearby private options increases the odds that parents
will bypass the local neighborhood school. Each additional private school increases the odds of
opting out by a factor of 1.13 (p<.001). Interestingly, the coefficient for public choice (i.e.
magnet and charter) is not statistically significant. It could be the case that parents residing in
different types of neighborhood vary in whether they use magnet and charter options, which I
explore in later models.
Next, I examine how the odds of non-neighborhood school enrollment differ between recent
movers and existing residents (Model 3). Relative to longer-tenured residents, recent movers to a
school attendance boundary are more likely to enroll in a school other than their assigned
neighborhood school, conditioning on public and private school choice availability and model
controls (OR=1.7; p<.01). Newcomers may avoid neighborhood schools because they are less
familiar with them than longer-tenured residents, or it may be the case that newcomers only
move to certain neighborhoods if they can opt out of the neighborhood school. Some of these

88

newcomers may be parents that prioritize school stability by moving to a new neighborhood but
continuing to send their children to their former neighborhood’s school.
25

Do Parents Bypass Neighborhood Schools More Often in Gentrifying Neighborhoods?
The next set of models examines whether parents are more likely to opt out in gentrifying
neighborhoods. Table 3.3 displays odds ratios for models examining whether school choice
context and residential tenure shape school enrollment outcomes differently in gentrifying
neighborhoods compared to other non-gentrifying neighborhood types. To be clear, these
analyses do not support causal claims about parents opting out because of gentrification, but
rather present a descriptive portrait of how patterns of non-neighborhood school enrollment play
out across neighborhoods experiencing various SES trajectories.
Table 3.3, Model 1 presents the odds of opting out conditioning on school choice availability
and neighborhood type. Here, the main effect for public choice is positive, but remains non-
significant. Public choice options have proliferated since the mid-1990s, roughly the same period
during which I observe neighborhood SES change. Qualitative work has found that public
options, such as charter and magnet schools, are viewed as attractive options for gentrifier
families. Therefore, I add interaction terms between nearby public choice and neighborhood type
in Model 2 and, after doing so, the null main effect for public choice in Model 1 gains statistical
significance. In this model, the main effect for nearby public choice, which represents the odds
for parents in gentrifying neighborhoods (since gentrification is my reference neighborhood
category), is statistically significant. The odds of opt-out in gentrifying neighborhoods increases
with each additional public choice option that is located nearby (OR=1.87; p<.01). Interestingly,

25
DeLuca and Rosenblatt (2010) suggest school stability as a reason why some movers in their study of Moving to
Opportunity families in Baltimore did not choose their new local school. About three percent of children in my
sample (n=33) enrolled in the assigned school of their former neighborhood.

89

when combining the main and interaction effects for public choice and neighborhood type, we
see that the association between nearby public choice options and opt-out is weaker in low/mid-
SES, upper-SES, and declining neighborhoods—i.e. the odds of opting out are lower in non-
gentrifying neighborhoods relative to gentrifying neighborhoods when there are public choice
options. Greater availability of nearby public choice options increases the odds of parents opting
out of the neighborhood school, but it increases the odds the most in gentrifying
neighborhoods—parents are most likely to bypass their assigned neighborhood school in
gentrifying neighborhoods relative to low/mid-SES, upper-SES, and declining neighborhoods
when there are nearby choice options. Taken together, results here suggest that opt-out increases
the most the in gentrifying neighborhoods as more public options are available.
Recall that fewer gentrifying neighborhoods in my sample have nearby choice options
compared to all non-gentrifying neighborhood types, so the null main effect for public choice in
Model 1 was concealing key differences in terms of where the public choice context of a
neighborhood was more likely to break the link between home residence and neighborhood
school enrollment (i.e. gentrifying neighborhoods). Although fewer gentrifying neighborhoods
had public choice options, parents residing in those neighborhoods were far more likely than
parents in non-gentrifying neighborhoods to opt out where those options were available. I also
tested whether the odds of enrolling in private school, a longstanding option that has not
expanded in recent decades, varied between different types of neighborhoods, but did not
observe significant differences between gentrifying and non-gentrifying neighborhoods (results
not shown). While there will always be some subset of families that choose private schools, thus
breaking the neighborhood-school link in all neighborhoods with nearby options, the private

90

school context of a neighborhood does not seem to be the main feature driving neighborhood
choices differently between families in gentrifying and non-gentrifying neighborhoods.
To better interpret the interaction terms from Model 2 (Table 3), Figure 1 presents predictive
probabilities of opting-out at various levels of public school choice availability in different
neighborhood types, drawing on those results. On average, parents residing in gentrifying
neighborhoods (square symbol) with at least some availability of public choice options are more
likely to bypass neighborhood schools compared to parents in non-gentrifying neighborhoods,
holding covariates at their respective means. While gentrifying neighborhoods have the greatest
expected probability of opting out with just one public school choice option, the probabilities are
similar across all neighborhood types.
26
Significant differences, however, emerge as the number
of public options increases—the upwards curve for gentrifying neighborhoods clearly illustrates
that the probability of opting out of neighborhood schools in gentrifying neighborhoods is higher
relative to declining, upper-SES, and low/mid-SES neighborhoods. Note that the average number
of public choice options for gentrifying neighborhoods that do have nearby choice is about 3.2
schools, so the higher probabilities for gentrifying neighborhoods relative to other neighborhood
types represent feasible outcomes. Moreover, the positive effect of public choice is greater in
gentrifying neighborhoods compared to non-gentrifying neighborhoods. For example, the
probability of opt-out is about 15 percentage points higher for parents residing in gentrifying
neighborhoods with one public school option (50) compared to gentrifying neighborhoods
without any public choice (35), on average. In contrast, comparing a neighborhood without any
public choice options to a neighborhood with one public option increases the probability of opt-

26
Considering the combined number of all nearby public and private choice options, parents in gentrifying
neighborhoods are most likely to opt out when there are at least two nearby magnet, charter or private schools
(Appendix Figure 3.1A).

91

out in low/mid-SES neighborhoods (circle symbol) by only about two percentage points (44 to
46). In stable upper-SES neighborhoods, it could be the case that residents find their local school
acceptable, so the availability of public choice does not decouple neighborhoods and schools in
the same way as gentrifying neighborhoods. Instead, when parents in upper-SES neighborhoods
do activate choice options, they may invest in private schooling as a way to accrue additional
advantages despite being zoned for high quality neighborhood schools (Loeb et al. 2011;
Mickelson et al. 2008).
Do Newcomers to Gentrifying Neighborhoods Bypass Local Schools?
Results thus far indicate that children residing in gentrifying neighborhoods are most likely to
opt out of their assigned neighborhood schools as the number of nearby school choice options
increases. In the previous set of analyses, I also showed that newcomers were more likely to opt
out of the neighborhood school. This may be particularly true in gentrifying neighborhoods
because newcomers, who in my sample are mostly white and higher-income, may differ most
from longer-tenured residents in terms of both their ability to access choice and their interest in
doing so. For example, higher-SES newcomers to gentrifying neighborhoods may view charter
options more favorably than neighborhood schools, which may still lag behind neighborhood
changes along socioeconomic and demographic lines.
27
On the other hand, newcomers to
neighborhoods that are more socioeconomically (and demographically) stable (i.e. stable
low/mid-SES and upper-SES neighborhoods), may not differ substantially from existing
residents in either their school preferences or economic conditions.

27
Charter schools vary widely in aims and prestige, and some qualitative scholarship finds that gentrifier parents’
preference for charters depends on the model (Brown and Makris 2018). One quantitative study finds that white and
middle-class families tend to perceive charter schools as higher quality than neighborhood schools even when
ranked lower on academic performance (Weiher and Teiden 2002).

92

In the next model, I explore heterogeneity in the odds of opt-out between newcomers and
long-time residents across neighborhood type. Are the odds of opt-out for newcomer parents in
gentrifying neighborhoods higher relative to parents that have resided in the neighborhood
longer? Table 3.3 Model 3 investigates whether the newcomer effect varies in different types of
neighborhoods.
28
In this interacted model, the main effect for newcomer represents the odds
ratio of gentrifying newcomers to existing residents, since gentrification is my reference
neighborhood category. Relative to existing residents, the odds of opting out of neighborhood
schools for recent in-movers to gentrifying neighborhoods are 160 percent greater, all factors
considered (OR 2.6; p<.05).
29
Additionally, results here indicate that residential tenure also
significantly shapes non-neighborhood schooling outcomes in declining neighborhoods. After
calculating odds ratios between newcomers and existing residents in declining neighborhoods
and performing post-estimation tests of significance, I find that newcomers in declining
neighborhoods are also more likely to enroll in non-neighborhood schools (OR 2.3; p<.05).
30

Figure 3.2 illustrates the results of Model 3 in a different way, comparing the average
predictive probabilities of opting out between newcomers (squares) and longer-tenured (circles)
residents at varying levels of nearby public choice options in gentrifying, low/mid-SES, upper-
SES, and declining neighborhoods. The trends illustrated in Figure 2 align with the hypothesis
that newcomers to gentrifying neighborhoods are more likely to opt out of local schools

28
I performed a series of robustness checks, including: models restricted to a single child per household (i.e.
youngest-child only; oldest-child only); models excluding middle grades (i.e. 7
th
and 8
th
); random effects models.
Results do not change, and the substantive interpretations hold across all models.
29
In additional analyses, I construct a binary measure that captures whether family income is above the census tract
median. Then, rather than using newcomer, I perform models interacting this binary measure of high-income with
neighborhood type to observe whether higher income families are more likely to opt out. Results are consistent with
the newcomer models in gentrifying neighborhoods.
30
I used post-estimation tests to identify whether coefficients for newcomers and non-newcomers were significantly
different in each neighborhood type. Odds ratios for newcomers versus non-newcomers were significant in both
gentrifying and declining neighborhoods (both p<.05). Full results available upon request.

93

compared to longer-tenured parents. Here, the expected probabilities for newcomers are most
dissimilar from existing residents in gentrifying and declining neighborhoods— the higher trend
lines for newcomers suggest that they opt out more often than existing residents when nearby
public choice options are available. It could be the case that newcomers to both types of
neighborhoods rely on much different networks than existing residents for information about
local schools. On the other hand, the adjusted probabilities for newcomers and existing residents
map onto each other almost identically in low/mid-SES and upper-SES neighborhoods—two
stable neighborhood types. That newcomers are most dissimilar from existing residents in
socioeconomically and demographically changing neighborhoods (i.e. gentrifying and declining
neighborhoods) suggests that while recent in-movers may be contributing to changes in the
demographic profile of their neighborhoods, they may not be contributing similarly to their
neighborhood schools. Results here suggest that the main effect for newcomer from earlier
models may be driven largely by parents moving in to socioeconomically and demographically
changing (i.e. gentrifying and declining) neighborhoods.

Descriptive Analyses: Are Newcomers in Gentrifying Neighborhoods Opting Out for Whiter
Schools?
Findings thus far indicate that in places with at least one available nearby public school
choice option, families residing in gentrifying neighborhoods (relative to non-gentrifying types)
are more likely to bypass the local school, with newcomers to gentrifying neighborhoods most
likely to opt out. Gentrifying neighborhoods tend to have greater in-flows of upper-SES and
white residents, which contribute to the changing demographic and socioeconomic profiles of
those neighborhoods (Owens and Candipan 2018). Most prior work on school preferences
indicates that whites are particularly sensitive to school racial composition, preferring schools

94

with substantial representation of white students and seeking alternate schooling options when
there is a perceptible share of nonwhite (particularly black) students. Building off regression
analyses of newcomers, who are most likely to opt out, the following analyses further examine
school features that may be associated with non-neighborhood enrollment among newcomers.
Specifically, I interrogate how race may factor into enrollment decisions of newcomers by
comparing the school racial composition of schools to which students are assigned to those that
students actually attend. When newcomers to gentrifying neighborhoods opt out, are they
enrolling in schools with a greater share of white students than their assigned neighborhood
school?
Table 3.4 compares the racial composition of children’s assigned neighborhood schools to
the schools in which they actually enroll.
31
The first two rows indicate that when parents send
their children to non-neighborhood schools, the average white racial composition of the enrolled
school is higher than the assigned school while the opposite is true for black racial
composition—the share of black students is lower in enrolled schools relative to the assigned
schools. Newcomers in all types of neighborhoods opt out for schools with greater shares of
white students and fewer black students, on average (Rows 3 and 4). Among newcomers that opt
out, the difference in racial composition in assigned schools compared to enrolled schools is
most evident in gentrifying and low/mid-SES neighborhoods. The contrast in racial composition
of assigned and enrolled schools is greatest among white newcomers. While white newcomers to
stable low/mid-SES and declining neighborhoods also opt out for schools with more white
students than their assigned neighborhood school, white newcomers to gentrifying

31
Although homeschoolers are represented as opt-out children in my sample, they are excluded from this analysis
since the racial composition of the enrolled school is not relevant (5.5 percent of opt-outs). I also exclude three
children without racial composition data for the enrolled school (< 1 percent of opt-outs).

95

neighborhoods opt out for schools with a greater share of white students (64.7 to 78.2 percent),
as well as a smaller proportion of black students (10.6 vs 4.6 percent). This suggests that white
newcomers to gentrifying neighborhoods may be more sensitive to racial composition of
neighborhood schools than the racial composition of their neighborhood. This aligns with a
recent single-year report showing that white kindergarten students from gentrifying
neighborhoods in New York City enrolled in schools with substantially greater proportions of
whites relative to the student body of their assigned neighborhood schools (Mader et al. 2018).
These unadjusted analyses, though descriptive (and limited by a small sample size), reveal
patterns suggestive of meaningful associations by race and neighborhood type that are consistent
with past work on school preferences. The implication here is that the segregating choices of
white and upper-SES newcomers may produce segregated social networks and separate
educational experiences even among students of the same neighborhood.
32

Interestingly, opt-out white newcomers to socioeconomically declining neighborhood enroll
in schools with both a higher percentage of white and black students relative than their assigned
schools. Although the proportion of white opt-out newcomers to declining neighborhoods is
relatively small (~33 percent), results nonetheless suggest nuance around typical “avoidance”
explanations in the neighborhood and school choice literature. Future work should further
explore how contextual features of schools vary between neighborhood types, and how school

32
In supplementary analyses, I place a similar question into a multivariable framework, performing logistic
regression to predict whether opt-out parents enroll their children in schools with a higher share of white students
than their assigned school. I find that the odds of enrolling in schools with a greater proportion of white students
than the assigned school are higher for white parents and increase as percent school black of the assigned school
rises. Relative to longer-tenured residents, opt-out newcomers overall have substantially greater odds of enrolling in
schools with a higher percentage of white students than the local school, with the largest disparity occurring in
gentrifying neighborhoods (Appendix Table 3.1A). Though suggestive of patterns showing an association between
school race and opt-out, these analyses are limited by sample size, and future work should continue to pursue these
questions.

96

characteristics shape enrollment decisions differently between parents in gentrifying and
declining neighborhoods.

Discussion
Since neighborhoods have historically been segregated along economic and racial/ethnic
lines, schools have also endured legacies of segregation. When neighborhoods change, do
schools also follow suit? Using a novel combination of individual, school, and neighborhood
data that links children to both assigned and enrolled schools in the U.S., this study set out to
understand whether the availability of school choice shapes school enrollment outcomes in
gentrifying neighborhoods— where the inflow of white and higher-SES parents to lower-SES
and minority urban communities presents an opportunity to observe whether commonly
understood narratives of race- and income-based school preferences are upheld even when
families choose more diverse neighborhoods. I find that when there are nearby choice options to
opt out of the neighborhood school, parents in gentrifying neighborhoods are more likely than
parents in socioeconomically stable and declining neighborhoods to activate non-local options,
bypassing neighborhood schools in the process. Since gentrifying neighborhoods are those
experiencing change, this decoupling between neighborhoods and schools implies that whatever
racial/ethnic and socioeconomic change occurring at the neighborhood level may not result in
corresponding changes in schools. Findings are consistent with recent aggregate-level research
that describes growing dissimilarity between neighborhood and school composition in areas
experiencing socioeconomic ascent (Bischoff and Tach 2018; Candipan 2019; Mader et al.
2018).

97

As past qualitative work has shown, processes of residential and school choice in gentrifying
neighborhoods are complex and dynamic. Whether gentrifier parents opt out is thus highly
contingent on neighborhood and school contexts, the geographic distribution of educational
opportunities, and the ability of parents to actualize those opportunities. On the one hand,
particular features of assigned local schools could draw gentrifier parents into gentrifying
neighborhoods. On the other hand, gentrifiers opt out the most when there are non-local options
suggesting that the choice context of a neighborhood may also draw families into neighborhoods
where they may hold reservations about the local school.
Residential newcomers opt out of neighborhood schools more than parents that have lived in
the neighborhood longer, particularly in gentrifying neighborhoods. When newcomers,
particularly white newcomers, opt out of neighborhood schools in gentrifying neighborhoods,
they tend to seek non-neighborhood schools that serve higher proportions of white students
relative to their assigned neighborhood schools. These results imply that neighborhood
integration does not guarantee school integration—the diversity produced via gentrification may
also uphold school segregation as incoming white and higher-SES families choose more
frequently to bypass neighborhood schools when alternative school options are nearby. This has
important implications for thinking about how urban and school policy is linked, and for
considering how efforts to address school segregation via neighborhood integration may not
always produce intended results.
Findings suggest that neighborhood schools in gentrifying areas may grow increasingly
dissimilar from the neighborhoods they serve, particularly as school choice options proliferate.
One implication of parents opting out in gentrifying neighborhoods is that while the

98

neighborhoods may change, their key institutions may not. This results in non-intersecting social
networks—even while neighborhood diversity increases, the segregated social networks within
neighborhoods lead to separate educational (and everyday) experiences (Burdick-Will 2018).
Another concern is that school choice may divert both money and political power away from
traditional public schools in gentrifying neighborhoods (as Ladd and Singleton (2018) find
regarding charter schools and local school funding in North Carolina), or that upper-SES and
white parents may divert their social and economic resources to non-neighborhood schools. That
said, while school integration is often a stated goal of districts, integrating schools along
demographic lines alone does not guarantee meaningful inclusion, as other scholars have noted
(Lewis and Diamond 2015; Lewis-McCoy 2014). While my data cannot speak directly to these
nuanced dynamics, future work should consider examining within-school processes, such as
whether specialized programming and curricula that may draw gentrifier parents into
demographically diverse urban schools ultimately sort students into segregated tracks within
schools.
Partly due to data limitations, few quantitative studies have analyzed connections between
residential and school choice nationally and from a micro-level perspective. In this study, I
examined individual school enrollment outcomes among children residing in different types of
neighborhoods to understand where, and for whom, school choice occurs. I focused on schooling
decisions in gentrifying neighborhoods to understand whether parents that move into integrating
neighborhoods embrace or avoid schools that may still be relatively diverse and less advantaged
in some way. Findings suggest that while some families may welcome diversity in
neighborhoods undergoing demographic change, that same embrace does not always extend to
schools in changing neighborhoods.

99

This study presents a first step towards understanding how individual decisions about where
to live and attend school aggregate to shape broader patterns of inequality across neighborhood
and schools, two key contexts for children’s well-being. The school choice “liberation” model
posits that severing the neighborhood-school link is a way of expanding educational
opportunities for lower-SES families anchored to lower performing schools under a strict
residence-based assignment system. Results from my study, however, find that the expansion of
school choice breaks the neighborhood-school link more frequently in gentrifying
neighborhoods, allowing white and higher-SES in-movers the opportunity to make residential
decisions independently of school considerations. Future work should further explore whether
some families consider a wider array of residential options in certain neighborhoods despite
concerns about the zoned school, since school choice effectively decouples neighborhoods from
the traditional public schools that serve them. Do parents make residential decisions increasingly
on the basis of neighborhood characteristics without regard to neighborhood schools?
This study has limitations that future research should address. First, although I gather
information on families’ neighborhoods and residential moves over time, this study can only
examine school enrollment outcomes in a single year. Although school enrollment data is
available in earlier CDS surveys, the lack of comprehensive SABS data for the 1997-2007
periods prohibits my ability to match kids precisely to their assigned schools in earlier CDS
years. As collection of school attendance boundary data improves, future work should
incorporate subsequent waves of CDS-II to address some of these limitations. Second, I lack
comprehensive district school assignment policy data that would allow for more extensive
examinations of how non-neighborhood options shape enrollment outcomes through softer forms
of school choice. Collecting comprehensive, detailed data on district programs and other forms

100

of soft choice over time would allow scholars to fully consider all types of choice options when
examining links between housing and education markets. Third, because PSID sample families
rarely live in the same neighborhood, I was unable to directly compare newcomers and existing
residents from the same neighborhood. Moreover, the data I used to construct residential
histories was reported once every two years, so my analyses may be missing nuances in the
timing of moves. Additionally, the origin and design of the PSID has resulted in a largely Black-
White sample over time, limiting extensive examinations on how multiethnic diversity shapes
neighborhood and school processes. Fourth, because census data are limited in their ability to
capture the exact timing of neighborhood change, I lose some of the specific timing in terms of
how school choice affects gentrification (and vice versa). Finally, this study was limited
somewhat by sample size, including relatively smaller group sizes for gentrifying and
(especially) upper-SES neighborhoods. A larger sample would allow for additional analyses (e.g.
models examining newcomers with children at similar grade levels; etc.).
Despite these limitations, this study makes several substantive contributions, combining
multiple data sources in new ways to address previously unanswered questions with national-
level quantitative data. While prior work demonstrates that choice happens for certain types of
people, this study finds that choice also happens in certain types of places. On a theoretical level,
this work engages with and extends theories of residential and school sorting and speaks more
broadly to how individual-level processes, such as decisions about where to live and attend
school, aggregate to create wider patterns of spatial inequality. Findings here thus raise new
questions about how race- and class-based sorting upholds existing racial and class hierarchies
by way of neighborhood and school stratification. Although focused largely on gentrifying
neighborhoods, this study sheds light more broadly on the nuanced and complex process of

101

neighborhood change. Results bear on housing and school policies, including affordable
housing, neighborhood revitalization, school choice, neighborhood and school integration,
economic development, student assignment policies and many other urban and education
policies.

102

Tables and Figures

Table 3.1. Summary Statistics of Key Variables in Analysis Sample

Notes: Analysis sample restricted to school-age children residing in census-defined MSAs, and that have with complete school
assignment and neighborhood SES data. Newcomers refers to households that moved school attendance boundaries in the past
two years. Public choice refers to magnet and charter schools. All choice refers to magnet, charter and private schools. The
number of public choice options (among neighborhoods with options) reports the average count of nearby public options for
neighborhoods with any public choice.

103

Table 3.2. Effects of School Choice Proximity and Residential Tenure on Non-Neighborhood
School Enrollment
Model 1 Model 2 Model 3
Baseline Controls School Choice Newcomer
Family Background

Family Income (in ten thousands) 1.033* .015 1.034* .017 1.035* .017

Age of Youngest Child .952+ .028 .956 .028 .967 .028

College Degree .967 .170 .979 .187 .995 .194

Married 1.019 .199 1.078 .210 1.127 .221

Owns Home 1.114 .216 1.106 .215 1.368 .284

Number of Kids (Under 18)

1 ref

ref

ref

2 .815 .153 .793 .150 .790 .150

3+ .553** .120 .545** .119 .551** .121

Child Factors

Non-Hisp. White (Child) .983 .200 1.049 .216 1.066 .221

Grade Level

Grades KG-8 ref

ref

ref

Grades 9-12 1.189 .236 1.250 .246 1.218 .243

School Factors

Lagged Pct School Black 1.691+ .486 1.500 .434 1.606 .472

MSA Controls

Population Density 1.096 .098 .970 .090 .969 .089

School Choice

Num. Public Choice Schools <2m

1.089 .060 1.088 .061

Num. Private Schools <2m

1.129* .054 1.134** .054

Residential Tenure

Newcomer

1.704** .300

_cons .891 .281 .718 .232 .438* .161

n(clusters) 784

784

784

N(observations) 1094 1094 1094
Notes: Logistic regressions results reported as odds ratios. Robust standard errors (clustered by household) displayed below
odds ratios in italics. All models include family, child and school covariates. Newcomer refers to households that moved
school attendance boundaries within the past two years. School percent black is lagged by one year.
+ p<0.10, * p<0.05, ** p<0.01, *** p<0.001 (two-tailed).
104

Table 3.3. Effects of School Choice Proximity and Residential Tenure on Non-Neighborhood
School Enrollment by Neighborhood SES Type
Model 1 Model 2 Model 3
Neighborhood Type Public School Choice Newcomer
Family Background

Family Income (in ten thousands) 1.036* .018 1.038* .018 1.039* .018

Age of Youngest Child 0.957 .027 .949+ .027 0.958 .028

College Degree .971 .187 .935 .182 .962 .192

Married 1.075 .211 1.103 .218 1.129 .223

Owns Home 1.100 .215 1.164 .229 1.478+ .312

Number of Kids (Under 18)

1 ref

ref

ref

2 .798 .152 .784 .151 .778 .151

3+ .550** .121 .532** .118 .527** .118

Child Factors

Non-Hisp. White (Child) 1.047 .217 1.014 .215 1.020 .218

Grade Level

Grades KG-8 ref

ref

ref

Grades 9-12 1.236 .243 1.285 .256 1.237 .250

MSA Controls

Population Density 0.967 .092 0.965 .094 0.957 .092

Neighborhood Factors

Neighborhood Type

Gentrifying ref

ref

ref

Low/Mid-SES 1.219 .280 1.531+ .383 2.062* .676

Declining 1.082 .278 1.498 .436 1.589 .604

Upper SES 1.243 .338 1.807* .530 2.683** 1.003

School Choice

Num. Public Choice Schools <2m 1.086 .060 1.869** .414 1.755** .349

Num. Private Schools<2m 1.126* .054 1.129* .054 1.133* .053

NeighType x Pub. Choice

Gentrifying*Pub. Choice

ref

ref

Low/Mid-SES*Pub. Choice

.584* .133 .622* .128

Declining*Pub. Choice

.521** .127 .549** .123

Upper SES*Pub. Choice

.428** .111 .455** .110

Residential Tenure

Newcomer

2.640* 1.071

NeighType x Newcomer

Gentrifying*Newcomer

ref

Low/Mid-SES*Newcomer

.565 .262

Declining*Newcomer

.886 .446

Upper SES*Newcomer

.382+ .209

School Factors

Lagged Pct School Black 1.487 .441 1.350 .407 1.463 .452

_cons .627 .230 .529+ .200 .267** .124

n(clusters) 784

784

784

N(observations) 1094 1094 1094
Notes: Logistic regression results reported as odds ratios. Robust standard errors (clustered by household) displayed to
the right of odds ratios in italics. All models include family, child and school covariates. Reference neighborhood
category is gentrifying neighborhoods. Newcomer refers to households that moved school attendance boundaries within
the past two years. School percent black is lagged by one year.
+ p<0.10, * p<0.05, ** p<0.01, *** p<0.001 (two-tailed).

105

Table 3.4. Mean School Racial Composition of Assigned and Enrolled Schools for Opt-Out
Newcomers

Low/Mid-SES
(n=450)
Gentrifying
(n=224)
Declining
(n=247)
Upper-SES
(n=173)

Assigned
School
Enrolled
School
Assigned
School
Enrolled
School
Assigned
School
Enrolled
School
Assigned
School
Enrolled
School
All Opt-Out Students % White 28.1% 33.9% 56.4% 59.7% 28.3% 36.5% 54.0% 56.9%

% Black 40.6% 36.5% 18.8% 17.4% 38.7% 36.0% 19.9% 17.7%
Opt-Out Newcomers % White 21.5% 28.9% 49.2% 56.0% 30.4% 33.1% 52.3% 51.9%

% Black 46.0% 42.1% 25.6% 22.0% 36.5% 39.9% 24.9% 22.2%
Opt-Out White
Newcomers % White 41.9% 50.2% 64.7% 78.2% 44.5% 54.6% 66.2% 64.9%

% Black 8.8% 7.9% 10.6% 4.6% 8.4% 11.8% 13.1% 9.8%
Note: Sample restricted to opt-out students only. Homeschooled students (about 5 percent of opt-out students) and students
with missing racial composition data for the enrolled school (n=3) are also excluded from these analyses. Assigned school
refers to the neighborhood school to which each student is assigned based on grade and residence. Enrolled school refers the
school a student actually attends. Data on school racial composition calculated from 2013-14 Common Core of Data (CCD) and
2013-14 Private School Universe Survey (PSS). Newcomer refers to households that moved school attendance boundaries in
the past two years.

106

Figure 3.1. Predicted Probabilities of Non-Neighborhood School Enrollment in Gentrifying and
Non-Gentrifying Neighborhoods Given Public Choice Options

Notes: Sample is KG-12
th
Grades. Expected probabilities correspond to results in Table 3, Model 2.
Public choice refers to the number of grade-specific magnet and charter schools located within two miles
of a child’s residence. All covariates held at their respective means.

107

Figure 3.2. Predicted Probabilities of Opting Out Between Residential Newcomers and Long-
time Residents (by Neighborhood Type)

Notes: Sample is KG-12
th
Grades. Probabilities presented for Table 3, Model 3. Public choice refers to the
number of grade-specific magnet and charter schools located within two miles of a child’s residence. All
covariates held at their observed values. Newcomer refers to households that moved school attendance
boundaries in the past two years.

108

References
Berends, Mark. 2015. “What We Know After Two Decades of Charter Schools.” Annual Review
of Sociology 41:159-180.
Bifulco, Robert and Helen F. Ladd. 2007. “School Choice, Racial Segregation, and Test-Score
Gaps: Evidence from North Carolina’s Charter School Program.” Journal of Policy
Analysis and Management 26(1):31-56.
Bifulco, Robert, Helen F. Ladd, and Stephen Ross. 2009. “Public School Choice and Integration
Evidence from Durham, North Carolina.” Social Science Research 38(1):71-85.
Billingham, Chase M. and Shelley M. Kimelberg. 2013. “Middle-Class Parents, Urban
Schooling, and the Shift from Consumption to Production of Urban Space.” Sociological
Forum 28(1):85–108.
Billingham, Chase M. and Matthew O. Hunt. 2016. “School Racial Composition and Parental
Choice: New Evidence on the Preferences of White Parents in the United States.”
Sociology of Education 89(2):99–117.
Bischoff, Kendra, and Laura Tach. 2018. “The Racial Composition of Neighborhoods and Local
Schools: The Role of Diversity, Inequality, and School Choice.” City & Community
17(3):675-701.
Blagg, Kristin and Matthew Chingos. 2017. Who could benefit from school choice? Mapping
access to public and private schools. Washington, DC: Urban Institute.
Boterman, Willem. 2013. "Dealing with Diversity: Middle-class Family Households and the
Issue of ‘Black’ and ‘White’ Schools in Amsterdam." Urban Studies 50(6): 1130-1147.

109

Brown, Elizabeth and Molly V. Makris. 2018. “A Different Type of Charter School: In Prestige
Charters, a Rise in Cachet Equals a Decline in Access.” Journal of Educ. Policy 33: 85–
117.
Brown-Saracino, Japonica. 2017. “Explicating Divided Approaches to Gentrification and
Growing Income Inequality.” Annual Review of Sociology 43:515-539
Brown-Saracino, Japonica. 2009. A Neighborhood That Never Changes: Gentrification, Social
Preservation and the Search for Authenticity. Chicago, IL: University of Chicago Press.
Brunner, Eric, Sung-Woo Cho, J. and Randall Reback. 2012. “Mobility, housing markets and
schools: Estimating Effects of Inter-district Choice Programs.” Journal of Public
Economics 96(7/8):604-614.
Burdick-Will, Julia, Micere Keels, and Todd Schuble. 2013. “Closing and Opening Schools: The
Association between Neighborhood Characteristics and Location of New Educational
Opportunities in a Large Urban District.” Journal of Urban Affairs 35(1): 59-80.
Butler, J.S., Douglas A. Carr, Eugenia F. Toma, and Ron Zimmer. 2013. “Choice in a World of
New School Types.” Journal of Policy and Management 32(4): 785-806.
Butler, Tim. 2003. “Living in the bubble: Gentrification and its “others” in North London.”
Urban Studies 40(12): 2469–2486.
Candipan, Jennifer. 2019. “Neighbourhood change and the neighbourhood-school gap.” Urban
Studies, 0042098018819075.
Charles, Camille. 2003 “The Dynamics of Racial Residential Segregation.” Annual Review of
Sociology 29(1):167–207.
Cucchiara, Maia B. 2013. Marketing Schools, Marketing Cities: Who Wins and Who Loses When
Schools Become Urban Amenities? Chicago: University of Chicago Press.

110

Cucchiara, M. B., & Horvat, E. M. 2009. Perils and promises: Middle-class parental involvement
in urban schools. American Educational Research Journal 46(4): 974–1004.
DeLuca, Stefanie and Peter Rosenblatt. 2010. “Does Moving To Better Neighborhoods Lead to
Better Schooling Opportunities? Parental School Choice in an Experimental Housing
Voucher Program.” Teachers College Record 112(5): 1443-1491.
Denice, Patrick and Bethany Gross. 2016. “Choice, Preferences, and Constraints: Evidence from
Public School Applications in Denver.” Sociology of Education p.1-21.
Ellen, Ingrid G. 2000. Sharing America’s Neighborhoods: The Prospects for Stable Racial
Integration. Harvard University Press.
Ellen, Ingrid Gould, and Katherine M. O’Regan. 2011. “How Low-Income Neighborhoods
Change: Entry, Exit, and Enhancement.” Regional Science & Urban Economics 41(2):
89–97.
Fairlie, Robert W. and Alexandra Resch. 2002. “Is There ‘White Flight’ into Private Schools?
Evidence from the National Educational Longitudinal Survey.” Rev of Econ and Stat.
84(1):21–33.
Frankenberg, Erica, Genevieve Siegel-Hawley and Gary Orfield. 2008. The Forgotten Choice:
Rethinking Magnet Schools in a Changing Landscape. Los Angeles, CA: The Civil
Rights Project.
Glass, Ruth. 1964. Introduction: Aspects of Change. In London: Aspects of Change, ed. Centre
for Urban Studies, London: MacKibbon and Kee, xiii–xlii.
Glazerman, Steven and Dallas Dotter. 2017. “Market Signals: Evidence on the Determinants and
Consequences of School Choice from a Citywide Lottery.” Educational Evaluation and
Policy Analysis 39(4):593-619.

111

Goyette, KA. 2008. Race, Social Background, and School Choice Options: Equity & Excellence
in Education, 41:1, 114-129.
Goyette, Kimberly, Danielle Farrie, and Joshua Freely. 2012. “This School's Gone Downhill:
Racial Change and Perceived School Quality Among Whites.” Social Problems
59(2):155-176.
Grady, Sarah and Stacey Bielick. 2010. Trends in the Use of School Choice: 1993 to 2007.
Statistical Analysis Report NCES 2010-004.
Harris, David R. 1999. “Property Values Drop When Blacks Move In, Because…Racial and
Socioeconomic Determinants of Neighborhood Desirability?” American Sociological
Review 64(3):461–479.
Holme, Jennifer J. 2002. “Buying Homes, Buying Schools: School Choice and the Social
Construction of School Quality.” Harvard Educational Review 72(2):177–205.
Hyra, Derek. 2015. “The Back-to-the-City Movement: Neighbourhood Redevelopment and
Processes of Political and Cultural Displacement.” Urban Studies 52(10): 1753-1773.
Jordan, Reed and Megan Gallagher. 2015. “Does School Choice Affect Gentrification?”
Washington, DC: Urban Institute.
Keels, Micere, Julia Burdick-Will and Sara Keene. 2013. “The Effects of Gentrification on
Neighborhood Public Schools.” City & Community 12(3):238–59.
Kimelberg, Shelley. M. 2014. “Beyond Test Scores: Middle-Class Mothers, Cultural Capital, and
the Evaluation of Urban Public Schools.” Sociological Perspectives 57(2):208–28.
Kimelberg, Shelley M. and Chase M. Billingham. 2013. “Attitudes Toward Diversity and the
School Choice Process: Middle-Class Parents in a Segregated Urban Public School
District.” Urban Education 48(2):198–231.

112

Krysan, Maria. 2002. “Whites Who Say They’d Flee: Who are They and Why Would They
Leave?” Demography 39(4):675–96.
Ladd, Helen F. and John D. Singleton. 2018. “The Fiscal Externalities of Charter Schools:
Evidence from North Carolina.” Economic Research Initiatives at Duke (ERID) Working
Paper No. 261. Available at SSRN: https://ssrn.com/abstract=3082968 or
http://dx.doi.org/10.2139/ssrn.3082968
Lareau, Annette. 2014. “Schools, Housing, and the Reproduction of Inequality.” Pps. 169-206 in
A. Lareau and K. Goyette (eds.) Choosing Homes, Choosing Schools. NY: Russell Sage.
Lareau, Annette & Kimberly Goyette (eds). 2014. Choosing Schools, Choosing Neighborhoods:
Residential Segregation and the Search for a Good Schools. NY: Russell Sage.
Lavery, Lesley and Deven Carlson. 2015. “Dynamic Participation in Interdistrict Open
Enrollment.” Educational Policy 29(5):746-779.
Lewis, Amanda and John Diamond. 2015. Despite the Best Intentions: How Racial Inequality
Thrives in Good Schools. New York, NY: Oxford University Press.
Lewis-McCoy, L’Heureux. 2014. Inequality in the Promised Land. Stanford, CA: Stanford
University Press.
Loeb, Susanna, Jon Valant, and Matt Kasman. 2011. “Increasing Choice in the Market for
Schools: Recent Reforms and their Effects on Student Achievement.” National Tax
Journal. 64(4):141-164.
Logan, John, Zengwang Xu, and Brian Stults. 2017. “A Longitudinal Tract Database.” Database,
Brown University.
https://s4.ad.brown.edu/Projects/Diversity/researcher/ltbddload/datalist.aspx.

113

Lubienski, Christopher. 2003. “Innovation in Education Markets: Theory and Evidence on the
Impact of Competition and Choice in Charter Schools.” American Educational Research
Journal 40(2): 395–443.
Mader, Nicole, Clara Hemphill and Qasim Abbas. 2018. "The Paradox of Choice: How School
Choice Divides New York City Elementary Schools." Center for NYC Affairs. Policy
Report.
Makris, Molly V. 2018. “The Chimera of Choice: Gentrification, School Choice, and
Community” Peabody Journal of Education 93(4): 411-429.
Makris, Molly V. 2015. Public Housing and School Choice in a Gentrified City. New York:
Palgrave Macmillan.
McKinnish, Terra, Randall Walsh, and T. Kirk White. 2010. “Who Gentrifies Low-Income
Neighborhoods?” Journal of Urban Economics 67:180–93.
Mickelson, Roslyn A., Martha Bottia, and Stephanie Southworth. 2008. “School Choice and
Segregation by Race, Class and Achievement.” Policy brief, Education Policy Research
Unit, Arizona State University, Tempe, and Education and the Public Interest Center,
University of Colorado, Boulder, http://nepc.colorado.edu/publication/school-choice-and-
segregation-race-class-and-achievement
Murnane, Richard. and Sean Reardon. 2018. “Long Term Trends in Private School Enrollments
by Family Income.” AERA Open 4(1):1-24.
National Center for Education Statistics. 2014. “State Support for School Choice and Other
Options, Tables 4.2-4.4.” Retrieved (http://nces.ed.gov/programs/statereform/sss.asp.).
National Center for Education Statistics. 2016. “The Condition of Education: Public Charter
School Enrollment.” NCES Publication No. 2016144.

114

Neild, Ruth C. 2005. “Parent Management of School Choice in a Large Urban District.” Urban
Education 40(3): 270-297.
Oberti, Marco. 2007. "Social and school differentiation in urban space: inequalities and local
configurations." Environment and Planning A 39:208-227.
Owens, Ann. 2012. “Neighborhoods on the Rise: A Typology of Neighborhoods Experiencing
Socioeconomic Ascent.” City & Community 11(4):345-369.
Owens, Ann. 2016. “Inequality in Children’s Contexts: Income Segregation of Households With
and Without Children”. American Sociological Review 81(3):549-574.
Owens, Ann and Jennifer Candipan. 2018. “Racial/Ethnic Transition and Hierarchy Among
Ascending Neighborhoods.” Urban Affairs Review doi:1078087418770810.
Phan, Tai. 2015. Documentation for the School Attendance Boundary Survey: School Year 2013-
14 (NCES 2015-118). Washington, DC: U.S. Department of Education, National Center
for Education Statistics.
Posey-Maddox, Linn. 2014. When Middle-Class Parents Choose Urban Schools: Class, Race,
and the Challenge of Equity in Public Education. Chicago, IL: University of Chicago
Press.
Posey-Maddox, Linn, Shelley McDonough Kimelberg, and Maia Cucchiara. 2014. “Middle-
Class Parents and Urban Public Schools: Current Research and Future Directions.”
Sociology Compass 8(4):446–56.
Pearman, Frances A. and Walker Swain. 2017. “School Choice, Gentrification, and the Variable
Significance of Racial Stratification in Urban Neighborhoods.” Sociology of Education
90(3): 213-235.

115

Renzulli, Linda and Lorraine Evans. 2005. “School Choice, Charter Schools, and White Flight.”
Social Problems 52(3):398–418.
Rhodes, Anna and Stefanie DeLuca. 2014. “Residential Mobility and School Choice Among
Poor Families.” Pp. 137-166 in A. Lareau and K. Goyette (eds.) Choosing Homes,
Choosing Schools. NY: Russell Sage.
Roda, Allison, and Amy S. Wells. 2013. “School Choice Policies and Racial Segregation: Where
White Parents’ Good Intentions, Anxiety, and Privilege Collide.” American Journal of
Education 119(2):261–93.
Saporito, Salvatore. 2003. “Private choices, public consequences: Magnet School Choice and
Segregation by Race and Poverty.” Social Problems 50(2):181–203.
Saporito, Salvatore. 2009. “School Choice in Black and White: Private School Enrollment
Among Racial Groups, 1990–2000.” Peabody Journal of Education 84(2):172–90.
Saporito, Salvatore and Daneesh Sohoni. 2006. “Coloring Outside the Lines: Racial Segregation
in Public Schools and Their Attendance Boundaries.” Sociology of Education 79(2):81–
105.
Schneider, Mark, Paul Teske, Christine Roch and Melissa Marschall. 1997. “Networks to
nowhere: Segregation and stratification in networks of information about schools.”
American Journal of Political Science 41(4):1201–1223.
Sikkink, David. and Emerson, Michael O. 2008. “School Choice and Racial Segregation in US
Schools: The Role of Parents’ Education.” Ethnic and Racial Studies 31(2):267–93.
Stillman, Jennifer. 2012. Gentrification and Schools: The Process of Integration When Whites
Reverse Flight. New York: Palgrave MacMillan.

116

Tach, Laura. 2009. “More than Bricks and Mortar: Neighborhood Frames, Social Processes, and
the Mixed‐Income Redevelopment of a Public Housing Project.” City & Community 8(3):
269-299.
Tissot, Sylvie. 2011. “Of Dogs and Men: The Making of Spatial Boundaries in a Gentrifying
Neighborhood.” City & Community 10(3): 265-284.
Weiher, Gregory and Kent Teden. 2002. “Does choice lead to racially distinctive schools?
Charter schools and household preferences.” Journal of Policy Analysis and Management
21(1):79-92.
Wells, Amy S. 2015. Diverse Housing, Diverse Schooling: How Policy Can Stabilize Racial
Demographic Change in Cities and Suburbs. Boulder, CO: National Education Policy
Center.
Wells, Amy S., Lauren Fox, Diana Cordova-Cobo, and Douglas Ready. 2018. "A Shared Future:
Addressing the Patterns of Resegregation in Urban and Suburban Contexts: How to
Stabilize Integrated Schools and Communities Amid Metro Migrations." Joint Center for
Housing Studies.
Wells, Amy S., Alejandra Lopez, Janelle Scott, and Jennifer J. Holme. 1999. “Charter Schools as
Postmodern Paradox: Rethinking Social Stratification in an Age of Deregulated School
Choice.” Harvard Educ. Review 69(2):172–204.

117

Research Ethics Statement

This research uses restricted data files from the Panel Study of Income Dynamics (PSID) and
was conducted in accordance with guidelines set forth under a restricted-use contractual
agreement designed to protect the anonymity of respondents. More information about restricted-
use data files can be obtained via the PSID website at psidonline.isr.umich.edu.

118

Appendices

Appendix Table 3.1A. Odds Ratios Predicting Greater Percent School White in Enrolled Versus
Assigned School Among Opt-Out Students
Model 1

Neighborhood Type
Family Background

Family Income (in ten thousands) 1.013 .010

Age of Youngest Child 0.973 .044

College Degree .904 .253

Married .912 .262

Owns Home 1.376 .431

Number of Kids (Under 18)

1 ref

2 .922 .276

3+ 1.137 .417

Child Factors

Non-Hisp. White (Child) 3.483*** 1.229

Grade Level

Grades KG-8 ref

Grades 9-12 1.505 .513

MSA Controls

Population Density .806+ .100

Neighborhood Factors

Neighborhood Type

Gentrifying ref

Low/Mid-SES 2.986* 1.507

Declining 3.651* 2.048

Upper SES 3.600* 2.080

Residential Tenure

Newcomer 4.587* 2.847

Neigh. Type x Newcomer

Gentrifying*Newcomer ref

Low/Mid-SES*Newcomer .281+ 0.203

Declining*Newcomer .153* 0.119

Upper SES*Newcomer .101** 0.086

School Factors

Lagged Percent School Black 4.892*** 2.357

_cons .147** .096

N(observations) 460
Notes: Logistic regression results reported as odds ratios. Robust standard errors (clustered by household)
displayed to the right of odds ratios in italics. Sample restricted to opt-out students only. The dependent
variable is a dichotomous measure capturing whether the proportion of white students in the enrolled
neighborhood school is higher than the assigned school. All models include family, child, and school
covariates. Reference neighborhood category is gentrifying neighborhoods. Newcomer refers to households
that moved school attendance boundaries within the past two years. School percent black is lagged by one
year. Homeschooled students (about 5 percent of opt-out students) and students with missing racial
composition data for the enrolled school (<1 percent) are also excluded from these analyses.
+ p<0.10, * p<0.05, ** p<0.01, *** p<0.001 (two-tailed).

119

Appendix Figure 3.1A. Predicted Probabilities of Non-Neighborhood School Enrollment in
Gentrifying and Non-Gentrifying Neighborhoods Given Public and Private Choice Options

Notes: Sample is KG-12
th
Grades. Expected probabilities correspond to results using the same
analytic strategy as used for Table 3, Model 2. All Choice refers to the number of grade-specific
magnet, charter, and private schools located within two miles of a child’s residence. All covariates
held at their respective means.

120

Chapter 4
Neighborhood Effects on Educational Outcomes: Have They Changed
Over Time?

Introduction
In this chapter, I examine how neighborhoods matter over time. I briefly review the literature
on neighborhood effects on educational outcomes and discuss analytic tools that recent work has
adopted to better align empirical models with underlying theory. I then lay out an analysis plan
for a cross-period cohort comparison of neighborhood effects on student achievement and
persistence.
While much research has examined how neighborhood context affects student outcomes, the
neighborhood effects literature has largely ignored how these outcomes may differ between
different periods. Differences in neighborhood effects by period may reflect changes in larger
societal contexts in recent decades. Linking census data on neighborhood characteristics to three
decades of data on households’ residential histories from the Panel Study of Income Dynamics
(PSID), a nationally representative longitudinal survey of households, I estimate whether the
effects of neighborhood context on children’s outcomes have changed substantially over time,
given social and demographic changes in neighborhoods. I then consider whether changes in
neighborhood effects over time could be attributed to a changing educational market, given that
schools are a key neighborhood institution but the neighborhood-school link has weakened.
To address broad questions about how the mechanisms of neighborhood inequality have
changed over time, I rely specifically on Census, NCES and restricted-use PSID data for

121

successive generational cohorts. I also draw on restricted PSID-Child Development Supplements
(CDS-I and CDS-II) data, which provide detailed information on children’s schooling experiences,
family demographic and economic data, and indicators of children’s well-being. I compare two
cohorts representing children in 1997 (CDS-I) and 2014 (CDS-II). I explore analytic models that
incorporate the indirect effects of neighborhood advantage and disadvantage on educational
outcomes while also adjusting for possible confounding (due to non-random residential selection
into different neighborhoods across time) by time-varying covariates, following recent work on
neighborhood effects (Wodkte 2018). These models also permit the exploration of multiple
potential mediators, which subsequent analyses will undertake.
Theoretically, my project casts a temporal framework on the neighborhoods effects literature
by modeling changes in effects of cumulative exposure to neighborhoods over time, thus providing
a richer understanding of how neighborhoods matter for children’s well-being.

Research Questions: Neighborhood effects over time on educational outcomes
Using successive generational cohort data from the PSID to document whether
neighborhoods have a stronger or weaker effect on children’s educational outcomes in 2014 than
they did twenty years earlier. I ask:

RQ1. How have neighborhood effects on students’ academic achievement changed over
time?
RQ2. How have neighborhood effects on students’ academic persistence changed over time?
The first step in these analyses involves simply identifying whether the association between
children’s exposure to residential contexts on educational outcomes have changed substantially
over time, given social and demographic changes in neighborhoods. Then, building off findings

122

from the first two empirical chapters of this dissertation, I lay out a plan for future analyses to
explore whether this is because schools are increasingly not a neighborhood institution. If
schools and neighborhoods are becoming dissimilar in terms of their compositions, but both are
promoting inequality in some way, what are the implications for children’s educational
outcomes?
Background
Mechanisms of Neighborhood Effects
A large body of work has sought to understand contextual explanations for persistent
disparities in educational outcomes over and above purely individual-level explanations. The
literature has generally supported the argument that residential environments matter for child
outcomes, with some work demonstrating strong positive correlations between community
affluence and academic performance (Reardon 2011; Reardon et al. 2019) and others showing
how living in disadvantaged residential contexts limits educational attainment (Jargowsky and
Komi 2011; Sastry and Pebley 2010; Sampson 2008; Wodtke et. al 2011).
The neighborhood effects literature hypothesizes several mechanisms to understand how
residential contexts influence a host of quality-of-life indicators (Jencks and Mayer 1990;
Sampson 2008; Sharkey and Faber 2014; Galster 2012). Scholars tend to group these into four
main areas: social-interactive, environmental, geographic, and institutional mechanisms (Galster
2012; Sharkey and Faber 2014). I briefly summarize these potential mechanisms through which
neighborhoods effects operate and describe how exposure to disadvantaged residential contexts
might shape educational outcomes, the focus of this study.

123

Social-Interactive Mechanisms
This broad set of mechanisms describes the influence of social ties and social processes
endogenous to neighborhoods in shaping individual outcomes. This includes peer and collective
socialization, social contagion, social cohesion and control, and social networks. Disadvantaged
neighborhoods may have lower levels of mutual trust, social capital, and social cohesion
stemming from the structural conditions of these neighborhoods (e.g., poverty, high levels of
unemployment and residential instability) (Hicks et al. 2018). As a result, children may receive
decreased collective support and social control to stave off behavior that might have a negative
impact on educational outcomes (e.g. delinquency; depression; high-risk behavior) (Sampson
2008). Disadvantaged neighborhoods may be more socially isolated than higher-resourced
neighborhoods, and their social networks may have less economic and political power to
leverage their demands (e.g. advocating for more school resources, better public services, etc.) or
less access to information and opportunities.
Environmental Mechanisms
Environmental mechanisms refer to the physical surroundings, natural and human-made, that
may shape residents’ physical and mental health. For example, physical proximity to freeways
that increase exposure to air pollution or toxic exposure to industrial infrastructure could have
negative effects on residents’ health (see also Harding et al. 2011). Toxic exposure and
environmental stress can influence student outcomes through their effects on children’s mental
health. Exposure to neighborhood violence may also negatively affect community members’
mental health and well-being by inducing stress, which has negative effects on children’s
cognition and academic performance (Burdick-Will 2018; McCoy et al. 2015; Sharkey 2010;
Sharkey et al. 2012; Wheaton and Clarke 2003). For example, Sharkey (2010) finds that children

124

exposed to acute violence in their communities exhibited less focus and performed lower on
academic tests administered in the aftermath.

Geographical Mechanisms
Geographical mechanisms refer to how individual outcomes are shaped by their
neighborhood’s location in relation to public services, resources, and opportunities. For example,
disadvantaged neighborhoods tend to have greater spatial mismatch between where they are
located and where quality employment and public services are located. Educational opportunities
are not evenly distributed across space, and socioeconomically disadvantaged neighborhoods
tend to have lower resourced schools. The quality of after-school programs and recreational
groups may also be limited in less advantaged neighborhoods.
Extreme spatial isolation leaves groups vulnerable to extreme social isolation, along with all
its accompanying ills (Mare 2011, Sharkey and Elwert 2011). Institutionalized residential
isolation has the potential to persist and weigh heavily on successive generations. Thus, social
and spatial isolation can breed persistent hardship as neighborhoods themselves have the
potential to transfer a legacy of advantage or disadvantage intergenerationally (Mare 2011;
Sampson et al. 2008; Sharkey and Elwert 2011).

Institutional Mechanisms
Institutional mechanisms refer to the local institutions that shape the social life of
neighborhoods. Local schools are key neighborhood institutions that shape children’s lives.
Schools located in disadvantaged neighborhoods tend to have fewer social, instructional, and
economic resources than schools located in higher-SES neighborhoods. The disparity in
educational resources between the most and least disadvantaged neighborhoods is even greater in

125

highly segregated metropolitan regions (Owens and Candipan 2019). Disadvantaged
neighborhoods often have decreased access to services on which parents often rely such as child
care and quality transportation, thus constraining the choices of parents residing in disadvantaged
neighborhoods to move to higher-resourced schools.

Injecting temporal dimensions into the study of neighborhood inequality
Urban theory has long suggested that it is not only one's exposure to disadvantaged
neighborhood contexts, but also the timing and duration of that exposure that can have enduring
effects on individual outcomes (Jencks and Mayer 1990; Massey and Denton 1993; Wilson
1987). Until recently, however, most studies measured neighborhood effects using cross-
sectional designs that capture only a single point in time. Neighborhoods are not static units and
the mechanisms by which residential contexts shape individual outcomes are likely to take shape
over a longer period of time. In recent years, a growing number of studies have incorporated
temporal dimensions of neighborhood effects into empirical models to align better with theory.

Time as Recency
Some work has examined time in terms of whether the timing or sequence of neighborhood
exposures matters. For example, recent exposure may have a stronger influence on student
outcomes. Recency of exposure may matter in the short-term if we think that there could be
acute shocks experienced at the community level, but only a handful of studies have modeled the
timing of short-term exposure into their analytical design. In a Chicago study examining
whether child test takers were affected by a recent homicide having taken place within the
community, Sharkey (2010) finds for black students that the negative effects were strongest in
the days following a homicide in the community, but that the magnitude decayed the further

126

away the homicide took place and diminished as days passed. Burdick-Will et al. (2011) find that
moving to less disadvantaged neighborhoods has a positive effect on test scores, aligning with
work from Hicks et al. (2018) which observes that recent exposure to neighborhood contexts
significantly affects student achievement.

Time in a Critical Age framework
Other neighborhood effects studies have examined time through a critical age framework.
These studies investigate whether neighborhood effects are stronger during different stages of
early childhood and adolescence. The evidence is mixed. While some find that children may be
more sensitive to neighborhood institutional resources during childhood as compared to
adolescence and that early exposure to neighborhood disadvantage may have larger effects on
academic performance in the short- and longer-term (Heckman 2006; Johnson and Schoeni 2011;
Sharkey 2013), other work finds that disadvantaged residential contexts matter more during
adolescence (Chetty et al. 2014; Wodtke et al. 2011). These differences are likely the result of
differentiated mechanisms primary to each age period and setting (Ellen and Turner 1997).
Time as Intergenerational
Research also suggests that neighborhood context can be transmitted intergenerationally
(Sharkey 2013). Those that reside in the most disadvantaged neighborhoods are more likely to
have children and grandchildren that also live in disadvantaged neighborhoods, suggesting that
duration of neighborhood exposure has lagged effects that are far reaching. This work highlights
ways in which neighborhood environments have the potential to create multigenerational
legacies of advantage and disadvantage (Sampson and Sharkey 2008; Mare 2011; Sharkey and
Elwert 2011).

127

Time as Cumulative Exposure
Although early foundational theoretical arguments implied the importance of extended
exposure to disadvantaged neighborhood contexts for generating negative effects, research has
only recently heeded calls to inject a temporal dimension into the empirical literature analyzing
the effects of exposure to residential contexts on individual outcomes (Sharkey and Faber 2014).
Earlier research examining neighborhood effects on educational outcomes at a single point in
time was mixed and often generated null or modest effects (Sampson et al. 2002; Sharkey and
Faber 2014). Aside from their empirical limitations, such point-in-time approaches assume that
individuals are influenced by their residential environments in isolation, overlooking how the
impact of exposure to disadvantaged residential contexts neighborhood accumulates and persists
throughout the life course (Crowder and South 2011; Sampson et al. 2008; Sharkey & Elwert
2011; Wodtke et al. 2011; Wodtke et al. 2016).
Recent scholarship has attempted to examine how long-term exposure to disadvantaged and
advantaged contexts affects a host of quality-of-life indicators (economic, social, occupational,
residential, educational, health, etc.). Generally, these studies have found a stronger magnitude of
neighborhood effects on individual outcomes when accounting for the length of exposure to
different types of neighborhood contexts. (Crowder and South 2011; Sharkey and Elwert 2011;
Wodtke et al 2011).
In a review of neighborhood effects literature since 2002, Sharkey and Faber (2014: 567)
conclude that the most consistent finding from studies explicitly modeling duration of
neighborhood exposure “is that the effect of neighborhood disadvantage on cognitive and
academic outcomes is more severe if disadvantage is persistent, experienced over long periods of
a family’s history.” The authors called on future research to build on these studies by
incorporating a temporal lens that accounts for the duration of exposure to residential contexts,

128

taking advantage of improved longitudinal data sources and advancements in analytic tools that
allow for richer analyses that align better with neighborhood effects conceptual models. The
present study answers calls to inject a temporal perspective into neighborhood effects studies by
examining cumulative exposure to residential contexts. Importantly, this study addresses an
empirical hole by also incorporating an overlooked dimension of time: how neighborhood effects
have changed over time.

Why might neighborhood effects have changed over time?
While much research has examined how neighborhood context affects student outcomes, the
neighborhood effects literature has largely ignored how these outcomes may differ between
different periods. For example, in their study on the effect of cumulative childhood exposure to
neighborhood disadvantage on high school graduation, Wodtke et al. (2011) pool all children that
were age 1 between 1968 and 1978, and follow them through age 20. Their study period thus
spanned a time during which high school completion rates began increasing overall. In their study
examining associations between neighborhood SES on high school graduation, Crowder and South
(2011) combine data on all black and white survey participants born between 1968 and 1980 for
whom educational attainment was known at age 25. In both studies, pooling data on children’s
outcomes over several years may mask how broader secular trends could potentially alter the
association between residential environments and student outcomes.
The present study casts a new temporal framework on the neighborhoods effects literature by
modeling changes in effects of exposure to neighborhoods over time, thus providing a richer
understanding of how neighborhoods matter for children’s well-being. Differences in
neighborhood effects by period may reflect broader secular changes in societal contexts in recent
decades. There are many possible explanations for changing neighborhood effects over time.

129

Drawing on hypothesized mechanisms identified in the neighborhood effects literature, I briefly
discuss just a few.
The decline in public space in recent decades might limit children’s peer interaction with
other families and children living the same neighborhood. In disadvantaged neighborhoods, this
may exacerbate social isolation. Rising residential income segregation may also result in
segregated activity spaces between more and less advantaged children. Increased income
segregation between families with children may further widen resource disparities between
families residing in high- and low-SES neighborhoods. Persistent residential segregation may
further exacerbate geographic mismatches between disadvantaged neighborhoods and access to
quality public services—e.g. childcare, libraries, and public transportation—which may
negatively impact student outcomes. On the other hand, the decline in violent crime in U.S. cities
may indirectly weaken neighborhood effects on educational outcomes over time for some
children through its influence on children’s mental health and cognition.
The ways in which families sort into neighborhoods and schools has also changed over time,
reconfiguring the social and spatial configuration of cities. Rising income inequality is further
concentrating wealth within and between cities and increasingly isolating affluent communities
from the rest (Reardon and Bischoff 2016). Broader residential processes such as gentrification
are also reshaping the composition of neighborhoods in many cities, producing demographically
integrated neighborhoods while maintaining segregated neighborhood institutions, such as
schools, as white and higher-SES parents increasingly bypass local schools for non-local options
(as demonstrated in Chapters 2 and 3 of this dissertation).

130

School and neighborhood decoupling
The changing relationship between neighborhoods and local schools is perhaps most relevant
to this study’s focus on changing neighborhood effects on educational outcomes. Institutional
resource theory views schools as key neighborhood institutions and potential mediators of
neighborhood effects (Galster 2012; Jencks and Mayer 1990; Sharkey and Faber 2014). In past
decades, when neighborhoods were more tightly linked, the composition of a child’s
neighborhood correlated more closely with the composition of a child’s school. However,
expanding school choice since the 1990s has weakened the tight match between neighborhoods
and neighborhood schools. Recent work examining the neighborhood-school link finds that the
demographic composition of neighborhoods and schools in the U.S. has increasingly diverged
since 2000—more white students reside in neighborhoods than attend the neighborhood school
(Bischoff and Tach 2018; Mader et al. 2018; Candipan 2019). If the neighborhood-school link
has weakened over time, and schools are no longer key neighborhood institutions, then the
mediating effects of neighborhood schools may not be as strong relative to periods when most
children attended their neighborhood school. Children may be less affected by their home
neighborhood environments if they travel outside of the neighborhood for school. Moreover,
children residing in the same neighborhood may have completely non-intersecting lives, thus
potentially diminishing the social-interactive role of neighborhoods (Burdick-Will 2017; Sharkey
and Faber 2014).

Methodological Challenges in Neighborhood Effects Research
Providing an unbiased estimate of the magnitude of neighborhood effects has been a
methodological challenge for researchers. The central challenge has been avoiding selection
bias—the notion that there is a systematic spatial selection process in which individuals sort into

131

certain types of neighborhoods differently based on particular measured and unmeasured
characteristics. Analyses that do not account for this selection process properly will therefore be
biased if there are unmeasured characteristics that shape individuals’ sorting preferences that are
left unaddressed (Galster 2012). Scholars hoping to study cumulative exposure to neighborhood
effects are often faced with the additional challenge of accounting for non-random selection into
neighborhoods over the life course. Researchers must account for time-varying individual and
household factors that affect both selection into current, as well as future, neighborhoods.
33

Early studies employing conventional regression, often relying on cross-sectional data,
resulted in findings that were inconclusive and inconsistent, with many studies producing small
or no effects. Other studies attempted to address selection bias by naively conditioning on time-
varying household characteristics that shape residential decisions, but this approach ignores how
these time-varying confounders may also be affected by past treatment, thereby inducing
potential bias by overcontrolling the indirect pathways associated with child outcomes. This can
result in underestimating the effects of neighborhood disadvantage on child outcomes (for
reviews, see Sampson et al. 2002 and Sharkey and Faber 2014). Subsequent work employed
methods such as multilevel modeling and matching techniques (e.g. propensity score matching
with sensitivity analyses), yet such static models insufficiently treated selection bias and, in the
case of multilevel models, also overcontrolled for contextual effects beyond the neighborhood
(e.g. family; school; peer; etc.) (Harding 2003).
Experimental and quasi-experimental studies, such as those examining the outcomes of
children participating in Moving to Opportunity (MTO) and Gatreaux, attempted to address the
issue of non-random selection into neighborhoods through more “natural” designs, but voucher

33
Moreover, current “treatments” may also affect future time-varying “risk factors.” In this case, residence in
particular neighborhoods affect time-varying individual and household factors in subsequent years.

132

families were not always able to move freely to more advantaged neighborhoods, thus disrupting
the effectiveness of these studies’ experimental design. Study families living in high-poverty
neighborhoods were often constrained in ways that did not allow them to activate housing
vouchers or move to less disadvantaged neighborhoods, instead making lateral residential moves,
thus leading to underestimated or null effects of moving out of a high-poverty neighborhood
(Clampet-Lundquist and Massey 2008). Moreover, the effect of moving to a more advantaged
neighborhood may actually represent a short-term “shock” and overlooking how the duration of
prior exposure influences child outcomes (Sampson 2008; Sharkey and Faber 2014).
Pulling out neighborhood effects is thus challenging, in large part because of the data
required to match empirical models to theoretical frames that highlight the role of cumulative
exposure to neighborhood disadvantage (i.e. information on children’s family background and
residential locations over time; etc.) while also accounting for endogenous selection bias. In
subsequent work, better longitudinal data sources and analytic tools enabled researchers to
address some of these methodological limitations of earlier studies using a variety of techniques
(Crowder and South 2011; Sampson et al. 2008; Sharkey and Elwert 2011; Wodtke et al 2011).
Two recent approaches, borrowed from epidemiology, address earlier methodological concerns
regarding non-random neighborhood selection: 1) marginal structural models (MSM) with
inverse probability of treatment (IPT) weights; and 2) regression with residuals (RWR) on a
structural nested mean model (Robins et al. 2000; Almirall et al. 2013). These models adjust for
observed confounding in ways that permit estimation of marginal effects in a longitudinal
setting. I briefly describe these models below, then further discuss features and assumptions for
both in subsequent sections.

133

Marginal structural models with inverse probability of treatment weighting (MSM-IPTW)
Standard methods are biased when time-varying “risk” factors also predict subsequent
treatment (e.g. neighborhood disadvantage) and when past treatment history (e.g. prior exposure
to various levels of neighborhood disadvantage) predicts subsequent risk factor levels.
34
Using
standard methods introduces confounding if we do not adjust for time-varying risk factors, but
introduces selection bias if we do attempt to adjust for these same factors. Inverse probability of
treatment (IPT) weighting on a marginal structural model (MSM) provides a solution to adjust
for both confounding and selection bias.
35
Confounding adjustment is achieved through IPT
weights—i.e. the untreated and treated are made comparable across all levels of confounders—
which are then placed into a weighted regression model to estimate either marginal or
conditional effects of treatment.

Regression with residuals on a constrained structural nested mean model (SNMM-RWR)
Marginal structural models with inverse probability of treatment weighting are constructed
using binary or ordinal “treatments” at each study wave. Performing MSM-IPTWs thus comes
with a loss of precision from binning continuous treatments (e.g. recategorizing continuous
neighborhood disadvantage scores into a five-category ordinal treatment). Recent work proposes
a new method for neighborhood effects researchers that better handles continuous treatments
and, while able to account for time-varying confounders that affect both the treatment and the
outcome, is similar to conventional regression is its implementation. Regression-with-residuals
(RWR) estimation of a constrained structural nested mean model (SNMM) provides an

34
For example, family income is a “risk” factor that predicts current treatment (residence in a child’s current
neighborhoods), but a child’s current neighborhood (“treatment”) also affects future family income (future “risk”).
35
The model is considered marginal because the effect may be estimated for the entire population, and structural
because models for counterfactual outcomes are sometimes referred to as “causal” or “structural” (Robins and
Hernan 2019).

134

alternative method of estimating marginal effects, extending the SNMM-RWR estimator
originally “designed for analyses of moderated, or subpopulation average, treatment effects
(Robins 1994; Almirall, Ten Have, and Murphy 2010; Almirall et al. 2013; Wodtke et al. 2016;
Wodtke and Almirall 2017). This two-stage method: (1) residualizes time-varying confounders
in the first step, thus purging confounders of their association with prior treatments; (2) then
regresses the outcome on those residualized confounders and past treatments in the second step.
Under the assumption of no effect moderation across levels of time-varying confounders, the
moderated effects in a constrained SNMM are equivalent to marginal effects of interest (Wodtke
2018).

Data
To answer questions about how neighborhood effects on children’s educational outcomes
have changed over time, I link neighborhood data from Census and American Community
Survey (ACS) to the following:
• Restricted household- and individual-level data from the Panel Study of Income Dynamics
(PSID), a nationally representative longitudinal survey of households that began in 1968
with annual follow-up interviews until 1996, and biennial interviews starting in 1997. The
restricted-use PSID Geomatch files contain geographic identifiers down to the census block
level which allow me to observe households’ residential history across survey waves. The
geospatial identifiers also allow me to merge neighborhood contextual data from the census
onto households and individuals.

• Restricted PSID-Child Development Supplements (CDS-I and CDS-II) data, a
complementary component to the Main Interview administered to sample children. The
original CDS-I began interviewing children ages 0-12 in 1997 (N=3563), with follow-up
interviews conducted for age-eligible (<18 y/o) children in 2002-3 (N=2907) and 2007
(N=1608). A new CDS-II survey was introduced in 2014 (N=4333) to follow children ages

135

0-17 born since the start of CDS-I in 1997.
36
The CDS-97 and CDS-2014 represent
successive generational cohorts (what PSID calls an “intergenerational” cohort design),
facilitating a clean cross-cohort comparison (i.e. children eligible for the CDS-14 sample are
the offspring of CDS-97 (and earlier) children).
37

I restrict my sample to black and white CDS children for whom residential histories were
recorded. Following most quantitative work on neighborhood effects, I use census tract as my
proxy for neighborhood.
A benefit of the longitudinal design of the PSID-CDS is that I can link children’s households
to their neighborhood contexts in each wave using a nationally representative sample. Being able
to capture children’s full residential histories allows for a richer analysis on how cumulative
exposure to neighborhood environments shapes children’s educational outcomes. Because I
know the neighborhood in which children reside in each wave, I can also account for changes in
neighborhood contexts, including transitions into and out of disadvantaged neighborhoods, the
duration and age at which children reside in various types of neighborhoods, and the recency of
exposure. Moreover, while much past work has documented neighborhood effects on children’s
outcomes in two cities—Chicago and Los Angeles (Burdick-Will et al. 2011; Hicks et al.
2018)—the current study takes advantage of PSID’s broader geographic scope which allows for
an examination of neighborhood effects over time across at a national level.

36
Although CDS-2014 collected data on a total of 4333 children, the in-home interview and educational assessment
data was only registered for a subset of children from this overall sample.
37
See https://psidonline.isr.umich.edu for more information.

136

Measures
Cumulative Neighborhood Disadvantage
My research question asks how children’s cumulative exposure to disadvantaged
neighborhood contexts affects educational outcomes, and how this has changed over time. To
answer this question, I must first construct a measure that captures long-term exposure to
neighborhood contexts.
My key independent variable is an indicator of cumulative exposure to neighborhood
disadvantage, a composite measure summarizing several highly correlated characteristics of
neighborhood socioeconomic status identified in past work as influential factors on child
outcomes (Wodtke et al. 2011). I construct this measure in three main steps.
First, I use census tract identifiers (provided via restricted-use Geomatch files) for each wave
to observe the residential histories of each child in my sample. Second, I use factor analysis, a
method of dimension reduction, on Census/ACS data to construct a tract-level composite
measure of neighborhood disadvantage score based on indicators widely used in previous
research (Sampson et al. 1997; Wodtke et al. 2011): proportion of female-headed households,
family-level indicators of poverty rates, public assistance receipt, unemployment, education, and
occupational structure. I thus reduce several census tract components into a single neighborhood
SES score in 1980, 1990, 2000, 2010, and 2015.
38
I restrict my analyses to tracts located in
census-defined metropolitan regions. The raw neighborhood disadvantage score in each year
loads on to a single component, with a mean of zero and a standard deviation of 1 based on the
overall sample of census tracts in each year. After constructing my measure of neighborhood
disadvantage in 1980, 1990, 2000, 2010, and 2015, I then register neighborhood disadvantage

38
For 1980-2000, I use the long-form decennial census; I use ACS 5-year estimates for 2008-12 and 2013-2017,
using 2010 and 2015 as the midpoints.

137

scores for each survey wave using linear interpolation to impute tract-level data for intercensal
years between 1980 and 2015.
While children may be exposed to different residential environments by relocating to a new
neighborhood, neighborhoods may also change, thus exposing children to varied residential
contexts for some that physically stay in place. By calculating a neighborhood disadvantage
score in each study year, I am able to capture some children’s exposure to changing residential
contexts simply from staying in place while their neighborhood conditions change.
Next, following Wodtke et al. (2011), I use my neighborhood disadvantage raw scores (a
continuous measure) to construct a new 5-category indicator ordered by neighborhood
disadvantage quintile. I do this by ranking each neighborhood’s raw score, for each wave, to
other neighborhoods within the same MSA. This accounts for differences between metro regions
(in each year) on the tract-level components used to generate factor scores. Higher number
quintiles indicate the most disadvantaged neighborhoods, and vice versa (i.e. quintile category
5= most disadvantaged neighborhood; quintile category 1= least disadvantaged neighborhood).
I observe residential histories for children as they age from 0 until the age 18 (for models
examining persistence) or from 0 until the age at which they take the test (for models examining
achievement), which I then link to my measure of neighborhood disadvantage for each wave.
Then, for each child, I register the wave-specific neighborhood disadvantage quintile at each
wave and calculate the mean across all study years to construct my duration-weighted measure of
cumulative neighborhood disadvantage. The resulting measure is the average of ordinal wave-
specific “treatments” which indicates the degree to which children are exposed to disadvantaged
neighborhood contexts during early childhood.
39

39
Constructing this measure presents two challenges. First, there is a mismatch between waves per cohort; children
from earlier cohorts have more waves of data because PSID administered the survey annually prior to 1998. To

138

Measures of student persistence and achievement
My analyses focus on two types of educational outcomes: (1) one measure of academic
achievement, as measured by test scores on the Woodcock-Johnson Revised Tests of
Achievement (WJ-R); and (2) one measure of academic persistence, as measured by the odds of
on-time high school graduation.

Student Achievement
In the first set of models, my dependent variables are measures of academic achievement.
The WJ-R test is comprised of a series of nine subtests that measure several dimensions of
academic achievement. W-JR tests were administered to age-eligible children in each CDS wave,
with raw scores standardized in each year based on age and national norms ensuring
comparability between children in different age groups (McGrew, Werder, and Woodcock 1991).
Among all test-takers, standardized scores have a mean of 100 and a standard deviation of 15.
For my analyses, I focus on the Letter Word Recognition subtest, administered during all CDS
waves to children ages 3-17. In my analysis sample, children’s standardized scores have a mean
of 104 and standard deviation of 19, both slightly above the national average.
Student Persistence
After examining changing neighborhood effects on student achievement over time, I then
perform a series of parallel models examining academic persistence, as measured by on-time

maintain consistency between cohorts, I record neighborhood context for children every other year regardless of
whether they have single-year data. Second, the age at which children registered their test scores was substantially
later, on average, for the earlier cohort compared to the later cohort (i.e. CDS-97 vs. CDS-14 children). This means
that the duration-weighted neighborhood disadvantage score is based on a shorter period of time for children in the
later cohort.

139

high school (HS) graduation. I define on-time HS graduation as twelve years of educational
attainment without any break in grade.

Period cohorts
To examine how the effect of children’s exposure to neighborhood disadvantage on
educational outcomes has changed over time, I pool data on two period cohorts of children. For
both set of analyses—models examining neighborhood effects on (1) student achievement and
(2) student persistence—I compare children from the CDS-97 cohort to children in the CDS-
2014 cohort. I construct a binary cohort indicating a child’s cohort membership.
For models examining neighborhood effects on student achievement (RQ1), I compare
children from the CDS-97 cohort to children in the CDS-2014 cohort. I construct a
binary cohort indicating a child’s cohort membership. For models examining student persistence
(RQ2), I again pool data on two period cohorts, but use a slightly different definition for cohort
membership: 1) children ages 2-4 y/o at the start of CDS-97; 2) children ages 10-12 y/o at the
start of CDS-97. Children in the first cohort (2-4 y/o in 1997) would have graduated high school
in 2011-2013, while children in the second cohort would have graduated in 2003-2005, while
children in the second cohort would have graduated high school in 2011-2013.
40

Time-invariant individual and family covariates
I include a battery of time-varying individual and family factors, including child age at time
of test (for “achievement” models), mother’s marital status, unemployment status, family
income, number of children in the household, home ownership, and whether the family received

40
Because CDS-14 children span the ages of 0-17 in 2013, they are largely ineligible to graduate high school by
2015, thus preventing me from including them in my analysis. As data releases are made available for later years, I
hope to eventually incorporate these children into my analyses examining neighborhood effects on academic
persistence.

140

public assistance in that year. I also incorporate time-invariant measures, such as sex, mother’s
age at birth, and whether a child’s primary caregiver is college-educated. All models are
stratified by race, which I restrict to non-Hispanic white and non-Hispanic black.

Analysis Plan
I perform two types of models recently adopted by neighborhood effects researchers,
outlined next, which are both appropriate for handling confounding in longitudinal settings.
MSM-IPTW
For MSM-IPTW models, the analysis entails two main steps: First, to incorporate the indirect
effects of neighborhood disadvantage on educational outcomes while also adjusting for possible
confounding (due to non-random residential selection into different neighborhoods across time)
by time-varying covariates, I use inverse probability treatment weights, a propensity score
technique for longitudinal data. IPT weights are used to create a “pseudo-population,” which is
generated by simulating random assignment across different levels of neighborhood exposure at
each wave based on observed characteristics of individuals (e.g., parent education, family size,
household income, family homeownership, etc.).
41
These weights are called inverse probability
of treatment weights because they capture, for each subject, the inverse of the probability of
receiving the treatment they actually received, conditional on their covariate history. The benefit
of IPT weights is that they ensure that these observed individual-level factors are balanced in
expectation across different levels of neighborhood disadvantage for each wave. In practice, I
construct these weights by using ordinal logistic regression (my “weighting model”) with my 5-

41
Earlier cohorts have more available waves of data because PSID administered the survey annually prior to 1997.
To maintain consistency between cohorts, I record neighborhood context for children every other year regardless of
whether they have single-year data.

141

category indicator of neighborhood disadvantage to predict membership into the neighborhood in
which a child actually lives based on time-varying and time-invariant household factors that
shape residential decisions, and then taking the inverse of that predicted probability, following
Wodtke et al. (2011). In a longitudinal setting, this requires calculating IPT weights for each
study wave.
Effect estimates based on IPT weights can be large and unstable, resulting in findings that are
driven by just a few cases. Researchers suggest addressing this by constructing stabilized weights
which often result in reduced variance. I construct stabilized weights by replacing the numerator
with the predicted probability of treatment (i.e. neighborhood disadvantage) in the prior wave
conditional only on the time-invariant baseline covariates included in the denominator of the
original weighting model.
42
Then, for each child, I multiply across waves to create my final
stabilized weight, which I truncate at the 1st and 99th percentiles to achieve greater consistency
of results and avoid disproportionate influence from outliers, following past work (Cole and
Hernán 2008; Sharkey and Elwert 2011; Williamson and Ravani 2017; Wodtke et al. 2011).
Since my outcome measure of achievement is continuous, I place the weights into a linear
OLS model. In these models, I include both time-invariant child and parent factors, as well as
covariates at k=0. Because the construction of stabilized IPT weights required baseline
confounders in both the numerator and denominator, I must also include these baseline controls
in my regression model.
43
In the second stage, I place my IPT weights into a linear regression
model that examines changes in the effects of exposure to early childhood neighborhood

42
In other words, the numerator of my stabilized weights is the probability of a child’s observed treatment
regardless of time-varying factors (Hernan and Robins 2019).
43
In these analyses, baseline refers to the wave in which a child first enters into the sample (i.e. the first recording
of residential contexts).

142

disadvantage on academic achievement (RQ1).
44
My key independent variable is my measure of
cumulative exposure to neighborhood disadvantage. The resulting coefficient for duration-
weighted neighborhood disadvantage in the full model indicates the effect of moving from
quintile q to quintile q-1 on children’s achievement scores (e.g. moving from most disadvantaged
neighborhood (q5) to the next most disadvantaged neighborhood (q4)).
Because prior research finds that neighborhood attainment varies substantially by race, I run
race-stratified models (Charles 2003; Iceland and Scopilliti 2008; Wodtke et al. 2011). I
construct an interaction term between cumulative neighborhood disadvantage and cohort to
compare differences between cohorts in neighborhood effects. The interaction term between
cohort and neighborhood disadvantage indicates the difference in how early childhood exposure
to neighborhood disadvantage predicts odds of high school graduation between the two period
cohorts. By race, these weighted regression models address the question of how neighborhood
effects on academic achievement has changed over time. I then perform a parallel set of analyses
examining children’s exposure to neighborhood disadvantage on academic persistence (RQ2).
Because my dependent variable for this model is a binary indicator for having graduated high
school, I place my weights into a logistic regression model that examines changes in the effects
of exposure to early childhood neighborhood disadvantage on academic persistence. In this
model, the interaction term between cohort and neighborhood disadvantage indicates the
difference in how early childhood exposure to neighborhood disadvantage predicts odds of high
school graduation between the two period cohorts. By race, these weighted regression models
address the question of how neighborhood effects on academic persistence has changed over
time.

44
In these weighted models, children underrepresented in treatment assignment are given proportionally more
weight, whereas higher-represented children are given proportionally less weight.

143

SNMM-RWR
Performing MSMs with IPT weights comes with a loss of precision from binning continuous
neighborhood disadvantage scores into a five-category ordinal treatment. Recent work proposes
a new method for neighborhood effects researcher that better handles continuous treatments and,
while able to account for time-varying confounders that affect both the treatment and the
outcome, is similar to conventional regression is its implementation. Regression-with-residuals
(RWR) estimation of a constrained structural nested mean model (SNMM) provides an
alternative method of estimating marginal effects. This method extends the original SNMM-
RWR estimator which is designed for analyses of moderated, rather than marginal, effects
(Robins 1994; Almirall, Ten Have, and Murphy 2010; Wodtke et al. 2016; Wodtke and Almirall
2017). Assuming there is no effect moderation of the treatment (i.e. neighborhood disadvantage)
on time-varying confounders, RWR on a constrained SNMM results are equivalent to marginal
effects, thus providing an alternative method of estimating the marginal effects of a time-varying
treatment (Wodtke 2018).
45

Under the right conditions, regression with residuals on a constrained structural nested mean
model is potentially more efficient and precise than MSMs with IPT weights. Implementing
SNMM-RWR is similar to conventional regression and simpler than MSMs with IPT weights.
This method relies on a potential outcomes framework and proceeds in two steps. In the first
step, I residualize all time-varying confounders at each wave, which effectively purges
confounders of their association with prior treatments. In the second step, I regress the outcome

45
In situations where time-varying effect moderation occurs, Wodtke and Almirall (2017) propose hybrid
approaches that combine features of IPT weighting and RWR (IPTW-RWR), such as weighting to adjust for a
subset of moderating time-varying confounders, while the remaining non-moderating confounders are adjusted for
directly in an SNMM fit to an weighted pseudo-sample by the method of RWR. (Wodtke 2018).

144

on prior treatments (i.e. neighborhood disadvantage quintile in each study wave), as well as the
residualized confounders controls. Incorporating these residualized transformations in my
second-step regression model enables me to account for confounding while avoiding bias from
overcontrol and endogenous selection (Wodtke 2018). Results from these models thus indicate
the effect of moving from more disadvantaged neighborhoods to less disadvantaged
neighborhoods.

Assumptions of MSM-IPTW and SNMM-RWR
In recent years, MSMs with IPT weights have been a commonly adopted method of
addressing selection bias in the neighborhood effects literature. While these models mark a vast
improvement to earlier studies using naïve regression and matching estimators, ensuring
unbiased results requires meeting a set of assumptions (Robins et al. 2000). First,
exchangeability implies the absence of unmeasured confounding inducing correlation between
treatment (exposure) and residuals. Second, consistency requires that each individual’s observed
outcome is the causal outcome resulting from each individual’s set of observed treatments.
Third, positivity requires that the control group must have a non-zero chance of being treated,
and vice versa (for the treatment group). Finally, MSMs requires that the model used to generate
IPT weights must be correctly specified, which is a general assumption of most statistical
models.
Regression with residuals on a structural nested mean model improves on MSM in some key
ways, but these models also carry strong assumptions. In addition to the assumptions outlined
earlier for MSM-IPTW models, SNMM-RWR models require correct model specification for
both the causal and nuisance functions. Moreover, to estimate unbiased marginal effects, the
constrained SNMM assumes no effect moderation between the treatment and time-varying

145

confounders in each wave. This may be a particular strong assumption to meet given that prior
urban theory and research has suggested that, while exposure to neighborhood advantage may
further boost educational outcomes for already-advantaged children, it may simultaneously result
in relative deprivation for nonwhite (particularly black) and lower-SES children. Thus, although
SNMM-RWR provides a potentially more efficient and precise estimator than MSMs with IPT
weights, these models do come with potential bias-variance tradeoffs when models are
misspecified or unable to satisfy these strong assumptions.
Given that assumptions for both models may be difficult to meet in practice (e.g. models with
large numbers of study waves and confounders), Wodtke (2018) recommends employing both
methods to determine which performs best.

Preliminary Findings: Neighborhood Effects on Student Achievement
In this section, I present preliminary results from student achievement models (RQ1), then
discuss next steps for student persistent analyses (RQ2).

Descriptive Statistics
Table 4.1 presents sample characteristics for my analysis sample. CDS-97 children (Cohort
0) represent roughly two-thirds of the overall analysis sample. The age at which children took the
LW subtest ranged from 13 to 17 for CDS-97 children and 3 to 17 for CDS-14 (Cohort 1)
children. There were slightly more black students in my sample overall (52.8 percent black
compared to 47.2 percent white students, but the disparity is more pronounced in the CDS-14
cohort (56.9 vs. 43.1 percent). Both cohorts have roughly the same proportion of male and
female students. Recall that higher numbered quintiles indicate relatively more disadvantaged
neighborhoods. Most neighborhoods in my sample are categorized in the quintile 5 (most

146

disadvantaged neighborhood), which reflects the design of the PSID which initially oversampled
low-income black families in 1968 for its first study wave. Neighborhoods in upper quintiles
have a majority black resident population, on average, while racial composition in the lower
quintile neighborhood is predominantly white. Though not directly measured in my analysis, it is
important to remember that residential segregation by race and income are highly correlated, and
likely play a role in exacerbating disparities in student outcomes between white and black
students residing in advantaged and disadvantaged neighborhood settings.

Academic achievement across neighborhood types
Next, I explore the distribution of test score percentile rankings among children in the CDS-
97 and CDS-14 cohorts. Table 4.2 presents preliminary descriptive statistics and unadjusted
mean achievement (LW subtest percentile ranking) for (1) the overall analysis sample and (2) by
period cohort. Overall, children in my analysis sample performed slightly better than the national
average (among all test-takers) in terms of average test score percentile rank (53.8).
46

The next rows display unadjusted mean achievement (i.e. percentile rank) by cohort and
neighborhood disadvantage quintile.
47
Not surprisingly, overall test score rankings are highest in
the least disadvantage neighborhoods (quintile 1), but decline in upper quintiles across both
cohorts. These averages, however, may mask heterogeneous trends between white and black
students over time. Next, I capture mean achievement separately for white and black students (by

46
The majority of CDS-97 children completed the LW subtest in multiple waves. For children with multiple test
scores, I consider only the most recent test.
47
To calculate mean achievement rankings by neighborhood quintile, I assign a child’s same test score ranking to
each neighborhood quintile in which a child matches during the study period. Children may only be exposed to a
particular type of neighborhood for one wave, and they may reside in another for the majority of childhood. My
approach may therefore give more weight to neighborhoods in which children reside for shorter periods of time,
which is a limitation to consider when interpreting these results. Moreover, while I have full residential histories for
almost all CDS-97 children (i.e. neighborhood data from age 0-17), I have shorter residential histories for CDS-14
children because most took their W-JR LW tests well before age 17.

147

cohort and neighborhood quintile). Overall, white students, on average, ranked substantially
higher than the national average (64.6 percentile rank). This largely reflects the disproportionate
number of higher-SES white students in my sample exposed to advantaged neighborhood
contexts (descriptive results now shown).
Black students performed substantially better in 2014 relative to earlier decades. Test score
rankings were also much higher for black children in Cohort 1 compared to black children in
Cohort 0. Black students in Cohort 1 narrowed the gap in percentile test rankings with white
students in Cohort 1. The disparity between black and white students’ test score rankings, on
average, was about three points in the most disadvantaged neighborhoods and roughly seven
points in the least disadvantaged neighborhoods in 2014. Nonetheless, the persistent disparity in
test score performance reflects the enduring impact that decades of an unequal opportunity
structure and accumulated educational debts have heaped on students of color (Ladson-Billings
2006).
Inter-cohort trends in achievement for white and black students also diverged. Among white
students, the achievement gap in test score rankings between students residing in the most and
least disadvantaged neighborhoods narrowed over time. On the other hand, while test scores
among black students in later years increased overall, the gap in test score ranking between the
most and least disadvantaged neighborhood quintiles widened substantially, suggesting the
stickiness of negative impacts on education for black children exposed to the most disadvantaged
neighborhood environments.
48

48
In preliminary descriptive analyses (not shown), there appears to be further subgroup heterogeneity by sex with
black girls in 2014 ranking substantially higher than earlier decades, and girls generally ranking higher than boys
across both cohorts, on average. Since there are relatively fewer black children residing in the lowest (least
disadvantaged) quintiles in my sample, this could be a case of a very small subgroup of children driving results.

148

Multivariate Analyses
This study compares results across three types of models: (1) conventional OLS with
covariate adjustment; (2) MSM with IPT weights; and (3) RWR on a constrained SNMM. I
estimate the effect of children’s cumulative exposure to neighborhood disadvantage first by
pooling the entire sample, then by isolating the effects on my two period cohorts (i.e. CDS 97 vs.
CDS 14 children).
I present predicted values from three modeling approaches to analyze how neighborhood
effects on academic achievement have changed over time (RQ1). First, I perform conventional
regression using covariate adjustment (Figure 4.1). Second, I estimate the parameters of a
marginal structural mean model with a continuous treatment (i.e. duration-weighted
neighborhood disadvantage) using stabilized IPT weights (Figure 4.2). Third, I use the
regression-with-residuals estimator on a constrained structural nested mean model (Figure 4.3).
Since patterns are generally similar across models, I focus my discussion on MSM-IPTW,
the more frequently adopted analytic approach in recent work on neighborhood effects in
longitudinal settings. Figure 4.2 presents preliminary predictive margins from race-stratified
models regressing academic achievement (LW percentile rank) on cumulative neighborhood
disadvantage, cohort, and their interaction, as well as family and individual covariates.
49
The top
panel displays results for white students. Here we see a smaller spread in white students’ test
score percentile ranking between neighborhood quintiles in Cohort 1 (CDS-14 children),
indicating less inequality between the most and least disadvantaged neighborhoods over time
conditioning on family and individual factors. Moving from the fourth quintile neighborhood to
the fifth (most disadvantaged) decreases test score percentile rankings by ~4.5 for Cohort 0

49
All covariates (described earlier) are held at their means. Full results from all models available upon request.

149

(CDS-97 children), but only ~2.5 points for Cohort 1. White students exposed to the most
disadvantaged neighborhoods in Cohort 1 performed substantially better than white children
exposed to disadvantaged residential environments in earlier decades. Moreover, the gap
between the most and least disadvantaged neighborhoods narrows among white children in
Cohort 1 compared to Cohort 0 (~10 vs. 16.5 percentile points). One potential explanation is that
this indicates the influence of broader changes to the educational market, such as school choice
expansion, which may enable white students to reside in relatively lower-SES neighborhoods but
travel outside of the neighborhood to attend higher-SES non-local schools. Test score percentile
rankings for white children exposed, on average, to the two least disadvantaged neighborhood
quintiles were not significantly different between early and later cohorts. Percentile rankings for
white students in the least disadvantaged quintile (q1) were substantially higher relative to white
children in the most disadvantaged neighborhoods in 1997 (~69), and their percentile rank
remained very high in 2014 (~72), demonstrating the durable advantages transmitted to white
students exposed to higher-SES neighborhood contexts. Across all neighborhoods, white
students’ test score percentile rankings were higher for Cohort 1 relative to Cohort 0. The
predicted rankings for both cohorts of white students were substantially higher than test score
rankings for both cohorts of black students, reflecting the varied ways in which white students
continually manage to accrue advantages in an unequal opportunity structure.
A different story emerges for black students in 2014 relative to earlier decades. Although I do
not find significant differences between cohorts in terms of neighborhood effects on student
achievement, patterns suggest potentially widening inequality between black children exposed to
the most and least disadvantaged neighborhoods. Among black students (bottom panel), there is
a wider gap in test score percentile rank between neighborhood quintiles for students in Cohort 1.

150

Relative to blacks in Cohort 0, black students in Cohort 1 performed substantially better in more
advantaged neighborhoods, closing the test score gap with white students residing in the same
types of neighborhoods. While test score rankings for blacks in Cohort 1 were higher across all
quintiles relative to blacks in Cohort 0, their rankings were still lower whites (from both cohorts)
of all neighborhood quintiles. White students in Cohort 1 exposed to the most disadvantaged
residential contexts still had higher predicted test score rankings than black students exposed to
the least disadvantaged neighborhoods (62 vs. 61). While the gap in test score percentile rank
narrowed between black and white students in 2014, the persistent disparity in test scores
highlights the persistent structural constraints that black students face in their neighborhood and
school environments (e.g. unequal resources; racialized treatment; structural racism; choice
constraints; etc.) that are missing from these empirical models.
These preliminary results provide early evidence of changing neighborhood effects.
Moreover, the differences between race-stratified models suggest potential heterogeneity in
terms of changing neighborhood effects between racial/ethnic subgroups. For white students, the
effects of exposure to neighborhood disadvantage on student outcomes seems to be weakening
over time. For black students, the descriptive patterns suggest that while exposure to
neighborhood advantage may increase educational achievement more in recent years, the
negative impact of exposure to disadvantaged residential contexts remains durable and severe.
Subsequent analyses will continue to investigate how cumulative neighborhood disadvantage
affects student outcomes. After finalizing models examining academic achievement, I will
perform parallel models of academic persistence that investigate potential changing
neighborhood effects on on-time high school graduation.

151

Next Steps and Extensions

Can the weakening neighborhood-school relationship explain changes in neighborhood effects
on education over time?
This study takes the first step of identifying whether neighborhood effects on educational
outcomes have changed over time. Evidence of changes in effects over time could indicate that
there are other key neighborhood mechanisms mediating effects. The next step involves testing
factors associated with any change.
The neighborhood effects literature identifies a neighborhood’s institutional and social
contexts as two key overarching mechanisms that affect individual outcomes across various
measures of well-being (Galster 2012). Local schools are viewed as key neighborhood institutions
(Jencks and Mayer 1990) and mediators of neighborhood effects (Galster 2012; Sharkey and Faber
2014). Historically, the composition of a child’s neighborhood is correlated closely with the
composition of a child’s school. My dissertation, however, shows that this tight link between
where a child lives and enrolls in school has weakened and that neighborhoods and schools are
growing more dissimilar along demographic lines. My dissertation identified whether
neighborhood effects have changed over time and speculated that the weakened neighborhood-
school link may contribute to these changes.
Taking these findings as a starting point, and guided by the theoretical framework that views
schools as key neighborhood institutions that mediate neighborhood effects, I plan to test the
mechanisms for why neighborhood effects on student outcomes have changed over time and
explore how effects vary by race and gender. Specifically, I plan to perform mediation models that
explicitly explore whether this is because schools are increasingly not a neighborhood institution.
Evidence of changes in effects over time could indicate that there are other key neighborhood

152

mechanisms mediating effects or that neighborhoods and schools are independent sources of
inequality.
If schools and neighborhoods are becoming dissimilar in terms of their compositions, but
both are promoting inequality in some way, what are the implications for children’s educational
outcomes? I will rely on school contextual data provided by the CDS, as well as information on
school choice options, to test whether any changes in neighborhood effects over time is mediated
by the decoupled relationship between neighborhoods and neighborhood schools during this
period.
The genealogical sampling design of the CDS allows for clean cross-cohort comparison
among successive generational cohorts (i.e. children eligible for the CDS-14 sample are the
offspring of CDS-97 (and earlier) children). Future mediation analyses will investigate whether
neighborhood effects have changed because the neighborhood-school relationship has also
changed over time. The timing of these two cohorts neatly coincide with periods before and after
the expansion of school choice alternatives, and during a time when the neighborhood-school
relationship decoupled further (Candipan 2019), which I plan to exploit in mediation models.
Recent techniques, such as residual covariate balancing, extend the SNMM-RWR and MSM-
IPTW approaches to allow for the addition of mediators (Zhou and Wodtke 2019).

Extensions
Future analyses will also aim to examine whether particular neighborhood factors are more
or less influential on educational outcomes over time. Specifically, I plan to analyze how the
relative contribution of the neighborhood components used to generate neighborhood
disadvantage factor scores (e.g. neighborhood poverty) has changed over the three decades
comprising my study period. I also plan to examine the influence of other secular trends in

153

shaping educational outcomes over time. Specifically, I plan to decompose the effect of rising
educational attainment overall on student outcomes for children in my study.

Contributions
Overall, findings from this study will have implications for theories of neighborhood effects,
and for urban, housing, and educational policies. Importantly, this study addresses an empirical
hole in the neighborhood effect literature by analyzing changing effects over time, an overlooked
dimension of time in existing work. Additionally, this research contributes to our understanding
of the changing mechanisms of neighborhood inequality, and asks us to consider how
neighborhoods fit into a broader ecosystem that shapes inequality at multiple levels. Finally, this
research heeds calls to consider cumulative neighborhood effects and take spatial and temporal
dynamics seriously, and applies models that appropriately address confounding and collider
stratification bias in longitudinal settings.

Limitations of the Present Study
Certain data constraints limit the interpretation of current models. For current models
examining changing neighborhood effects on academic performance, I lack a lagged measure of
achievement for children in the CDS-14 cohort, which despite limitations (Morgan and Winship
2015), would provide stronger grounds for causal inference using observational data
(VanderWeele 2015). Models that control for lagged measures of achievement are called “value-
added” models (VAM) and are common in studies of academic achievement (Chetty et al. 2014;
Hanushek and Rivkin 2010).
Relatedly, while nearly all children in the CDS-97 cohort have multiple data points for test
scores (and thus have test score data in later ages/grades), the age at which CDS-14 children took

154

their tests ranges from 3-17 years. This means that I have much richer sequential measurements
of residential contexts and achievement for children in CDS-97 relative to CDS-14. Finally,
because only the oldest children in the CDS-14 sample are eligible for on-time high school
graduation by 2015, I have a relatively smaller subgroup of CDS-14 cohort members that I am
able observe in models examining changing neighborhood effects on academic persistence.

155

Tables and Figures

Table 4.1. Descriptive Characteristics of Analysis Sample
Overall (N=3515)
Cohort 0 (CDS-
97)
(n=2317)
Cohort 1 (CDS-14)
(n=1198)
N % N % N %
Race

White 1659 47.2 1143 49.3 516 43.1

Black 1856 52.8 1174 50.7 682 56.9
Sex

Male 1785 50.8 1181 50.4 604 50.4

Female 1730 49.2 1136 49.6 594 49.6
Neighborhood Disadvantage

Quintile 1

15.2

15.4

14.3

Quintile 2

13.7

14.1

12.1

Quintile 3

16.3

16.1

17.4

Quintile 4

20.0

19.9

20.1
Quintile 5 34.8 34.5 36.2
Notes: Analysis sample pertains to models examining neighborhood effects on student
achievement. This sample consists of non-Hispanic white and black children in CDS-97 and
CDS-14 cohorts that registered an assessment score on the Woodcock-Johnson Revised Letter
Writing (LW) test and for whom residential histories during childhood were observed.

156

Table 4.2. Inter-Cohort Comparison of Mean Achievement (by Neighborhood Disadvantage and
Race)
Overall Cohort 0 (CDS-97) Cohort 1 (CDS-14)
mean sd mean sd mean sd

Mean LW Test Percentile Rank

by Race

All

54.9 30.0 53.9 30.0 58.5 29.6

White

64.6 27.3 64.4 27.4 65.5 26.7

Black

45.8 29.5 43.5 28.8 53.2 30.5

by Neighborhood Disadvantage

Quintile 1 68.3 26.7 67.3 27.0 72.5 24.6

Quintile 2 63.7 27.1 63.4 27.1 65.2 27.0

Quintile 3 57.7 29.1 57.3 29.6 59.1 27.5

Quintile 4 51.4 30.0 49.5 29.7 58.6 29.7

Quintile 5 43.7 28.9 41.7 28.1 49.8 30.5

by Race and Neighborhood

White

% Neigh. Disadvantage

Quintile 1 71.0 24.9 70.4 25.1 73.8 23.8

Quintile 2 67.5 25.6 67.2 25.7 69.2 25.5

Quintile 3 62.9 27.6 63.2 28.3 61.7 25.3

Quintile 4 59.1 28.3 58.2 28.0 62.5 29.2

Quintile 5 49.6 28.4 48.5 28.7 52.5 27.7

Black

% Neigh. Disadvantage

Quintile 1 53.5 30.7 49.7 30.6 67.0 27.6

Quintile 2 53.3 28.1 51.6 27.9 58.4 28.4

Quintile 3 49.5 29.6 47.5 29.2 55.5 30.1

Quintile 4 46.5 30.0 43.8 29.5 56.2 29.8
Quintile 5 42.7 28.9 40.6 27.8 49.3 31.0
Notes: Mean achievement refers to percentile rank on the Woodcock-Johnson Revised Letter
Writing (LW) test, which is administered to children of all ages and grades. Percentile ranks are
age/grade-specific and range from 0 to 100.

157

Figure 4.1. Effects of Cumulative Neighborhood Disadvantage on Academic Achievement
Among White and Black Students (using Conventional OLS Regression)

158

Figure 4.2. Effects of Cumulative Neighborhood Disadvantage on Academic Achievement
Among White and Black Students (using Marginal Structural Models with Inverse Probability of
Treatment Weighting)

159

Figure 4.3. Effects of Cumulative Neighborhood Disadvantage on Academic Achievement
Among White and Black Students (using Regression with Residuals and a Structural Nested
Mean Model)

160

References

Almirall, D., Ten Have, T., & Murphy, S. A. 2010. “Structural nested mean models for assessing
time‐varying effect moderation.” Biometrics, 66(1), 131-139.
Almirall, D., McCaffrey, D. F., Ramchand, R., & Murphy, S. A. 2013. “Subgroups analysis
when treatment and moderators are time-varying.” Prevention Science, 14(2), 169-178.
Bischoff, Kendra, and Laura Tach. 2018. “The Racial Composition of Neighborhoods and Local
Schools: The Role of Diversity, Inequality, and School Choice.” City & Community
17(3):675-701.
Burdick-Will, J. 2017. “Neighbors but not classmates: neighborhood disadvantage, local violent
crime, and the heterogeneity of educational experiences in Chicago.” American journal of
education 124(1): 37-65.
Burdick-Will, J. 2018. “Neighborhood violence, peer effects, and academic achievement in
Chicago.” Sociology of education 91(3):205-223.
Burdick-Will, J., J. Ludwig, S. Raudenbush, R.J. Sampson, L. Sanbonmatsu, and P. Sharkey.
2011. “Converging evidence for neighborhood effects on children’s test scores: An
experimental, quasi-experimental, and observational comparison.” Pp. 255-276 in
Whither opportunity? Rising inequality, schools, and children's life chances, edited by
G.J. Duncan and R. Murnane. New York: Russell Sage.
Candipan, Jennifer. 2019. Neighbourhood change and the neighbourhood-school gap. Urban
Studies, 0042098018819075.
Charles, C. Z. 2003. The dynamics of racial residential segregation. Annual review of sociology,
29(1), 167-207.

161

Chetty, R., N. Hendren, P. Kline, and E. Saez. 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.
Clampet‐Lundquist, S., and D.S. Massey. 2008. “Neighborhood effects on economic self‐
sufficiency: A reconsideration of the moving to opportunity experiment.” American
Journal of Sociology, 114(1):107-143.
Cole, S. R., & Hernán, M. A. 2008. Constructing inverse probability weights for marginal
structural models. American journal of epidemiology, 168(6), 656-664.
Crowder, K., & South, S. J. 2011. Spatial and temporal dimensions of neighborhood effects on
high school graduation. Social science research, 40(1), 87-106.
Galster, George C. 2012. “The Mechanism(s) of Neighbourhood Effects: Theory, Evidence, and
Policy Implications.” Pp. 23–53 in Neighbourhood Effects Research: New Perspectives,
edited by M. Van Ham, D. Manley, N. Bailey, L. Simpson, and D. Maclennan.
Dordrecht, Neth.: Springer.
Harding, D. J. 2003. “Counterfactual models of neighborhood effects: the effect of neighborhood
poverty on dropping out and teenage pregnancy1.” American Journal of Sociology
109(3)676-719.
Harding, D.J., L. Gennetian, C. Winship, L. Sanbonmatsu, and J.R. Kling. 2011. “Unpacking
Neighborhood Influences on Education Outcomes: Setting the Stage for Future
Research.” Pp. 277-298 in Whither Opportunity? Rising inequality, schools, and
children's life chances, edited by G.J. Duncan and R. Murnane. New York: Russell Sage
Foundation.

162

Heckman, J. J. 2006. “Skill formation and the economics of investing in disadvantaged
children.” Science 312(5782): 1900-1902.
Hernán MA and Robins JM. 2019. Causal Inference. Boca Raton: Chapman & Hall/CRC,
forthcoming.
Hicks, A. L., Handcock, M. S., Sastry, N., & Pebley, A. R. 2018. “Sequential Neighborhood
Effects: The Effect of Long-Term Exposure to Concentrated Disadvantage on Children’s
Reading and Math Test Scores.” Demography, 55(1), 1-31.
Iceland, J., & Scopilliti, M. (2008). Immigrant residential segregation in US metropolitan areas,
1990–2000. Demography, 45(1), 79-94.
Jargowsky, P. A., & El Komi, M. 2011. Before or after the bell? School context and
neighborhood effects on student achievement. In H. B. Newburger, E. L. Birch, & S. M.
Wachter (Eds.), Neighborhood and life chances: How place matters in modern America
(pp. 50–72). Philadelphia: University of Pennsylvania Press.
Jencks C, Mayer SE. 1990. “The social consequences of growing up in a poor neighborhood,”
pp. 111–186 in Lynn LE Jr, McGeary MGH (eds.) Inner-City Poverty in the United
States. National Academy Press: Washington, DC.
Johnson, R. C., & Schoeni, R. F. 2011. “The influence of early-life events on human capital,
health status, and labor market outcomes over the life course.” The BE journal of
economic analysis & policy, 11(3).
Ladson-Billings, G. 2006. “From the achievement gap to the education debt: Understanding
achievement in US schools.” Educational researcher, 35(7), 3-12.

163

Mader, Nicole, Clara Hemphill and Qasim Abbas. 2018. "The Paradox of Choice: How School
Choice Divides New York City Elementary Schools." Center for NYC Affairs. Policy
Report.
Mare, R. D. 2011. “A multigenerational view of inequality.” Demography, 48(1), 1-23.
Massey, D. S., & Denton, N. A. 1993. American apartheid: Segregation and the making of the
underclass. Harvard University Press.
McCoy, D. C., Raver, C. C., & Sharkey, P. 2015. “Children’s cognitive performance and
selective attention following recent community violence.” Journal of health and social
behavior 56(1): 19-36.
McGrew, K. S., Werder, J. K., & Woodcock, R. W. 1991. Woodcock-Johnson [R]: Technical
Manual. DLM.
Owens, A., & Candipan, J. 2019. “Social and spatial inequalities of educational opportunity: A
portrait of schools serving high-and low-income neighbourhoods in US metropolitan
areas.” Urban Studies, 0042098018815049.
Reardon, Sean. 2011. “The Widening Academic Achievement Gap Between the Rich and the
Poor: New Evidence and Possible Explanations,” Pp. 91-116 in Duncan, G. J., &
Murnane, R. J. (eds.) Whither opportunity?: Rising inequality, schools, and children's life
chances. Russell Sage Foundation.
Reardon, Sean and Bischoff, Kendra. 2016. “The Continuing Increase in Income Segregation,
2007–2012.” Palo Alto, CA: Stanford Center for Education Policy Analysis.
Reardon, S. F., Kalogrides, D., & Shores, K. 2019. The geography of racial/ethnic test score
gaps. American Journal of Sociology, 124(4), 1164-1221.

164

Robins JM, Hernan MA, Brumback B. 2000. “Marginal structural models and causal inference in
epidemiology”. Epidemiology 11: 550–560.
Robins, J. M. (1994). Correcting for non-compliance in randomized trials using structural nested
mean models. Communications in Statistics-Theory and methods, 23(8), 2379-2412.
Robins James M., Hernan Miguel A., Rotnitzky Andrea. 2007. “Effect Modification by Time-
varying Covariates.” American Journal of Epidemiology 166:994–1002.
Sampson, R. J. 2008. “Moving to inequality: Neighborhood effects and experiments meet social
structure.” American journal of sociology 114(1), 189-231.
Sampson RJ, Morenoff JD, Gannon-Rowley T. 2002 “Assessing “neighborhood effects”: Social
processes and new directions in research.” Annual Review of Sociology 28:443–478.
Sampson, R. J., Raudenbush, S. W., & Earls, F. (1997). Neighborhoods and violent crime: A
multilevel study of collective efficacy. Science, 277(5328), 918-924.
Sampson, R. J., Sharkey, P., & Raudenbush, S. W. 2008. “Durable effects of concentrated
disadvantage on verbal ability among African-American children.” Proceedings of the
National Academy of Sciences, 105, 845–852.
Sastry, N., & Pebley, A. R. 2010. “Family and neighborhood sources of socioeconomic
inequality in children’s achievement.” Demography, 47(3):777-800.
Sharkey, P. 2010. “The acute effect of local homicides on children’s cognitive performance.”
Proceedings of the National Academy of Sciences, 107, 11733–11738.
Sharkey, P. 2013. Stuck in place: Urban neighborhoods and the end of progress toward racial
equality. University of Chicago Press.
Sharkey, P., & Elwert, F. 2011. “The legacy of disadvantage: Multigenerational neighborhood
effects on cognitive ability.” American journal of sociology 116(6): 1934-81.

165

Sharkey, P. T., Tirado-Strayer, N., Papachristos, A. V., & Raver, C. C. 2012. “The effect of local
violence on children’s attention and impulse control.” American Journal of Public
Health, 102, 2287–2293.
Sharkey Patrick, Faber Jacob W. 2014. “Where, When, Why, and for Whom Do Residential
Contexts Matter? Moving Away From the Dichotomous Understanding of Neighborhood
Effects.” Annual Review of Sociology 40:559–79.
Wheaton, B., & Clarke, P. 2003. “Space meets time: Integrating temporal and contextual
influences on mental health in early adulthood.” American Sociological Review, 680-706.
Williamson, T., & Ravani, P. 2017. “Marginal structural models in clinical research: when and
how to use them?” Nephrology Dialysis Transplantation 32(suppl_2): ii84-ii90.
Wilson, W.J. (1987) The Truly Disadvantaged: The Inner City, the Underclass and Public
Policy. Chicago: University of Chicago Press.
Wodtke, G. T. 2018. “Regression-based adjustment for time-varying confounders.” Sociological
Methods & Research, 0049124118769087.
Wodtke Geoffrey T., Almirall Daniel. 2017. “Estimating Moderated Causal Effects with Time-
varying Treatments and Time-varying Moderators: Structural Nested Mean Models and
Regression-with-residuals.” Sociological Methodology 47:212–45.
Wodtke Geoffrey T., Elwert Felix, Harding David J. 2016. “Neighborhood Effect Heterogeneity
by Family Income and Developmental Period.” American Journal of Sociology
121:1168–222.
Wodtke, G. T., Harding, D. J., & Elwert, F. 2011. Neighborhood effects in temporal perspective:
The impact of long-term exposure to concentrated disadvantage on high school
graduation. American Sociological Review, 76(5), 713-736.

166

Zhou, X., & Wodtke, G. T. 2019. Residual Balancing: A Method of Constructing Weights for
Marginal Structural Models.

167

Chapter 5
Conclusion

Overall Findings and Broad Implications
This dissertation examined the relationship between neighborhoods and schools in terms of
their racial/ethnic and socioeconomic composition; contextual effects on student outcomes; and
how changes in both contexts contribute to inequality. Neighborhoods and schools are linked—
historically, under a strict residence-based school assignment system, where one lived determined
where one attended school. The neighborhood-school link weakened as both policy changes
expanded school choice and largescale urban changes produced gentrification. I examined the
implications of the changing neighborhood-school link in three ways: (1) mismatches in
neighborhood and school composition; (2) families’ residential and school enrollment choices; and
(3) neighborhood effects on children’s educational outcomes.
This project examined links between neighborhood and school composition over time, how
neighborhood and school composition matter for children’s educational outcomes, and how
changes in both contexts contribute to inequality, with a particular focus on neighborhoods
undergoing economic change. A focus on neighborhoods experiencing economic change (e.g.
gentrifying and ascendant neighborhoods), provides insights into how institutions, such as
schools, change as neighborhoods diversify. If neighborhoods and schools are so closely linked,
why do public schools remain segregated, even as neighborhoods around them become more
diverse? Observing patterns and trends in neighborhoods that are diversifying along racial/ethnic
and socioeconomic lines sheds light on this question. Results from this dissertation suggest that

168

schools will not necessarily change alongside neighborhoods, and that policies, such as school
choice, may not provide equal educational access and opportunities to less advantaged students
unless they take into account the segregating choices of white and upper-SES families.
The final empirical chapter of this dissertation began investigating the consequences of
neighborhood exposure for children’s educational outcomes and presented a plan for subsequent
analyses to explore whether changes in effects over time can be explained by the weakening link
between neighborhoods and neighborhood schools. Findings have both theoretical and
methodological implications for future neighborhood effects research.
Taken together, the analyses encompassing this dissertation take a step toward understanding
trends, mechanisms, and effects of the changing neighborhood-school link, and how this
relationship plays out in a context of urban and neighborhood change. Future work will build off
these analyses. Although this project focuses largely on gentrifying neighborhoods, this project
aims more broadly to understand the process of racial transition and neighborhood change, and
how race and class-based residential and school sorting upholds longstanding systems of
inequality by way of neighborhood and school stratification.
My overall dissertation project contributes to a small, but emergent literature examining
trends, causes and consequences of the changing relationship between neighborhoods and
schools over time. In doing so, this dissertation brings together the fields of urban sociology, the
sociology of education, and social stratification. This dissertation also represents one of first
quantitative studies exploring the relationship between gentrification and schools, and one of few
national-level quantitative studies exploring the relationship between residential and school
enrollment outcomes. Additionally, this project examines neighborhoods and schools jointly and

169

takes spatial and temporal dynamics seriously, which few studies do, investigating how changes
in one context affect changes in the other.
Broadly, findings from my dissertation contribute to our understanding of the mechanisms
upholding neighborhood and school segregation, the consequences of educational policy and
urban changes for neighborhoods and schools, and the processes of residential and school choice,
with implications for policies in the housing, urban, education, and economic development
arenas.

Abstract (if available)

Abstract This dissertation examines the changing relationship between neighborhoods and schools in the United States in terms of their racial/ethnic composition

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Redlining, neighborhood change, and individual outcomes: an exploration of how space shapes the landscape of inequality from past to present

PDF

Racial formation and gentrification in Los Angeles

PDF

Investigating the association of historical preservation and neighborhood status in Detroit, 1970-2015

PDF

Navigating the struggle: neoliberal meritocracy, Latinx charter school graduates, and the college transition

PDF

Changing workplace and inequality in work-family policies

PDF

Educating hearts and minds through high-quality service learning curriculum in U.S. high schools

PDF

Levels of interest: the effects of teachers' unions on Congress, statehouses, and schools

PDF

Racial disparities in U.S. K-12 suburban schools: an evaluation study

PDF

Equity and inclusion of veteran students with traumatic brain injury in higher education classrooms: a gap analysis

PDF

Code 'war' theorizing: information and communication technology's impact on international relations theorizing, negotiation, and cyber relations

PDF

Household mobility and neighborhood impacts

PDF

Lack of alumni giving in an international school in Asia

PDF

Using the UNESCO Historic Urban Landscape framework: a case study of the Pico-Union neighborhood

PDF

(Re)Imagining STEM instruction: an examination of culturally relevant andragogical practices to eradicate STEM inequities among racially minoritized students in community colleges

PDF

Location choice and the costs of climate change

PDF

Inequality regimes in sexual labor

PDF

Maintaining alignment between strategic and tactical goals in an innovative healthcare organization

PDF

School resource allocation in times of economic boom and bust

PDF

Assessment and teaching improvement in higher education: investigating an unproven link

PDF

Reshaping Los Angeles: housing affordability and neighborhood change

Asset Metadata

Creator Candipan, Jennifer (author)

Core Title Spatial dimensions of stratification: neighborhood change, urban inequality, and the neighborhood-school link in the U.S.

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Sociology

Publication Date 07/23/2021

Defense Date 05/23/2019

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

Tag gentrification,Inequality,neighborhood change,neighborhood effects,neighborhoods and schools,OAI-PMH Harvest,policy,quantitative analysis,school choice,sociology of education,stratification,Urban sociology

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Owens, Ann (committee chair), Ailshire, Jennifer (committee member), Painter, Gary (committee member), Pastor, Manuel (committee member)

Creator Email candipan@usc.edu,jennifer.candipan@alumni.usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-187741

Unique identifier UC11662914

Identifier etd-CandipanJe-7595.pdf (filename),usctheses-c89-187741 (legacy record id)

Legacy Identifier etd-CandipanJe-7595.pdf

Dmrecord 187741

Document Type Dissertation

Format application/pdf (imt)

Rights Candipan, Jennifer

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