Long-Term Neighborhood Effects

on Low-Income Families: Evidence

from Moving to Opportunity
The Harvard community has made this
article openly available. Please share how
this access benefits you. Your story matters

Citation Ludwig, Jens, Greg J. Duncan, Lisa A. Gennetian, Lawrence F.

Katz, Ronald C. Kessler, Jeffrey R. Kling, and Lisa Sanbonmatsu.
2013. “Long-Term Neighborhood Effects on Low-Income Families:
Evidence from Moving to Opportunity.” American Economic Review
103, no. 3: 226–231.

Published Version doi:10.1257/aer.103.3.226

Citable link http://nrs.harvard.edu/urn-3:HUL.InstRepos:12553723

Terms of Use This article was downloaded from Harvard University’s DASH
repository, and is made available under the terms and conditions
applicable to Open Access Policy Articles, as set forth at http://
nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-
use#OAP
 Long-Term Neighborhood Effects on Low-Income Families: Evidence
from Moving to Opportunity
By JENS LUDWIG, GREG J. DUNCAN, LISA A. GENNETIAN, LAWRENCE F. KATZ, RONALD C.
KESSLER, JEFFREY R. KLING, AND LISA SANBONMATSU*

* Ludwig: University of Chicago, 1155 East 60th Street, concern that neighborhood environments may
Chicago, IL 60637 and National Bureau of Economic Research
(NBER), jludwig@uchicago.edu; Duncan: University of California, exert independent causal effects on people’s
Irvine, School of Education, 2056 Education Building, Irvine, CA
92697, gduncan@uci.edu; Gennetian: New York University, Institute
of Human Development and Social Change, 246 Greene Street, Floor
long-term life chances. Living in a
6E, New York, NY 10003 and NBER, gennetl@nber.org; Katz:
Harvard University, Department of Economics, Cambridge, MA disadvantaged social environment may
02138 and NBER, lkatz@harvard.edu; Kessler: Harvard Medical
School, Department of Health Care Policy, 180 Longwood Avenue, depress life outcomes by, for example,
Boston MA 02115, kessler@hcp.med.harvard.edu; Kling:
Congressional Budget Office, 2nd and D Streets, SW, Washington DC shaping exposure to peer norms or access to
20515, and NBER, jeffrey_kling@nber.org; Sanbonmatsu: NBER,
1050 Massachusetts Avenue, Cambridge, MA,
lsanbonm@nber.org. Support for this research was provided by a
02138, resources such as schools or job referrals.
contract from the U.S. Department of Housing and Urban
Development (HUD; C-CHI-00808) and grants from the National However some theories yield the opposite
Science Foundation (SES-0527615), National Institute for Child
Health and Human Development (R01-HD040404, R01-HD040444), prediction about the effects of moving into a
Centers for Disease Control (R49-CE000906), National Institute of
Mental Health (R01-MH077026), National Institute for Aging (P30- more affluent area, since more affluent areas
AG012810, R01-AG031259, and P01-AG005842-22S1), National
Opinion Research Center’s Population Research Center (through
R24-HD051152-04 from the National Institute of Child Health and
could have greater discrimination and
Human Development), University of Chicago’s Center for Health
Administration Studies, U.S. Department of Education/Institute of competition from advantaged peers and fewer
Education Sciences (R305U070006), Bill & Melinda Gates
Foundation, John D. and Catherine T. MacArthur Foundation, Russell social services for the poor.
Sage Foundation, Smith Richardson Foundation, Spencer
Foundation, Annie E. Casey, and Robert Wood Johnson Foundation. Isolating the causal effects of neighborhood
Outstanding assistance with the data and analysis was provided by
Joe Amick, Ryan Gillette, Ray Yun Gou, Ijun Lai, Jordan Marvakov,
Nicholas Potter, Matt Sciandra, Fanghua Yang, Sabrina Yusuf, and
environments on behavior and well-being is
Michael Zabek. The survey data collection effort was led by Nancy
Gebler of the University of Michigan’s Survey Research Center under complicated by the fact that most people have
subcontract to our research team. We thank Janet Currie and many
seminar participants for helpful comments. MTO data were provided at least some degree of choice over where they
by HUD. The data used in this paper are available through the
Interuniversity Consortium on Political and Social Research (ICPSR) live. Observational studies may confound
at the University of Michigan. The views expressed in this work are
those of the authors and should not be interpreted as those of the
Congressional Budget Office or HUD.
neighborhood influences with those of hard-
to-measure individual- or family-level
Research dating back to at least the 17th
attributes that affect both residential sorting
century has shown that people living in more
and the behavioral outcomes of interest.
disadvantaged neighborhoods fare worse with
Evidence about “neighborhood effects” is
respect to earnings, education, health, crime
important in part because neighborhood
involvement and other life outcomes
residential segregation by income has been
(Sampson 2012). These patterns have led to
increasing in the United States since 1970 several key adult mental and physical health
beyond the amount expected from rising outcomes, but have no consistent detectable
income inequality alone (Reardon and impacts on adult economic self-sufficiency or
Bischoff 2011). Nearly 9 million Americans children’s educational achievement outcomes,
live in “extreme-poverty” neighborhoods in even for children who were pre-school age at
which at least 40 percent of residents are poor baseline. We also find signs of the same
(Kneebone, Nadeau, and Berube 2011). gender difference in the effects of MTO
Knowledge of neighborhood effects (and the moves on youth risky behaviors and health
mechanisms behind such effects) is relevant found in the interim (4-7 year) follow-up, with
for evaluating policies that affect how people girls doing better in some ways while boys do
are sorted across neighborhoods and for worse. Despite the mixed MTO impacts on the
assessing housing market efficiency. standard outcomes that have dominated the
This paper examines the long-term effects neighborhood-effects literature, MTO moves
on low-income parents and children of generate a large gain in subjective well-being
moving from very disadvantaged to less (SWB) for adults (Ludwig et al. 2012).
distressed neighborhoods, using data from a
I. The Moving to Opportunity Experiment
unique, large-scale randomized social
experiment – the U.S. Department of Housing From 1994 to 1998 MTO enrolled 4,604
and Urban Development’s (HUD’s) Moving low-income public housing families living in
to Opportunity (MTO) demonstration. Via high-poverty neighborhoods within five U.S.
random lottery, MTO offered housing cities: Baltimore, Boston, Chicago, Los
vouchers to families with children living in Angeles, and New York. Families were
high-poverty public housing projects that randomized into three groups: i) the
facilitate moves to less-distressed areas. MTO Experimental group, which received housing
randomization generates large, persistent vouchers that subsidize private-market rents
differences in neighborhood conditions for and could only be used in census tracts with
otherwise comparable groups and enables us 1990 poverty rates below 10 percent, and
to attribute group differences in post-baseline additional housing-mobility counseling; ii) the
outcomes to the offer to move through MTO. Section 8 group, which received regular
We find that 10-15 years after housing vouchers without any MTO relocation
randomization, MTO-assisted moves improve constraint; and iii) a control group, which
received no assistance through MTO. Some were interviewed slightly later than other
48% of households assigned to the adults because funding for this activity was
Experimental group and 63% of those secured later during the project; we discuss
assigned to the Section 8 group moved implications of this delay below.
through MTO (the MTO “compliance rate”). To measure neighborhood conditions we
Data from baseline surveys show that these collected self-report address information and
families were quite economically passive tracking data, which we linked to
disadvantaged when they applied for MTO census tract-level data from the 1990 and 2000
(see Appendix Table 1). Most household censuses and the 2005-09 American
heads were African-American or Hispanic Community Surveys. We focus on duration-
females; fewer than 40% had completed high weighted average tract characteristics over the
school. Around three-quarters of applicants 10-15 year study period, since people’s life
reported getting away from gangs and drugs as outcomes may depend on cumulative exposure
the most important reason for enrolling in to neighborhood environments. Our surveys
MTO. As one would expect from a properly- also asked MTO adults and youth to self-
conducted random assignment, the distribution report about their neighborhood conditions.
of baseline characteristics is balanced between Our primary focus is on indices of adult
the treatment and control groups. outcomes in the domains of economic
outcomes, physical health, and mental health,
II. Measures and Methods
and youth outcomes in the domains of

To measure long-term outcomes, our education, physical health, mental health, and

research team subcontracted with the Institute risky behavior. The outcome indices are

for Social Research at the University of constructed from a set of individual outcomes

Michigan to collect in-person data with 3,273 from our surveys that are rescaled so that

MTO adults and 5,105 youth who were ages higher values represent “better” outcomes and

10-20 at the end of 2007. Data were collected then converted to Z-scores using the control

between 2008 and 2010, or 10-15 years after group distribution. Aggregating outcomes

baseline. The effective response rates equaled improves statistical power to detect impacts

90% for MTO adults and 89% for youth, and and reduces the risk of “false positives” by

were generally similar across randomized reducing the number of statistical tests carried

MTO groups. Adults in the Section 8 group out. To further reduce the risk of false
positives due to data mining, the outcome
III. Results
indices we examine were pre-specified for the
interim MTO follow-up done in 2002 (Kling, One year after baseline, the average control
Liebman, and Katz 2007). group adult was living in a neighborhood with
We present intention-to-treat (ITT) an average tract poverty rate of 50 percent
estimates that capture the effect of being (Appendix Table 2). Moving with an
offered the chance to use an MTO voucher to Experimental voucher reduced average tract
move into a different neighborhood. These poverty rates one year after baseline by 35
estimates are calculated as the difference in percentage points (2.8 standard deviations in
average outcomes for families assigned to the 2000 census tract poverty distribution),
treatment versus the control condition, by while moving through MTO with a regular
regressing an outcome index against indicators Section 8 voucher reduced tract poverty rates
for treatment-group assignment and (pre- by 21 percentage points (1.8 standard
random assignment) baseline covariates that deviations). These differences across MTO
include indicators for MTO demonstration site groups in neighborhood conditions narrowed
and participant socio-demographic over time, mostly because the neighborhood
characteristics to improve precision (see poverty rates for controls declined.
Appendix Table 1). The estimates are Despite the convergence of neighborhood
weighted to account for changes over time in conditions across MTO groups over the study
the probability of treatment assignment due to period, MTO-induced differences in duration-
higher-than-expected compliance rates. weighted average tract poverty rates over the
We also present estimates of the effects of course of the 10-15 year follow-up period
treatment on the treated (TOT), which use were quite sizable. Figure 1 shows that a large
random assignment indicators as instruments share of adults who moved with an MTO
for moving through MTO in the Experimental Experimental voucher had an average tract
or Section 8 groups and assume the treatment poverty rate below 20%, which was true for
assignment only affects families who move few control group families. The effects of
using a MTO voucher. moving with a regular Section 8 voucher on
average tract poverty rates were somewhat
less pronounced. (Appendix Table 2 presents
MTO impacts on a broader set of of height in meters) by 7 percentage points.
neighborhood characteristics.) This was a decline of nearly 40% of the
control group mean of 18 percent (Ludwig et
[Insert Figure 1 Here]
al. 2011). For a five-foot-four woman, a BMI

Contrary to the widespread view that living of 40 would correspond to a weight of about

in a disadvantaged inner-city neighborhood 235 pounds. We also found the Experimental-

depresses labor market outcomes, Table 1 voucher TOT effect reduced the prevalence of

shows that being offered a voucher through diabetes, measured from blood samples and

MTO did not improve economic self- defined as having a level of glycosylated

sufficiency, at least for this study sample. hemoglobin (HbA1c) 6.5%, by 10 percentage

Although the ITT estimate for the Section 8 points, or one-half of the control group’s rate.

group was negative and marginally significant

[Insert Table 1 Here]
(p<.10), we believe this was most likely an
artifact of our interviewing the Section 8 We found no evidence that MTO had
group adults a bit later than control adults, beneficial impacts on youth educational
when labor market conditions were less outcomes. Effects on math and reading test
favorable (see Sanbonmatsu et al. 2011). scores were very close to zero both for youth
The results in Table 1 also hint at some who were pre-school age at baseline and for
potentially positive impacts of MTO on adult youth who were ages 6 and up at baseline.
mental and physical health outcomes, with MTO did tend to have some beneficial effects
ITT effects on these broad health outcome on female but not male youth in other
indices that were in the direction of better outcome domains (Table 2). Assignment to
health but not quite statistically significant. the Experimental and Section 8 groups
However some specific individual health improved physical health for girls, while the
outcomes showed large and statistically Experimental group effect on mental health
significant improvements in response to outcomes is also positive and statistically
MTO-assisted moves. For example, moving significant for girls. The estimated effects on
with an Experimental-group voucher (the TOT health outcomes for boys range from zero to
effect) reduced the prevalence of having a negative (worse health). We can reject the null
body mass index of 40 or more (BMI, defined hypothesis that the physical and mental health
as weight in kilograms divided by the square
impacts of the Experimental treatment are the other hand much of the scientific and policy
same by gender (Appendix Table 3). concern about “neighborhood effects” is
precisely with families living in the most
[Insert Table 2 Here]
distressed areas. And previous observational

IV. Discussion studies report finding impacts on samples

similar to the MTO sample.
The MTO long-term results did not provide Looking at broad indices of outcomes that
support for the view that high rates of school were pre-specified for the interim MTO data,
failure and non-employment in central city we see suggestive (but not always statistically
neighborhoods are due to the direct adverse significant) signs that physical and mental
effects of living in a poor neighborhood. The health outcomes improved for adult women
pattern of findings was consistent with the and female youth. We see very large MTO
results from the 4-7 year interim follow-up of impacts on specific health measures,
MTO adults and youth (Kling, Liebman, and particularly those related to extreme obesity
Katz 2007). Our long-term data also showed and diabetes. Although we acknowledge that
no detectable impacts on academic measuring candidate mechanisms like diet,
achievement for children of pre-school age at exercise and access to health care is
baseline even though MTO led to very large intrinsically challenging, and that our
changes in their neighborhood conditions at a available data on these factors are quite
life stage when they may be most limited, it is noteworthy that MTO moves
developmentally malleable. reduced extreme obesity and diabetes by fully
One obvious question involves 40-50% for adults while generating almost no
generalizability: Do neighborhood changes detectable changes in our measures of these
have no impact on earnings or educational candidate mediators. One hypothesis for why
achievement outcomes here because the MTO MTO improved physical health is because of
study sample is somehow unusual? MTO MTO’s beneficial impacts on neighborhood
families were drawn from extremely safety, and subsequent gains in mental health
distressed communities. The baseline census – including measures of psychological
tracts for MTO families were fully 3 standard distress. This safety-stress-health hypothesis is
deviations above the national average in the also consistent with our finding that the
2000 census tract-poverty distribution. On the majority of MTO households signed up for
MTO because of concerns about crime and impacts on the overall quality of life and well-
violence. being of low-income families despite the
The long-term MTO data did not show any mixed pattern of impacts on traditional
signs of the large drop in violent-crime arrests “objective” outcome measures, including null
that were found in the 4-7 year MTO follow- effects on earnings and education. Ludwig et
up among both male and female youth (Kling, al. (2012) show that a 1 standard deviation
Ludwig, and Katz 2005). However the long- decline in census tract poverty rates (about 13
term data did echo the interim data to some percentage points) is associated with an
extent in showing female youth may benefit increase in SWB that is about the same size as
from MTO moves in other outcome domains the difference in SWB between households
like mental health or risky behaviors, but male whose annual incomes differ by $13,000 – a
youth tended to do no better (or do worse) as a very large amount given that the average
result of such moves. The reason for these control group family’s annual income in the
gender differences remains unclear; they do long-term survey is just $20,000.
not seem to be due merely to gender
REFERENCES
differences in the prevalence of these
outcomes or behaviors.
Kling, Jeffrey R., Jeffrey B. Liebman, and
The sizes of these gender differences in Lawrence F. Katz. 2007. “Experimental
Analysis of Neighborhood Effects.”
MTO impacts were smaller in the long-term
Econometrica 75 (1): 83–119.
than interim data, just as the difference across
Kling, Jeffrey R., Jens Ludwig, and Lawrence
MTO groups in neighborhood conditions was
F. Katz. 2005. “Neighborhood Effects on
smaller at the time of the long-term surveys Crime for Female and Male Youth:
Evidence from a Randomized Housing
than interim surveys. These patterns suggest
Voucher Experiment.” Quarterly Journal
youth outcomes may be more affected by of Economics 120 (1): 87–130.
contemporaneous neighborhood conditions
Kneebone, Elizabeth, Carey Nadeau, and Alan
than accumulated exposure to neighborhood Berube. 2011. “The Re-Emergence of
Concentrated Poverty: Metropolitan
environments, or what Sampson (2012) calls
Trends in the 2000s”. Washington, DC:
“situational” neighborhood effects as opposed The Brookings Institution, Metropolitan
Policy Program.
to “developmental” neighborhood effects.
The MTO data make clear that Ludwig, Jens, Greg J. Duncan, Lisa A.
Gennetian, Lawrence F. Katz, Ronald C.
neighborhood environments have important
Kessler, Jeffrey R. Kling, and Lisa
 Sanbonmatsu. 2012. “Neighborhood
Effects on the Long-Term Well-Being of
Low-Income Adults.” Science 337
(6101): 1505–1510.

Ludwig, Jens, Lisa Sanbonmatsu, Lisa

Gennetian, Emma Adam, Greg J.
Duncan, Lawrence F. Katz, Ronald C.
Kessler, Jeffrey R. Kling, Stacy Tessler
Lindau, Robert C. Whitaker, et al. 2011.
“Neighborhoods, Obesity, and Diabetes-a
Randomized Social Experiment.” The
New England Journal of Medicine 365
(16): 1509–19.

Reardon, Sean F., and Kendra Bischoff. 2011.

“Income Inequality and Income
Segregation.” American Journal of
Sociology 116 (4): 1092–1153.

Sampson, Robert J. 2012. Great American

City: Chicago and the Enduring
Neighborhood Effect. Chicago:
University of Chicago Press.

Sanbonmatsu, Lisa, Jens Ludwig, Lawrence F.

Katz, Lisa A. Gennetian, Greg J. Duncan,
Ronald C. Kessler, Emma Adam,
Thomas W. McDade, and Stacy Tessler
Lindau. 2011. Moving to Opportunity for
Fair Housing Demonstration Program:
Final Impacts Evaluation. Washington,
DC: U.S. Department of Housing and
Urban Development, Office of Policy
Development and Research.
 .05
Density
.025
0

0 20 40 60 80 100
Average Census Tract Poverty Rate
from Random Assignment through May 2008

Experimental Group Compliers

Section 8 Group Compliers
Control Group

FIGURE 1. D ENSITIES OF A VERAGE P OVERTY R ATE BY TREATMENT G ROUP

Notes: Duration-weighted average of census tract poverty at all addresses from random assignment through May 2008 (just prior to the long-term
survey fielding period), based on linear interpolation of 1990 and 2000 decennial census and the 2005-09 American Community Survey data.
Density estimates used an Epanechnikov kernel with a half-width of 2.
Source and Sample: The sample is all adults who were interviewed as part of the long-term survey (with Experimental and Section 8 group adults
limited to those who used an MTO voucher to move). Sample sizes in the Experimental, Section 8, and control groups are 711, 413, and 1,139.
 TABLE 1 — MTO I MPACTS ON A DULT O UTCOMES

Experimental vs. Section 8 vs.

Control Control
Panel A. Outcome Indices (z-scores)
Index for all outcomes 0.037 -0.010
(0.040) (0.059)
Economic self-sufficiency -0.029 -0.112*
(0.040) (0.059)
Absence of physical health problems 0.055 0.062
(0.042) (0.058)
Absence of mental health problems 0.069 0.063
(0.042) (0.062)
Panel B. Selected individual health outcomes
Psychological distress, K6 z-score -0.106** -0.081
(0.042) (0.060)
BMI 40 -0.036** -0.038*
(0.016) (0.023)
Blood test detected diabetes (HbA1c 6.5%) -0.050*** -0.015
(0.018) (0.026)

Notes: Estimates are the intent-to-treat effect sizes from an ordinary least squares
regression of each outcome on treatment indicators and the baseline covariates listed
in Appendix Table 1. Robust standard errors are in parentheses. Outcome indices and
psychological distress are z-scores using the mean and standard deviation for the
control group. Index components are as follows (positive outcomes (+) were included
as is, while the signs for negative outcomes (−) were reversed so that higher index
values indicate “better” outcomes): Economic self-sufficiency: + adult employed and
not on TANF + employed + 2009 earnings − on TANF − 2009 government income.
Mental health: − distress index − depression − Generalized Anxiety Disorder +
calmness + sleep. Physical health: − self-reported health fair/poor − asthma attack
past year – obesity − hypertension − trouble carrying/climbing. The index for all
outcomes includes the 15 measures in the self-sufficiency, physical health, and
mental health indices. Psychological distress consists of 6 items (sadness,
nervousness, restless, hopelessness, feeling that everything is an effort,
worthlessness) scaled on a score from 0 (no distress) to 24 (highest distress). Body
mass index (BMI) is weight in kilograms divided by height in meters squared (BMI
>= 40 indicates extreme obesity). Glycosylated hemoglobin (HbA1c) level is from a
blood sample, and a level >= 6.5% indicates diabetes.
Source and Sample: The sample is all adults who were interviewed as part of the
long-term survey. Sample sizes in the Experimental, Section 8, and Control groups
are 1,456, 678, and 1,139.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
 TABLE 2 — MTO I MPACTS ON Y OUTH O UTCOMES

Experimental vs. Section 8 vs. Experimental vs. Section 8 vs.

Control Control Control Control

Panel A. Outcome Indices (z-scores)

Female Youth Male Youth
Index for all outcomes 0.079 0.077 -0.016 -0.116*
(0.062) (0.065) (0.062) (0.069)
Absence of physical health problems 0.109* 0.124* -0.075 -0.058
(0.061) (0.065) (0.068) (0.078)
Absence of mental health problems 0.160*** 0.039 0.008 -0.062
(0.058) (0.065) (0.064) (0.071)
Absence of risky behavior -0.001 0.007 0.027 -0.069
(0.065) (0.066) (0.061) (0.067)
Education -0.043 0.027 -0.006 -0.082
(0.061) (0.072) (0.061) (0.069)
Panel B. Selected education outcomes
by age group (z-scores)
Under Age 6 Ages 6 and Over
Combined math/reading assessment -0.014 0.019 -0.018 0.043
(0.055) (0.056) (0.061) (0.072)

Notes: Estimates are the intent-to-treat effect sizes from an ordinary least squares regression of each outcome on
treatment indicators and the baseline covariates listed in Appendix Table 1 (the analyses also control for a series of
youth-specific covariates not listed in Appendix Table 1). Robust standard errors adjusted for household clustering
are in parentheses. All measures are z-scores using the mean and standard deviation for the control group. Index
components are as follows (positive outcomes (+) were included as is, while the signs for negative outcomes (−)
were reversed so that higher index values indicate “better” outcomes): Physical health: − self-reported health
fair/poor − asthma attack past year − overweight – non-sports injury past year. Mental health: − distress index −
depression − Generalized Anxiety Disorder. Risky behavior: − marijuana past 30 days − smoking past 30 days −
alcohol past 30 days − ever pregnant or gotten someone pregnant. Education: + graduated high school or still in
school + in school or working + Early Childhood Longitudinal Study-Kindergarten cohort study (ECLS-K)
reading score + ECLS-K math score. The index for all outcomes includes the 15 measures in the physical health,
mental health, risky behavior, and education indices. Combined math/reading assessment scores are the average of
the reading and math scores from ECLS-K assessments adapted for the MTO study.
Source and Sample: The sample in both panels is youth who were interviewed as part of the long-term survey.
Panel A is youth ages 15-20 as of December 2007, and Panel B is youth ages 13-20 as of the same date (in analysis
not shown, effects for youth ages 10-12 were similar to those for youth ages13-20). Sample sizes in the
Experimental, Section 8, and Control groups are 1,437, 1,031, and 1,153.
*** Significant at the 1 percent level.
* Significant at the 10 percent level.