The effect of a program-based housing move on employment: HOPE VI in Atlanta.
Anil, Bulent ; Sjoquist, David L. ; Wallace, Sally 等
1. Introduction
In 2000, there were approximately 4.9 million U.S. households
living in U.S. Department of Housing and Urban Development
(HUD)-subsidized housing. (1) In an effort to revitalize distressed
public housing and reduce the concentration of poverty, the federal
government enacted the HOPE VI program in 1992. The HOPE VI program
provides grants to local housing authorities in order to close existing
public housing projects and build mixed-income housing on the site. From
1993 through 2006, HUD reports funding grants totaling $6.2 billion for
implementation, demolition, and planning. (2)
HOPE VI, by design, often creates an exogenous shock to households
living in public housing by requiring households to move from their
current public housing unit to a housing unit at an alternative site.
(3) Under the HOPE VI program, displaced families may use a housing
voucher or move to a vacant public housing unit. The program creates a
natural experiment for analyzing the impact of an exogenous move from a
public housing unit on outcomes of those individuals who experience such
a move. The focus of our research is on the effect of program-based
moves ("HOPE VI moves") on employment. Knowing the effects of
HOPE VI is important for determining whether to continue the program, to
modify it, or to abandon it. We provide evidence on one possible
effect--employment.
To analyze the impact of program-based moves on employment, we
utilize a unique microlevel data set that includes the population of
Atlanta Housing Authority (AHA) clients from 1995 to 2003 matched with
Georgia Department of Labor (DOL) unemployment insurance (ES202)
records. The resulting data set is an administrative record that
contains both housing and employment status for all individuals in the
city of Atlanta who received housing assistance from AHA during the
period. To these data, we add a set of variables that reflect
neighborhood characteristics. We use these data to analyze the impact of
a "program-based move" out of public housing caused by a HOPE
VI or HOPE VI-like project on the employment of the affected AHA
clients. We compare the probability of employment of those who
experienced such a program-based move to those public housing tenants
who did not experience a program-based move. We find that a
program-based move caused a positive increase in the probability of
employment as compared to other public housing residents.
We also use these data to explore how employment status changes
after a voluntary move from a public housing project and find that the
increase in employment after a voluntary move is essentially the same as
the increase in employment associated with a program-based move when
compared to public housing residents that do not move.
The article proceeds as follows. In the next section, we review the
relevant literature on the effects of moves from public housing, which
is followed by a discussion of the data and methodology. Section 4
presents the empirical results, and summary statements conclude.
2. Voluntary and Involuntary Moves and Their Impacts on Employment
Public housing might affect the behavior of residents either
directly because of the housing subsidy, through the effect of
neighborhood characteristics, or through the availability of various
services such as job training. Federal housing programs have a variety
of documented impacts on recipient behaviors, including migration,
family size and formation, labor supply, and educational attainment
(see, e.g., Corcoran and Heflin 2003). (4) We focus here on those
studies most relevant to our research question--what is the impact of
involuntary versus voluntary moves from public housing on the
probability of employment; we do not consider the large and important
literature that analyzes the impact of a change in the housing subsidy
formula.
Impacts of Voluntary Relocation. Gautreaux and Moving to
Opportunity
There have been two major programs that have encouraged voluntary
relocation of public housing residents: the Gautreaux Program in Chicago
(which ended in 1998) and the Moving to Opportunity program (MTO), which
was run in five demonstration cities beginning in 1994. Johnson, Ladd,
and Ludwig (2002) provide a review of the research on the effects of the
Gautreaux and MTO programs on public housing tenants and conclude that
relocation of low-income residents reduces welfare dependency, increases
educational attainment, and improves the health status of its residents.
DeLuca and Rosenbaum (2003) study the long-run effects of the Gautreaux
Program on participants and find permanent positive labor market impacts
on participants. This result is consistent with the findings of
Rosenbaum (1995).
An interim evaluation report by Abt Associates and the National
Bureau of Economic Research for the U.S. Department of Housing and Urban
Development (HUD 2006) was done to assess MTO's effectiveness in
six "domains" related to education, employment, income, risky
behavior, mobility, and physical and mental health. Among the findings
of the interim report that are relevant to this study are the
significant amount of mobility and improved neighborhood characteristics
(lower poverty, which is a design element of MTO, and higher employment
rates) and an insignificant impact on employment. Rosenbaum and Harris
(2001) find that MTO participants experienced an increase in labor force
participation and employment relative to the control group. However,
Katz, Kling, and Liebman (2001) find that after five years MTO
households experienced no significant change in employment, earnings, or
public assistance receipts, although they lived in safer neighborhoods
that had lower poverty rates than the control group. Similarly, Ludwig,
Duncan, and Pinkston (2005) find that the MTO program resulted in no
change in earnings or employment, but there was a reduction in welfare
participation.
Participation in the Gautreaux and MTO programs was essentially
random but was voluntary since individuals selected could choose not to
participate. In addition, the MTO provides a control group. These
features allow the researchers to control for the unobserved
heterogeneity of living in public housing in the MTO studies; such
controls were not possible in the Gatreaux studies. The Hollman consent
decree for deconcentration of families in Minneapolis (following the
1995 court settlement) provides another example of voluntary and
involuntary moves from public housing. Goetz (2003) provides a detailed
account of the case and the resulting program where some households were
involuntarily moved from their demolished public housing projects (Goetz
likens this to a HOPE VI move). Other households voluntarily moved from
their residences into replacement housing or moved using a special
mobility certificate, similar to Gautreaux or MTO (Goetz 2003).
Evolution to HOPE VI
Throughout the 1980s and 1990s, a number of scholars provided a
renewed focus on the causes of urban poverty and implications of
concentrated poverty. For example, Wilson (1987) hypothesized that the
growth in urban poverty from the 1960s was due to changes in the
structure of the economy (less demand for low-skilled labor and
suburbanization of jobs) and the structure of inner-city neighborhoods.
Because of suburbanization and other factors, Wilson shows that the
inner-city neighborhoods became increasingly desperate. Curley (2005)
summarizes Wilson's argument that the inner-city neighborhoods
perpetuated isolation from jobs and networks needed to improve economic
conditions--and neighborhoods matter for economic well-being.
The development of this literature at a time of demonstration
projects such as Gautreaux and the MTO program provided new insights
into public policies aimed at alleviating poverty and provided fuel for
revisiting public housing rehabilitation. Congress adopted the Urban
Revitalization Demonstration Program in 1992, known as HOPE VI, to
renovate or demolish distressed public housing. The program has
demolished about 6% of all public housing units, that is, approximately
86,000 units (Popkin et al. 2004). The HOPE VI program targets the
physical conditions of public housing projects and the social and
economic environment of public housing residents.
Analyses of HOPE VI
We turn now to consideration of existing studies of HOPE VI,
focusing on results and methodological issues. Unlike the Gautreaux and
MTO programs, few studies have dealt with the effects of HOPE VI on
residents, particularly the effect on employment. The existing research
finds that the conditions of the HOPE VI project neighborhood improve
(Zielenbach 2003), that relocated residents perceive that they live in
better neighborhoods with less crime and lower poverty rates (Kleit and
Carlson 2003), that the relocated residents live in better housing
(Comey 2004), that educational outcomes for relocated children are
better (Popkin, Eiseman, and Cove 2004), and that mental health status
improved (Popkin and Cunningham 2002). Popkin et al. (2004) survey the
available research on these effects, including those that have used the
HOPE VI Panel Study Baseline and HOPE VI Panel Study Follow-Up, which
are extensive surveys of former residents of several HOPE VI projects
conducted by the Urban Institute.
The existing research finds no effect of a HOPE VI move out of a
public housing unit on the employment of former residents. These studies
include Levy and Kaye (2004), who used the Urban Institute's
surveys; Goetz (2002), who conducted a survey of 195 individuals in the
Minneapolis--St. Paul region; Clampet-Lundquist (2004), who conducted a
survey of 41 families from a HOPE VI project in Philadelphia; and Boston
(2005), who used AHA data to compare residents of four HOPE VI projects
with a control group consisting of residents of four other public
housing units.
Jacob (2004) addresses education outcomes from HOPE VI projects,
using agency data from the Chicago housing authority and school board.
Jacob uses public housing demolitions in Chicago to explore two
questions. First, does the program-based move affect the educational
outcomes of school-age residents, and, second, does living in public
housing relative to living in subsidized private housing (i.e., having a
housing voucher) affect educational outcomes of school-aged residents?
To address the first question, he compares those households that were
forced to move (HOPE VI participants) to those households in buildings
in the same housing project that were not demolished. (In Atlanta, all
units within a project were demolished, so we cannot make use of this
technique.) For the movers, Jacob finds no significant improvement on
academic achievement for younger children, a small negative effect on
academic achievement of older students, and a slight decrease in the
likelihood of dropping out for children under age 14. The former result
is contrary to the outcomes of other relocation programs. Jacob
interprets this contradiction as a result of the obligatory move feature
of relocation.
To explore the second question, Jacob must control for the
unobserved heterogeneity of being a public housing resident as opposed
to having a voucher. To address this issue, he uses demolitions as an
instrumental variable. He finds that being a public housing resident
relative to living in subsidized private housing has no effect on
educational outcomes.
Our empirical approach is similar to Jacob in that we use employee
records of employment and use the HOPE VI program-based move to control
for endogeneity. Our approach is an improvement over the studies that
rely on survey data. We use administrative records of employment as
reported to the DOL so that we capture the entire employed population.
This allows us to consider a much larger sample of tenants (in fact, the
entire population) and of HOPE VI projects and allows us to consider
employment over time rather than at one point in time. The studies using
survey data allow only a pre- and postmove comparison since there is no
control group. Boston (2005) uses AHA data on client-reported employment
status and compares the percentage of residents who are employed pre-
and postmove for HOPE VI movers and the control group. We compared AHA
client reported employment with DOL records and find substantial
underreporting of employment by Housing Authority clients. So we believe
that the DOL data provide better information on employment, and again
they allow us to analyze employment behavior long after a move.
None of the HOPE VI studies, other than Jacob (2004), control for
neighborhood effects. The literature suggests that in the case of
voluntary moves associated with the Gautreaux program, households
generally move to more middle-class neighborhoods, that is, those with
lower crime rates and less poverty (Rosenbaum 1995). However, while some
early work on MTO suggested that voluntary moves had a positive effect
on employment, more recent literature (Kling, Liebman, and Katz 2007)
suggests that there were no positive effects of such moves on
employment. (5) So a question remains as to whether a voluntary or a
program-based move has an impact on employment directly or because of
moving to a better neighborhood, characterized by lower levels of
poverty and crime and higher rates of employment and educational
attainment. We are able to explore this question given our unique
microdata for Atlanta.
3. Motivation, Data, and Empirical Methodology
The AHA has redeveloped seven public housing projects
("program public housing projects") through HOPE VI or HOPE
VI-like programs and is in the process of redeveloping the remaining
nonelderly housing projects. Each of these redevelopment projects
involves the demolition of the existing public housing project and the
construction of new housing as a mixed-income housing project. The
residents of these public housing projects were required to move but
were given a choice of moving to another public housing project or of
taking a rental voucher (called the housing choice program). Residents
were given some assistance in finding rental property, but in meetings
with participants in one HOPE VI project, it was suggested that the
assistance was not particularly helpful. AHA offered supportive
services, including job training and employment assistance to those who
were relocated, but participation in these programs was entirely a
decision of the resident, and participation was very low. Once the HOPE
VI project was completed, former residents could move into the rebuilt
project, subject to the availability of units and to meeting residency
requirements.
While previous studies of HOPE VI find no effect on employment, a
program-based move from an existing public housing project could, in
theory, have either a positive or a negative effect on employment. The
move could result in a loss of social networks and in psychological
costs that negatively affect employment. For example, if the resident
has to reestablish child care arrangements, it may increase absenteeism
resulting in a job loss. Or the loss of social networks that are a
source of information about job openings could reduce the probability of
finding employment. On the other hand, the residents' job search
pattern or proximity to potential jobs may improve in their new
location. At the same time, individuals at the margin of searching for a
job (new employment status or a change in job) might just find the
involuntary nature of the HOPE VI move the influence needed to find a
job or a better job.
The move can also result in a change in neighborhood conditions.
Wilson (1987, 1996) argues that distressed neighborhood conditions, such
as those in which public housing projects are located, negatively affect
the labor market outcomes, education, criminal behavior, and health of
public housing residents. There are empirical studies that focus on how
neighborhood characteristics affect families and children, but research
on neighborhood effects on employment is quite limited. (6) Most of the
studies focus on youth (e.g., O'Regan and Quigley 1996), and they
generally find that better neighborhood characteristics are associated
with improved labor market outcomes of youth; however, Oreopoulos (2003)
finds little impact of neighborhood characteristics on employment of
youth. In their literature review, Ellen and Turner (1997) identify only
two studies that focus on the relationship between neighborhood
characteristics and labor market outcomes of adults. Most studies that
consider neighborhood effects are in the context of an analysis of
relocation of public housing tenants. Kleit (2001) in fact finds
evidence that social networks of dispersed residents (vs. those
clustered in geographically close neighborhoods) provide better access
to information (including employment information). Her conclusions
suggest that dispersed housing (vs. clustered housing) may be a more
effective way to increase interactions between poor and nonpoor. These
results suggest that voucher users may (depending on the density of the
neighborhood) have more effective social network opportunities than
individuals who move back to public housing projects.
The data for this study are drawn from complete housing records of
the AHA and individual-level employment records of DOL. These data were
merged to construct a panel data set covering the period 1995-2003. The
data include employment status each quarter over the period (from the
DOL records) and characteristics of the individual, including age, race,
gender, and housing status (public housing or housing voucher) and
location (from the AHA records). (7) The AHA data report annual
conditions, and thus we match these annual data with the quarterly
employment data.
The AHA data do not contain information regarding education. This
is a limitation of our research since employment status is expected to
be associated with education level. We attempted to construct an
education variable by using Temporary Assistance for Needy Families
records,
which do contain education, and regressing that against employment
status. However, the equation did not perform very well. Instead, we use
employment in the previous quarter to reflect the effect of education on
employment. Those with more education are more likely to be employed,
and thus employment status should reflect education level. Of course,
employment status also reflects many other characteristics, such as
labor force attachment or skill level, so prior employment is much more
than a proxy for education. (8)
To this panel we added neighborhood characteristics, where
neighborhood is defined as the census tract in which the tenant lives.
The characteristics we use were selected on the basis of the
characteristics that other authors have included. Several variables were
obtained from the 2000 census and include percent of adults who are
employed, percent with less than a high school education, percent home
owner, percentage receiving public assistance, the poverty rate, and the
percent of households that are female headed. In addition, we include
the crime rate, which was constructed using individual crime records
from the city of Atlanta, which were geocoded to census tracts and
divided by the census tract's 2000 population. We expect the
probability of employment to be positively associated with the
employment and home ownership rates and negatively associated with the
other neighborhood characteristics. Finally, we include a measure of job
access, which was constructed using the following equation:
[Access.sub.i] = [j=J.summation over (j = 1)] [Emp.sub.j] x exp (-
[d.sub.ij]), (1)
where Accessi is access for a household living in census tract i,
Empj is the number of jobs in census tract j, and [d.sub.ij] is the
distance in miles between the centroids of the census tract i and j. (9)
We consider all census tracts within the Atlanta metropolitan
statistical area in measuring access. We expect that an increase in
access will be associated with an increase in the probability of
employment.
The panel data set is organized quarterly, although the
neighborhood characteristics for any census tract have the same value in
all quarters. The panel nature of the data allows us to follow the
variation in labor market participation of public housing residents
following the program-based move.
Our focus is on the impact of a program-based move on employment
and whether the impact is significantly different from a voluntary move
or no-move situation. We take advantage of the natural experiment of the
HOPE VI program to help identify the impact of a program-based move on
employment behavior. We can address this question by comparing the
employment for program-based movers to the employment of other tenants
of public housing projects. We can infer whether the program-based move
caused a change in the employment of those who moved, if being subject
to a program-based move is random. Among the 28 nonsenior public housing
projects in the city of Atlanta existing in 1995, six were demolished
between 1995 and 2003 (Clark Howell Homes, East Lake Meadows, Carver
Homes, Perry Homes, Harris Homes, and Capitol Homes) either as part of
the HOPE VI program or through a HOPE VI-like program. The first HOPE VI
project, Techwood Homes, was essentially completed by 1995 and thus is
not included in our data set.
In order to treat HOPE VI projects as random treatments, the
tenants of the HOPE VI projects must be a random draw of public housing
tenants, and the HOPE VI projects need to be a random draw of public
housing projects. The process for placing AHA clients into public
housing units implies that there is no bias in distribution of tenants
across housing projects. Applicants for public housing were placed on a
waiting list. Once the applicant reached the top of the queue, the
applicant was offered two units that met the applicant's need
(e.g., family size, disability condition, senior, and so on). If the
applicant rejected the units, the applicant had to reapply and would be
placed at the bottom of the waiting list. At one point, the applicant
even had to wait one year before reapplying. Once the tenant was in a
public housing unit, the tenant could move only if there was a change in
status, such as family size, although even this was not automatic. This
procedure certainly implies that there should be little bias in tenants
who occupied housing units in HOPE VI projects. We compared the
demographic characteristics of residents of the HOPE VI projects with
other nonelderly housing projects and found essentially no difference in
variables such as employment status, income, race, gender of head of
household, family size, and so on.
A potential concern with our sample of HOPE VI projects is that it
may not be representative of city of Atlanta public housing projects
generally--which means that the program-based move would not be random.
For example, the projects selected may have been among the worst
projects in the system. AHA certainly did not select the projects by
random draw. Rather, AHA selected projects that were in greatest need of
rehabilitation. For example, for the Harris Homes project, AHA stated in
its HOPE VI application that the reason for selecting this project for
HOPE VI was that the "buildings were obsolete" since it would
cost as much to modernize the buildings as to replace them (Housing
Authority of the City of Atlanta 1999, p. 2). That AHA is now redoing
all of the nonsenior housing projects suggests that the conditions of
housing in other public housing projects are not much better. In fact,
the justification for demolishing the last of the nonsenior projects was
that the buildings were obsolete.
There is no indication that the projects were selected on the basis
of neighborhood characteristics, either from our conversations with AHA
officials or from a comparison of neighborhood data. We compared the
neighborhood characteristics of the census tracts of HOPE VI projects
and the census tracts of non--HOPE VI, nonelderly projects and found
little difference. For example, the poverty rates were 39.5% and 40.6%
in the census tracts for the HOPE VI and non-HOPE VI projects,
respectively. Crime rates were within 10% of each other. The percentage
of adults who are employed is somewhat larger in the HOPE VI census
tracts, but the difference is not statistically significant.
Furthermore, based on the change in the number of owner-occupied housing
units, in median housing prices, and in the percentage of housing units
that were rental, there is no indication that the HOPE VI projects were
in neighborhoods that were more likely to be gentrifying. (10) The
selection of the first HOPE VI project, Techwood Homes, was driven in
large part by considerations associated with hosting the 1996 Summer
Olympics; we do not include Techwood Homes in our analysis. The other
projects reflect a geographic dispersal of the then existing public
housing projects. Thus, we believe that the HOPE VI projects reflect in
essence a random draw of the public housing projects.
If the selection of projects for HOPE VI is in essence random, then
we can assume that there is no unobserved heterogeneity. Thus, we can
use our results to infer whether the program-based move caused a change
in employment status relative to that of other public housing residents.
Of course, it is possible that HOPE VI projects do differ from other
housing projects in some unobservable way that we cannot account for.
We estimate a probit model of the probability of employment,
denoted ProbE, as a function of lagged employment, denoted [E.sub.t-1];
a set of demographic characteristics of the individual, denoted C; a set
of neighborhood characteristics, denoted N; a set of year dummies,
denoted T; and a program-based move dummy, denoted DF. DF equals one in
the quarter in which the resident moved from a public housing project as
a result of a HOPE VI project and for every subsequent quarter and zero
otherwise. A program-based move could be either to another public
housing project or to subsidized market housing (i.e., to the housing
voucher program). The equation is specified as
Pr([E.sub.it] = l | X) = F([[alpha].sub.0] + [[alpha].sub.1]
[E.sub.it-1] + [C.sub.it][[alpha].sub.2] + [[alpha].sub.3]D[F.sub.it] +
N[[alpha].sub.4] + [T.sub.it][[alpha].sub.5]), (2)
where i represents the ith individual, t is the tth quarter, and F
is a normal probability density function. Since the treatment group is a
random sample of the public housing population, we do not need to
include a treatment dummy variable, which would make Equation 2 a
standard difference-in-difference model. However, we did estimate
Equation 2 as a difference-indifference model by including a
treatment-group dummy. The coefficient on the treatment-group dummy was
not statistically significant, and its inclusion did not change the
coefficient on the other variables. Equation 2 is estimated using
quarterly data.
Our sample consists of any adult (16-65 years of age) who, in 1995
or when first observed in the AHA file at any point during the period
1995-2003, was a resident of a public housing project. We exclude anyone
who was in the AHA file for only one year. The sample size changes over
time as individuals enter public housing or cease being an AHA client.
There were 18,645 such unique individuals identified over the nine-year
period, 2191 of whom had a HOPE VI program-based move (72 of whom had
two such moves). Note that residents of other public housing could have
moved voluntarily. We observe employment each quarter, but since not all
individuals were in the sample every quarter and some observations could
not be geocoded, we have a total of 313,662 usable observations in the
full sample.
We lose observations over time because of attrition. For 1995,
there were 9231 individuals in the sample, which falls to 4007 in 2003.
In part this is due to individuals who reach 65 years of age and are
dropped from the sample. Most of the attrition is due to family members
moving out of the household, but there are also evictions due to
nonpayment of rent or violation of AHA regulations. The actual sample
increases to 10,660 by 2003 as a result of individuals moving into an
AHA client's household, new households moving into a public housing
unit, or existing residents reaching 16 years of age. There were many
individuals who were in the AHA file for just one year, typically
someone who moved into an existing AHA household. Because these
individuals are so transient, we excluded them from the sample. This
also reduces the attrition rate and excludes individuals who were in the
AHA file in 1995 but who left public housing earlier.
If the individual is employed in any quarter, the dependent
variable equals one in that quarter and zero otherwise. The demographic
variables include age, race, and gender. Marital status was missing for
a large number of observations, and thus we exclude marital status as an
independent variable. However, over 97% of the individuals for whom we
know marital status are single; the empirical results are essentially
the same if we include marital status using the reduced sample size.
Table 1 contains the names of the variables and their means and
standard deviations. Some of the variables are used not in the initial
regressions but in subsequent regressions.
4. Results
Recall that our focus is on the effect that a program-based move
has on the probability of employment. We empirically analyze this effect
by first comparing program-based movers to other public housing
residents (Table 2). In a second set of regressions, we compare the
employment outcomes of program-based movers to other types of movers
(Tables 4-7).
Analysis of Program-Based Moves
Table 2 presents the probit regression results of a program-based
move on the probability of employment relative to all other residents of
public housing units. We report both the estimated parameter from the
probit regression as well as the marginal effects calculated at the mean
of the variables. Since being subject to a program-based move is taken
to be random, we can simply compare the employment subsequent to the
program-based move for those subject to the program-based move relative
to the employment of other public housing residents.
The coefficient on the program-based move dummy (DF) is positive
and has a very small standard error. Thus, it appears that a
program-based move causes an increase in the resident's probability
of being employed relative to other residents of public housing units
who were not subject to the program-based move. The marginal effect of a
program-based move is to increase the probability of employment by 3.3
percentage points or by 8.3% over the mean employment rate--an
economically important impact.
The result for DF runs counter to the findings of other HOPE VI
studies, which find either a negative or no effect on employment from a
HOPE VI move. This may be due to our ability to control for other
characteristics or provide a more accurate measure of employment than
the self-reported status used in other studies, or it may be due to the
size of the database, which may give this study increased explanatory
power. Compared to the near doubling of employment that Rosenbaum and
Harris (2001) find in their study of the Chicago MTO program, 8.3% is
small, but they note that the change is strongly influenced by the
change in the economy. On the other hand, Ludwig, Duncan, and Pinkston
(2005) find a decrease in welfare receipts of 11% to 16% but no change
in employment in their study of the effects of MTO. (11)
The coefficients on the individual characteristics are significant,
but the signs are generally not what might be expected. We control for
employment in the previous quarter. If the individual was employed (not
employed) in the previous quarter, the coefficients on the individual
characteristics would measure the characteristic's effect on
remaining (becoming) employed in the subsequent quarter. If females and
blacks are more likely to change jobs and to have longer spells of
unemployment, we would expect the coefficients on gender and race to be
negative. However, the results suggest that females are more likely to
be employed than males and that nonwhites are more likely to be employed
than whites. These results could be due to the peculiarities of the
public housing population; there are very few whites and males, and they
may not be representative of the more general population. The
coefficient on lagged employment is positive and statistically
significant, as was expected. We also estimated the regression without
the lagged employment variable; there was no change in the sign or
statistical significance of the coefficient on the move dummy
coefficient, although its magnitude increased. (12)
Year dummies were included but not reported. They are statistically
significant. Excluding year dummies increases the coefficient on DF,
with an increase in the marginal effect to 4.2 percentage points. We
also tried quarterly dummies; the results did not change.
The coefficient on ACCESS has an unexpected negative sign, but it
is insignificant. The negative coefficient could be a consequence of A
CCESS reflecting access to all jobs and not just low-skilled jobs. Four
of the eight coefficients on the neighborhood characteristics are
statistically significant, but one, percent employed, has an unexpected
sign. There is some correlation among these variables (percent public
assistance, crime, and employment), so we experimented with dropping
some of the variables. The results were not substantially different. The
marginal effects of the significant variables are quite small--possibly
reflecting a limited effect of these neighborhood characteristics.
Furthering investigating the ACCESS variable, we find that the
within-year variation is relatively constant, but over time, the mean
value of the variable increases. The time dummies may have picked up
that year-to-year variation.
One of the possible effects of being subject to a program-based
move is that the individual moves to a neighborhood that is more
conducive to being employed; for example, access to employment may
improve. We calculated the mean value of our neighborhood
characteristics for individuals who were originally HOPE VI residents
before their move and neighborhood characteristics for the same
individuals after their move from the HOPE VI housing project (Table 3).
All but one of the neighborhoods characteristics would be considered to
have improved after the move from the HOPE VI project (i.e., lower crime
rate, higher average level of education, and so on). The only exception
was ACCESS, which measures access to employment. As noted previously,
ACCESS measures access to all jobs, not just low-skilled jobs.
Furthermore, many of the public housing projects are close to the
central business district, while HOPE VI residents who took a housing
voucher found housing further from the CBD. This is consistent with the
observed increase in the value of ACCESS for those who moved to another
public housing unit and the decrease in value for those who took a
voucher (Table 3). All but one of the differences are statistically
significant, the exception being the percent of households that are
female headed. Those HOPE VI residents who took a voucher moved to
neighborhoods whose characteristics were generally better than those who
moved to another public housing unit.
An issue that arises is whether the change in neighborhood
characteristics is the result of the program-based move. If the change
in neighborhood characteristics is caused by the program-based move and
neighborhood characteristics affect employment, then including the
neighborhood characteristics in the regression would be inappropriate
since these variables measure some of the effect of the program-based
move. Therefore, we estimate Equation 2 excluding the neighborhood
characteristics. The results are reported in the last two columns of
Table 2 and are very similar to the other regression reported in Table
2; in particular, the coefficient on DF is essentially the same,
although it is measured a bit more precisely in the regression without
the neighborhood characteristics.
Changes in the sample over time through attraction and attrition
are a potential empirical problem. To examine the possible effects of
attrition, we used a "no-attrition sample," which is a
subsample consisting of only those individuals who were in public
housing in 1995 and remained AHA clients through 2003. The size and
significance of the coefficient on DF for this regression is essentially
the same as that reported in Table 2. (13)
The principal concern with attrition is that those who leave the
sample may have employment patterns that differ systematically from
those who do not leave. For a given year, we compared the number of
quarters that each person was employed in the previous two years. We
found that there was no statistically significant difference in the
employment histories of those who left the sample and those who did not,
and that is the case for both those who were residents of a HOPE VI
project and those who were not.
These results suggest that attrition does not substantially affect
the empirical results. However, there are other methodological
approaches that would deal with attrition more directly; for example, we
might estimate an equation for the probability of attrition and use that
to adjust our estimates of Equation 2. Based on the variables we have
available, including employment history, we find nothing related to the
probability of attrition. This lack of identification means that we are
unable to use this technique. (14)
Individuals may learn that a project would be demolished before the
actual demolition begins and therefore move out earlier than the
demolition date. In order to control for this potential
"announcement effect," we changed the measured program-based
move date to four quarters prior to the program-based move date used in
the previous regressions. We also estimated other alternative
specifications to deal with potential issues of announcement effects. We
did not find any significant deviation from our original results.
In summary, our empirical results suggest that an involuntary
program-based move increases employment relative to the employment of
other residents of public housing units. These results seem robust to
alternative samples and choice of independent variables.
Analysis of Other Moves
It is of interest to determine how the effect of a program-based
move on employment compares to the change in employment associated with
a voluntary move from a public housing project. It is expected that
voluntary movers have a different level of motivation than program-based
movers, and thus a voluntary move is likely associated with unobserved
personal characteristics. Therefore, we cannot imply any causal effect
of any difference in the outcome of the two types of moves, but we can
still address the interesting empirical question of whether that
motivation to move voluntarily translates into better labor market
outcomes than does a program-based move.
Table 4 reports the probit estimation in which we include a dummy
for program-based moves (DF) and a dummy for voluntary moves from a
public housing unit (DV), using the same sample as for the results in
Table 2. DV equals one if the move is a voluntary move (non-HOPE VI
move) from public housing to either public housing or housing voucher,
where 4692 individuals made that move. In this case, the two dummy
variables measure the change in the probability of employment relative
to those individuals in public housing who did not move during the
period.
It is possible that a voluntary move and employment are related to
unobserved characteristics of the individuals. However, including lagged
employment should capture the effect of these unobserved
characteristics.
The coefficients on DF and DV are both positive and significant;
that is, both types of moves are associated with an increase in the
probability of employment relative to public housing tenants who were
nonmovers. The coefficient on DF is larger than the coefficient on DV,
as are the marginal effects (4.9 vs. 3.7 percentage points), but is
larger than the coefficient on DF reported in Table 2. These results are
not sensitive to excluding some or all of the neighborhood and
individual characteristics, including lagged employment.
Voluntary movers may not incur the same types of costs if the
individuals planned the move and thus prepared themselves by choosing a
new location that provides social support. Or a voluntary move could
simply be a signal of an individual's willingness to seek new
opportunities, including employment. Thus, a program-based move could
result in either an increase or a decrease in the probability of
employment, but our priors were that a voluntary move will result in a
more positive effect on employment than a program-based move, which is
consistent with the empirical results.
For this same reason, we expected a larger marginal effect for DF
for the equation in Table 4 than for the equation in Table 2. The
coefficient on DF in Table 2 measures the change in the probability of
employment relative to all other public housing residents, including
those who might make a voluntary move in the future, while in Table 4 it
measures the change relative to public housing residents who do not move
as a result of either a program-based move or a voluntary move.
Next, we considered the comparison between just movers (i.e.,
program-based moves and voluntary moves) in a second regression (Table
5). Here we calculated the change in employment pre- and postmove
(measured in the quarters before and immediately after the move) for
both program-based and Voluntary moves for the first move post-1995,
estimating a difference-in-differences model using ordinary least
squares. While the probit regression reported in Table 4 can be used to
compare the employment results for just movers, using the panel data to
estimate a difference-in-difference model provides a somewhat stronger
approach to the question. We excluded all nonmovers for this regression.
In this regression, DF equals one if the move was a program-based move
and zero if the move was voluntary. The coefficient on DF is positive,
which is consistent with the results in Table 4, but the coefficient is
statistically insignificant. Because there are likely to be unobserved
characteristics associated with the voluntary move, we cannot associate
any causality between the move and any change in employment. We can say
only that those who experience a program-based move experience a change
in employment at the time of the move that is the same as for those who
move voluntarily.
Finally, both voluntary and program-based moves can end in either
public housing or housing voucher. Again, while the motivation for the
postmove choice of housing program may differ, it is an interesting
empirical issue as to whether an individual who experiences a
program-based move from public housing and decides to move to the
housing voucher program has a different employment outcome than someone
who chooses to move from public housing to another public housing unit.
For example, do the employment outcomes for a program-based move differ
between those who choose housing voucher and those who choose another
public housing project? Again, we cannot imply a causal relationship
between the employment and the postmove housing choice.
The dummy variables for the various types of moves are therefore
classified as follows (the figures in parentheses are the number of
individuals observed making that move):
DF-PH = 1 if the move is a program-based move from public housing
to public housing (802).
DF-HC = 1 if the move is a program-based move from public housing
to housing voucher (1461).
DV-PH = 1 if the move is a voluntary move from public housing to
public housing (2295).
DV-HC = 1 if the move is a voluntary move from public housing to
housing voucher (2632).
DC-PH = 1 if the move is a voluntary move from housing voucher to
public housing (366).
Once an individual moves, he or she is reclassified as a mover and
remains identified that way unless he or she moves again, at which point
he or she is again reclassified.
The results of the probit regressions are reported in Table 6,
while Table 7 contains the marginal effects for dummy "move"
variables. The sample is the same as that used for the regressions
reported in Table 2. The coefficients on the demographic variables in
the regression in Table 6 are similar to those reported in Table 2, and
thus we do not discuss them. The coefficients on the neighborhood
characteristics are generally insignificant. The dummy variables in the
regressions are the main focus--what is the association between the
probability of employment and the nature of the move and postmove
housing choice?
The regression reported in column 1 of Table 6 includes dummy
variables for all five of the possible combinations of move types and
postmove outcomes; the excluded category is public housing nonmovers.
Thus, the coefficients on the dummy variables measure the change in the
probability of employment relative to public housing residents who do
not move. Dummy variables DF-PH and DV-PH reflect the effect associated
with program-based and voluntary moves from public housing units to
another public housing unit, respectively. The coefficient on DF-PH is
positive but statistically insignificant. Thus, relative to public
housing nonmovers, a program-based move from one public housing unit to
another is not associated with a change in the probability of
employment. The coefficient on DV-PH, a voluntary move from one public
housing unit to another, is positive and significant.
DF-HC and DV-HC measure the change in the probability of employment
of a move from a public housing unit to the housing voucher program for
program-based and voluntary moves, respectively. The coefficients for
both variables are positive and statistically significant. The
coefficient on DV-PH is much smaller than the coefficient on DV-HC.
These results suggest that the change in the probability of employment
associated with program-based and voluntary moves reported in Table 4 is
driven by those who move out of public housing and into a private market
housing unit using a housing voucher. (The coefficients on DF-HC and
DV-HC are not statistically different.)
Some AHA clients moved from public housing to housing voucher and
then back. The return move is reflected in DC-PH. The coefficient on
DC-PH (a move from the voucher program to a public housing unit) in
column 1 of Table 6 is positive and significant, suggesting that those
who move from the housing voucher program to a public housing unit
experience an increase in the probability of employment relative to
individuals in public housing who have not moved. This is an unexpected
result. We expected that someone who moved from housing voucher to a
public housing unit would have experienced a decrease in the probability
of employment and certainly not a change in the probability of
employment of the same magnitude as that associated with a voluntary
move from public housing to housing voucher.
For the regression reported in column 2 of Table 6, we restrict the
sample to those who were in public housing and either did not move or
moved to another public housing unit. Thus, DF-PH and DV-PH measure the
change in the probability of employment associated with a move to
another public housing unit relative to the employment of public housing
residents who did not move. The coefficients on DF-PH and DV-PH are
similar in magnitude and significance levels as the coefficients on
these dummy variables in column 1. A move from a public housing unit to
another public housing unit is associated with no change in the
probability of employment relative to those who stay in public housing
for program-based moves but an increase for voluntary moves.
The regression in column 3 restricts the sample to those who move
from public housing to the voucher program and public housing nonmovers
and thus exclude those who move to another public housing unit or move
from housing voucher to a public housing unit. The coefficients on DF-HC
and DV-HC are similar in magnitude and statistical significance to the
coefficients on these dummy variables in column 1. In other words, a
program-based move from public housing to a voucher program is
associated with the same positive change in employment as a voluntary
move from public housing to a voucher program.
Column 4 of Table 6 considers those who made a move to another
public housing unit and thus excludes all those who did not move, who
moved to housing voucher, and who move from housing voucher to public
housing. Thus, DF-PH measures the change in the probability of
employment associated with a program-based move to another public
housing unit relative to a voluntary move to another public housing
unit. The coefficient on DF-PH is negative and significant, implying
that those clients who experienced a program-based move and choose a
public housing unit as the postmove option had a reduction in employment
as compared to public housing tenants who voluntarily move to another
public housing unit. This result is consistent with the difference in
the values of the coefficients on DF-PH and DV-PH in column 1.
Finally, in column 5 of Table 6, we consider those who made a move
to housing voucher and thus exclude all those who did not move, who
moved to public housing, and who move from housing voucher to public
housing. Thus, DF-HC measures the effect on the probability of
employment associated with a program-based move to housing voucher
relative to a voluntary move to another housing voucher. The coefficient
on DF-HC is negative, small, and statistically insignificant, which
implies that there is no difference in the change in employment
associated with a program-based move to housing voucher in comparison to
voluntary moves to housing voucher. This result is consistent with the
similar sized coefficients on DF-HC and DV-HC in column 1.
We also estimated the equations in Table 6 using the no-attrition
sample, which includes only those individuals who were in public housing
in 1995 and were still AHA clients in 2003. The results for this sample
are consistent with the results reported in Table 6.
Overall, we find support for the hypothesis that a program-based
move can improve employment relative to other public housing residents
and that a program-based move has the same effect on the probability of
employment as that associated with a voluntary move. The impact is
larger for those who moved from a public housing unit to a housing
voucher unit, and that is the case for both program-based and voluntary
moves. These results are consistent with the hypothesis that public
housing has a negative effect on employment. However, those who choose
housing voucher as their postmove options are not a random sample of
program-based movers, so no causation can be implied.
How might we explain these results in a more intuitive sense?
Consider the following. Suppose there are two types of public housing
residents, denoted RH6 and RN, where RH6 are residents of a HOPE VI
project and RN are not. The HOPE VI project requires RH6 residents to
move and to select between another public housing project and a housing
voucher. Suppose some RH6 residents, who choose a housing voucher, move
to locations that increase the chance of obtaining employment.
Furthermore, suppose that employers attach a negative stigma to being a
public housing resident so that switching to a voucher would increase
the probability of obtaining employment. (The latter implies that being
a public housing resident reduces the probability of being employed, a
result found by Ong [1998], Fischer [2000], and Reingold, Van Ryzin, and
Ronda [2001].) RN residents, on the other hand, do not have the same
opportunity to obtain a housing voucher or to move to a different
housing project. And, in the context of behavioral economics, not being
required to choose, the RN residents do not pursue alternative housing
arrangements. The implication is that being a HOPE VI resident would
cause an increase in the probability of employment relative to non-HOPE
VI residents, which is what we find (Table 2).
Some RN residents do move and, in particular, switch to a housing
voucher. These residents could also move to locations that increase
employment and would no longer be subject to possible employer
discrimination against public housing residents. Thus, we would expect
that a voluntary switch to a housing voucher would be associated with an
increase in the probability of employment and an increase about equal to
the increase associated with HOPE VI residents who choose a housing
voucher. This is also consistent with our findings (Table 6).
5. Summary
The focus of this research is on the impact of an exogenous
program-based move on employment behavior. We find evidence that
program-based moves have a positive causal effect on employment
outcomes; that is, program-based moves result in an increase in the
probability of employment relative to other public housing residents.
This effect seems to be driven by program-based movers who choose the
housing voucher program as their postmove alternative.
Compared to the change in employment associated with a voluntary
move, program-based moves are associated with a slightly larger positive
change in employment. The change in the probability of employment
associated with those who were subject to a program-based move and
select housing voucher is similar to the change associated with those
who voluntarily moved to housing voucher.
The findings are important to the policy debate surrounding HOPE VI
revitalization projects. Our results suggest that HOPE VI has a positive
effect on the probability of employment relative to the employment of
other public housing residents. These results are different than those
found in previous studies. Whether our results are unique to Atlanta or
to the methodology is unknown, but certainly the results suggest that
similar studies in other city should be conducted to help sort out that
issue.
References
Boston, Thomas D. 2005. The effects of revitalization on public
housing residents. Journal of the American Planning Association
71(4):393-407.
Clampet-Lundquist, Susan. 2004. Moving over or moving up?
Short-terms gains and losses for relocated HOPE VI families. Cityscape:
A Journal of Policy Development and Research 7(1):57-80.
Comey, Jennifer. 2004. An improved living environment? Housing
quality for HOPE VI relocates. Brief No. 2 (September). Metropolitan
Housing and Communities Center, Urban Institute.
Corcoran, Mary, and Colleen Heflin. 2003. Barriers to work among
recipients of housing assistance. Cityscape." A Journal of Policy
Development and Research 6(2):73-87.
Curley, Alexandra. 2005. Theories of urban poverty and implications
for public housing policy. Journal of Sociology and Social Welfare
32(2):97-119.
DeLuca, Stefanie, and James E. Rosenbaum. 2003. If low-income
blacks are given a chance to live in white neighborhoods, will they
stay? Examining mobility patterns in a quasi-experimental program with
administrative data. Housing Policy Debate 14(3):305-45.
Edin, Kathryn, and Laura Lein. 1997. Making ends meet: How single
mothers survive welfare and low-wage work. New York: Russell Sage
Foundation.
Ellen, Ingrid Gould, and Margery Austin Turner. 1997. Does
neighborhood matter? Assessing recent evidence. Housing Policy Debate
8(4):833-66.
Fischer, Will. 2000. Labor supply effects of federal rental
subsidies. Journal of Housing Economics 9(3):150-74.
Goetz, Edward G. 2002. Forced relocation vs. voluntary mobility:
The effects of dispersal programmes on households. Housing Studies
17(1):107-23.
Goetz, Edward G. 2003. Clearing the way: Deconcentrating the poor
in urban America. Washington, DC: Urban Institute Press.
Housing Authority of the City of Atlanta. 1999. 1999 Joel C. Harris
homes HOPE VI grant application. Atlanta: Housing Authority of the City
of Atlanta.
Jacob, Brian A. 2004. Public housing, housing vouchers, and student
achievement: Evidence from public housing demolitions in Chicago.
American Economic Review 94(1):233-58.
Johnson, Michael P., Helen F. Ladd, and Jens Ludwig. 2002. The
benefits and costs of residential mobility programmes for the poor.
Housing Studies 17(1): 125-38.
Katz, Lawrence F., Jeffrey R. Kling, and Jeffrey B. Liebman. 2001.
Moving to opportunity in Boston: Early results of a randomized mobility
experiment. Quarterly Journal of Economics 116:607-54.
Kleit, Rachel Garshick. 2001. The role of neighborhood social
networks in scattered-site public housing residents' search for
jobs. Housing Policy Debate 12(3):541-73.
Kleit, Rachel Garshick, and Daniel Carlson. 2003. Holly Park and
Roxbury HOPE VI redevelopments: Evaluation report. Report submitted to
Seattle Housing Authority.
Kling, Jeffrey R., Jeffrey B. Liebman, and Lawrence F. Katz. 2007.
Experimental analysis of neighborhood effects. Econometrica
75(1):83-119.
Levy, Diane, and Deborah R. Kaye. 2004. How are HOPE VI families
faring? Income and employment. Washington, DC: Urban Institute Press.
Ludwig, Jens, Greg J. Duncan, and Joshua C. Pinkston. 2005. Housing
mobility programs and economic self-sufficiency: Evidence from a
randomized experiment. Journal of Public Economics 89:131-56.
Ong, Paul. 1998. Subsidized housing and work among welfare
recipients. Housing Policy Debate 9(4):775-94.
O'Regan, Katherine, and John Quigley. 1996. Spatial effects
upon employment outcomes: The case of New Jersey teenagers. New England
Economic Review (May-June): 41-59.
Oreopoulos, Philip. 2003. The long-run consequences of living in a
poor neighborhood. Quarterly Journal of Economics 118(4): 1533-75.
Popkin, Susan J., and Mary K. Cunningham. 2002. CHA relocation
counseling assessment: Final report. Washington, DC: Urban Institute.
Popkin, Susan J., Michael Eiseman, and Elizabeth Cove. 2004. How
are HOPE VI families faring? Children. Washington, DC: Urban Institute
Press.
Popkin, Susan J., Bruce Katz, Mary K. Cunningham, Karen D. Brown,
Jeremy Gustafson, and Margery Austin Turner. 2004. A decade of HOPE VI:
Research findings and policy challenges. Washington, DC: Urban Institute
Press.
Reingold, David A., Gregg G. Van Ryzin, and Michelle Ronda. 2001.
Does urban public housing diminish the social capital and labor force
activity of its tenants? Journal of Policy Analysis and Management
20(3):485-504.
Rosenbaum, James E. 1995. Changing the geography of opportunity by
expanding residential choice: Lessons from the Gautreaux program.
Housing Policy Debate 6(1):231-69.
Rosenbaum, James E., and Laura Harris. 2001. Residential mobility
and opportunity: Early impacts of the moving to opportunity
demonstration program in Chicago. Housing Policy Debate 12(2):321-46.
Schone, Barbara Steinberg. 1992. Do means tested transfers reduce
labor supply? Economics Letters 40(3):353-57.
Shroder, Mark. 2002. Does housing assistance perversely affect
self-sufficiency? A review essay. Journal of Housing Economics
11(4):381-417.
Turney, Kristin, Susan Clampet-Lundquist, Kathryn Edin, Jeffrey R.
Kling, and Gregory J. Duncan. 2006. Neighborhood effects on barriers to
employment: Results from a randomized housing mobility experiment in
Baltimore. Brookings-Wharton Papers on Urban Affairs, 137-72.
U.S. Department of Housing and Urban Development. 2006. Effects of
housing vouchers on welfare families. Washington, DC: Abt Associates.
Wilson, William J. 1987. The truly disadvantaged. Chicago:
University of Chicago Press.
Wilson, William J. 1996. When work disappears. New York: Vintage
Books.
Zielenbach, Scan. 2003. Assessing economic change in HOPE VI
neighborhoods. Housing Policy Debate 14(4):621-55.
Bulent Anil, Department of Economics, Labovitz School of Business
and Economics, University of Minnesota-Duluth, 412 Library Drive,
Duluth, MN 55812-3029, USA.
David L. Sjoquist, Department of Economics, Andrew Young School of
Policy Studies, Georgia State University, 14 Marietta Street NW, P.O.
Box 3992, Atlanta, GA 30302-3992, USA; E-mail
[email protected];
corresponding author.
Sally Wallace, Department of Economics, Andrew Young School of
Policy Studies, Georgia State University, 14 Marietta Street NW, P.O.
Box 3992, Atlanta, GA 30302-3992, USA; E-mail
[email protected].
We benefited from comments from Thomas Boston, Mike Proctor, Erdal
Tekin, Geoffrey Turnbull, Mary Beth Walker, John Pepper, participants at
the AREUEA Spring Meetings 2006 and Southern Economic Association
Meetings 2006, and two anonymous referees.
Received July 2007; accepted January 2009.
(1) HUD, "A Picture of Subsidized Housing"
(http://www.huduser.org/picture2000/index.html). However, as pointed out
by a referee, the total number of households receiving low-income
housing is actually much greater.
(2) http://www.hud.gov/offices/pih/programs/ph/hope6/about/fundinghistory.pdf.
(3) In some cases, but not for Atlanta, residents may remain
on-site during the redevelopment process.
(4) Shroder (2002) provides a concise review of the theoretical and
empirical literature regarding the impact of housing subsidies on the
"self-sufficiency" of families. Appealing to a traditional
neoclassical model of labor supply (with all other goods representing
the other argument in a consumer's utility function), he
demonstrates the result that a housing subsidy reduces labor supply. A
critique of the standard model by Schone (1992) demonstrates that with
relatively standard consumer preferences, subsidies can in fact induce
increased labor supply. Abt Associates in a study for HUD (2006) on the
effect of vouchers on families finds no significant long-term (three and
a half years) impact of vouchers on employment.
(5) However, Turney et al. (2006) find a potential upside to moving
to a better neighborhood. Using a set of interviews, they find that
unemployed respondents in the MTO treatment group were more committed to
remaining in the labor force. This may have positive long-term impacts
on employment opportunities.
(6) Popkin et al. (2004) and Ellen and Turner (1997) provide
reviews of the literature on neighborhood effects.
(7) Survey data notoriously underreport items such as employment
and income. A comparison between the survey reported employment and the
Georgia DOL file confirms this. The DOL data suggest employment rates
that are about double the self-reported employment. Of course, DOL data
underreport employment; not all employees are required to report
employment, and off-the-book employment is not reported (Edin and Lein
1997).
(8) If, as we argue later, our treatment group is a random sample
of housing tenants and if education levels do not change over time, the
absence of education should not affect the coefficient on the treatment
variable.
(9) It would have been preferred if we could have used low-skill
jobs. However, the ES 202 data used to construct this variable did not
allow this. We thank Joey Smith for making his calculations of
employment by census tract available to us.
(10) We computed the change for the two sets of neighborhoods over
the periods 1980 and 1990. The changes were the same for the HOPE VI and
non-HOPE VI neighborhoods. For example, in 1980, the percent owner
occupied was 24% for the HOPE VI neighborhoods and increased by 7.5% and
was 22% for the non HOPE VI neighborhoods and increased by 7.9%.
(11) Rosenbaum and Harris (2001) compare pre- and postemployment
means of MTO participants from Chicago 18 months after the housing move.
They note that at least part of the increase in employment can be
attributed to the robust economy. Ludwig, Duncan, and Pinkston (2005)
use administrative employment records to estimate a probit equation to
determine the intent to treat effect. They compare the MTO experimental
group with the control group and the Section 8 group with the control
group.
(12) We also estimated the model including the number of children
in the household, alternatively defined as those under 6 years of age
and under 18 years of age. This did not affect the results.
(13) We estimated virtually all the regression results for both the
full and the nonattrition subsamples and did not find significantly
different results between the two samples. In addition, we estimated the
regressions using a sample that include those individuals who were in
the AHA file for only one year and a sample that excluded the first
year. The results are consistent with those reported in the tables.
(14) A lack of identification means that we are not able to
statistically show that the attrition is not driving the results. Given
the consistency of results between the general and nonattrition samples,
there is evidence that the attrition sorting is not affecting our
results.
Table 1. Descriptive Statistics
All
Mean (Standard
Variable Deviation) N
Employment
(= 1 if employed that quarter) 0.397 (0.489) 360,995
Age 35.7 (13.8) 360,995
Race (= 1 if nonwhite) 0.978 (0.146) 360,995
Gender (= 1 if female) 0.792 (0.406) 360,995
Nonmovers 0.706 (0.456) 360,995
DF (program-based move) 0.090 (0.287) 360,995
% EMPLOYMENT (a) 42.1 (13.4) 346,155
% LESS THAN HIGH SCHOOL (a) 41.2 (14.0) 346,155
% HOME OWNERSHIP (a) 26.0 (19.1) 346,155
% PUBLIC ASSISTANCE (a) 22.4 (15.4) 346,155
% POVERTY (a) 45.2 (22.6) 346,155
% FEMALE HEAD (a) 38.8 (16.6) 346,155
ACCESS (a) 24,457.3 (21,058.3) 340,810
PER CAPITA CRIME (a) 0.379 (0.397) 337,304
DF-PH (program-based move: public
housing to public housing) 0.032 (0.175) 360,995
DF-HC (program-based move: public
housing to choice) 0.059 (0.235) 360,995
DV-PH (voluntary move: public
housing to public housing) 0.090 (0.287) 360,995
DV-HC (voluntary move: public
housing to choice) 0.106 (0.308) 360,995
DC-PH (voluntary move: choice
to public housing) 0.007 (0.086) 360,995
HOPE VI
Mean (Standard
Variable Deviation) N
Employment
(= 1 if employed that quarter) 0.457 (0.498) 32,652
Age 36.6 (11.7) 32,652
Race (= 1 if nonwhite) 0.985 (0.123) 32,652
Gender (= 1 if female) 0.871 (0.335) 32,652
Nonmovers -- --
DF (program-based move) -- --
% EMPLOYMENT (a) 46.5 (12.5) 31,249
% LESS THAN HIGH SCHOOL (a) 36.3 (12.7) 31,249
% HOME OWNERSHIP (a) 33.1 (22.0) 31,249
% PUBLIC ASSISTANCE (a) 16.8 (13.8) 31,249
% POVERTY (a) 37.2 (21.5) 31,249
% FEMALE HEAD (a) 36.1 (13.5) 31,249
ACCESS (a) 20,544.4 (19,961.7) 29,618
PER CAPITA CRIME (a) 0.288 (0.311) 28,922
DF-PH (program-based move: public
housing to public housing) 0.350 (0.477) 32,652
DF-HC (program-based move: public
housing to choice) 0.650 (0.477) 32,652
DV-PH (voluntary move: public
housing to public housing) -- --
DV-HC (voluntary move: public
housing to choice) -- --
DC-PH (voluntary move: choice
to public housing) -- --
Other
Mean (Standard
Variable Deviation) N
Employment
(= 1 if employed that quarter) 0.391 (0.488) 328,343
Age 35.6 (13.9) 328,343
Race (= 1 if nonwhite) 0.978 (0.148) 328,343
Gender (= 1 if female) 0.784 (0.412) 328,343
Nonmovers 0.784 (0.412) 328,343
DF (program-based move) -- --
% EMPLOYMENT (a) 41.7 (13.4) 314,906
% LESS THAN HIGH SCHOOL (a) 41.7 (14.0) 314,906
% HOME OWNERSHIP (a) 25.3 (18.7) 314,906
% PUBLIC ASSISTANCE (a) 22.9 (15.4) 314,906
% POVERTY (a) 46.0 (22.6) 314,906
% FEMALE HEAD (a) 39.1 (16.9) 314,906
ACCESS (a) 24,829.7 (21,122.0) 311,192
PER CAPITA CRIME (a) 0.387 (0.403) 308,382
DF-PH (program-based move: public
housing to public housing) -- --
DF-HC (program-based move: public
housing to choice) -- --
DV-PH (voluntary move: public
housing to public housing) 0.099 (0.299) 328,342
DV-HC (voluntary move: public
housing to choice) 0.1171 (0.321) 328,342
DC-PH (voluntary move: choice
to public housing) 0.008 (0.090) 328,342
(a) Census tract variables.
Table 2. Probit Regression Explaining Employment Effect of
Program-Based Moves (Dependent Variable: Quarterly Employment Dummy)
Coefficient
Variable (Standard Error) Marginal Effect
Constant -0.719 *** (0.105)
DF (program-based move) 0.092 *** (0.019) 0.033 *** (0.007)
Age -0.026 *** (0.001) -0.009 *** (0.0002)
Race (= 1 if nonwhite) 0.261 *** (0.051) 0.084 *** (0.015)
Gender (= 1 if female) 0.456 *** (0.018) 0.146 *** (0.005)
Lag employment 1.621 *** (0.008) 0.558 *** (0.003)
% EMPLOYMENT -0.003 *** (0.001) -0.001 *** (0.0004)
% LESS THAN HIGH SCHOOL -0.002 ** (0.001) -0.001 ** (0.0003)
% HOME OWNERSHIP -0.0008 (0.0005) -0.0003 (0.0002)
% PUBLIC ASSISTANCE -0.002 ** (0.001) -0.001 ** (0.0003)
% POVERTY -0.002 *** (0.001) -0.001 *** (0.0002)
% FEMALE HEAD 0.0003 (0.001) 0.0001 (0.0003)
ACCESS -2.57E-07 (4.86E-07) -8.94E-07 (1.0E-06)
PER CAPITA CRIME -0.011 (0.019) -0.004 (0.006)
N 313,662
Coefficient
Variable (Standard Error) Marginal Effect
Constant -1.04 *** (0.056)
DF (program-based move) 0.085 *** (0.016) 0.030 *** (0.006)
Age -0.026 *** (0.001) -0.009 *** (0.0002)
Race (= 1 if nonwhite) 0.236 *** (0.050) 0.078 *** (0.015)
Gender (= 1 if female) 0.451 *** (0.018) 0.147 *** (0.005)
Lag employment 1.625 *** (0.008) 0.561 *** (0.003)
% EMPLOYMENT
% LESS THAN HIGH SCHOOL
% HOME OWNERSHIP
% PUBLIC ASSISTANCE
% POVERTY
% FEMALE HEAD
ACCESS
PER CAPITA CRIME
N 335,659
Year dummies were included but not reported.
* p < 0.10.
** p < 0.05.
*** p < 0.01.
Table 3. Mean Value of Neighborhood Characteristics
Before HOPE After HOPE
Neighborhood Characteristic VI Move VI Move
Percent employed 30.5 46.5
Percent with less than high school
degree 50.4 36.3
Percent owner-occupied housing 21.4 33.1
Percent on public assistance 22.9 16.8
Family poverty rate 40.1 37.2
Percent female-headed households 38.8 36.1
Crimes per capita 0.613 0.288
ACCESS 332,951 20,544
Public Public
Housing to Housing to
Neighborhood Characteristic Public Housing Voucher
Percent employed 39.3 50.7
Percent with less than high school
degree 44.1 31.8
Percent owner-occupied housing 17.7 41.9
Percent on public assistance 26.2 11.3
Family poverty rate 56.0 26.3
Percent female-headed households 41.2 33.2
Crimes per capita 0.393 0.220
ACCESS 36,587 10,479
Table 4. Probit Regression for Program-Based and Voluntary Moves
Variable Probit (Standard Error)
Constant -0.667 *** (.099)
Age -0.026 *** (.001)
Race (= 1 if nonwhite) 0.245 *** (.051)
Gender (= 1 if female) 0.444 *** (.018)
Lag employment 1.620 *** (.008)
DV (voluntary move) (a) 0.106 *** (.014)
DF (Program-based 0.139 *** (.020)
move) (a)
% EMPLOYMENT -0.003 *** (.001)
% LESS THAN -0.002 * (.001)
HIGHSCHOOL
% HOME -0.001 (.001)
OWNERSHIP
% PUBLIC -0.001 (.001)
ASSISTANCE
% POVERTY -0.002 *** (.001)
% FEMALE HEAD -0.0005 (.001)
ACCESS -2.71E-07 (4.87E-07)
PER CAPITA -0.007 (.019)
CRIME
N 311,209
Excluded category Public housing
nonmovers
Year dummies were included but not reported.
(a) Excluded category: no move.
* p < 0.10.
** p < 0.05.
*** p < 0.01.
Table 5. Difference-in-Differences Ordinary Least Squares Regression
Variable Coefficient (Standard Error)
Constant -0.006 (0.111)
DF (program-based 0.016 (0.013)
move)
Age 0.0002 *** (0.0004)
Race (= 1 if nonwhite) 0.100 * (0.058)
Gender (= 1 if female) -0.016 (0.017)
% EMPLOYMENT -0.0003 (0.001)
% LESS THAN -0.001 (0.0008)
HIGHSCHOOL
% HOME -0.0005 (0.0005)
OWNERSHIP
% PUBLIC -0.001 (0.001)
ASSISTANCE
% POVERTY 0.0003 (0.0007)
% FEMALE HEAD 0.0001 (0.001)
ACCESS 1.22E-07 (5.38E-07)
PER CAPITA 0.004 (0.022)
CRIME
N 4296
Year duminies were included but not reported.
* p < 0.10.
** p < 0.05.
*** p < 0.01.
Table 6. Probit Regression: Postmove Housing Tenure (Standard
Errors in Parentheses) (Dependent Variable: Quarterly Employment
Dummy)
Sample of
Nonmovers and
Public-to-
Variable Full Sample Public Moves
(1) (2)
Constant -0.843 *** (0.106) -0.903 *** (0.129)
Age -0.026 *** (0.0006) -0.026 *** (0.001)
Race (= 1 if nonwhite) 0.234 *** (0.051) 0.271 *** (0.054)
Gender (= 1 if female) 0.440 *** (0.018) 0.449 *** (0.019)
Lag employment 1.598 *** (0.008) 1.590 *** (0.009)
DF-PH 0.011 (0.030) 0.012 (0.030)
DF-HC 0.218 *** (0.024)
DV-PH 0.030 * (0.017) 0.033 * (0.018)
DV-HC 0.212 *** (0.019)
DC-PH 0.242 *** (0.045)
% EMPLOYMENT -0.002 (0.001) -0.002 (0.001)
% LESS THAN -0.001 (0.001) 0.001 (0.001)
HIGHSCHOOL
% HOME OWNERSHIP -0.0004 (0.0005) -0.00004 (0.0007)
% PUBLIC ASSISTANCE -0.0003 (0.001) 0.001 (0.001)
% POVERTY -0.001 * (0.001) -0.001 (0.001)
% FEMALE HEAD -0.0004 (0.001) -0.002 * (0.001)
ACCESS 7.33E-07 (4.97E-07) 9.11E-07 (6.03E-07)
PER CAPITA CRIME -0.006 (0.019) -0.035 (0.021)
N 313,662 265,448
Excluded category Public housing Public housing
nonmovers nonmovers
Sample of
Nonmovers and Sample of
Public-to- Public-to-
Variable Choice Moves Public Moves
(3) (4)
Constant -0.785 *** (0.106) -0.669 ** (0.362)
Age -0.025 *** (0.001) -0.034 *** (0.002)
Race (= 1 if nonwhite) 0.242 *** (0.052) 0.499 *** (0.189)
Gender (= 1 if female) 0.432 *** (0.018) 0.497 *** (0.057)
Lag employment 1.613 *** (0.009) 1.517 *** (0.023)
DF-PH -0.111 ** (0.045)
DF-HC 0.227 *** (0.025)
DV-PH
DV-HC 0.215 *** (0.020)
DC-PH
% EMPLOYMENT -0.003 ** (0.001) 0.0001 (0.003)
% LESS THAN -0.002 ** (0.001) 0.001 (0.003)
HIGHSCHOOL
% HOME OWNERSHIP -0.0004 (0.0006) 0.0002 (0.002)
% PUBLIC ASSISTANCE -0.002 * (0.001) 0.008 *** (0.003)
% POVERTY -0.0001 (0.001) -0.007 *** (0.002)
% FEMALE HEAD 0.0003 (0.001) -0.004 (0.003)
ACCESS 3.47E-07 (5.69E-07) 2.06E-06 (1.55E-06)
PER CAPITA CRIME -0.016 (0.021) 0.059 (0.058)
N 269,503 41,706
Excluded category Public housing Public-to-public
nonmovers voluntary
Sample of Public-to
Variable Choice Moves
(5)
Constant -0.217 (0.303)
Age -0.026 *** (0.002)
Race (= 1 if nonwhite) -0.090 (0.202)
Gender (= 1 if female) 0.430 *** (0.060)
Lag employment 1.616 *** (0.021)
DF-PH
DF-HC -0.012 (0.036)
DV-PH
DV-HC
DC-PH
% EMPLOYMENT 0.001 (0.003)
% LESS THAN -0.005 ** (0.002)
HIGHSCHOOL
% HOME OWNERSHIP -0.002 (0.001)
% PUBLIC ASSISTANCE 0.001 (0.003)
% POVERTY -0.003 (0.003)
% FEMALE HEAD -0.002 (0.002)
ACCESS -4.15E-06 **
(1.97E-06)
PER CAPITA CRIME 0.188 ** (0.073)
N 45,761
Excluded category Public-to-choice
voluntary
Year dummies were included but not reported.
* p < 0.10.
** p < 0.05.
*** p < 0.01.
Table 7. Marginal Effects for Dummies (Standard Errors in Parentheses)
Sample of Sample of
Nonmovers and Nonmovers and
Full Sample Public-to-Public Public-to-Choice
Moves Moves
Variable (1) (2) (3)
DF-PH 0.004 (0.010) 0.004 (0.010)
DF-HC 0.079 *** (0.009) 0.083 *** (0.009)
DV-PH 0.011 * (0.006) 0.011 * (0.006)
DV-HC 0.077 *** (0.007) 0.078 *** (0.008)
DC-PH 0.089 *** (0.017)
N 313,662 265,448 269,503
Excluded Public housing Public housing Public housing
category nonmovers nonmovers nonmovers
Sample of Public- Sample of Public-
to-Public Moves to Choice Moves
Variable (4) (5)
DF-PH -0.037 ** (0.015)
DF-HC -0.005 (0.014)
DV-PH
DV-HC
DC-PH
N 41,706 45,761
Excluded Public-to-public Public-to-choice
category voluntary voluntary
* p < 0.10.
** p < 0.05.
*** p < 0.01.
