|
Introduction
After its first 18 months, the Minnesota
Family Investment Program (MFIP) produced substantial effects
on the employment and earnings of single-parent, long-term
recipients in urban areas (Miller, et al. 1997). Subsequent
analyses revealed that the program had notably different effects
on recipients who were in public or subsidized housing at
program entry compared with those who were not. Specifically,
MFIP's impacts on employment and earnings were larger for
the former group. This paper presents MFIP's 18-month impacts
by housing status and examines several possible reasons for
the pattern of impacts.
The results indicate that public and subsidized
housing does provide benefits, such as residential stability,
that may encourage employment, but that these benefits are
unlikely to account for the pattern of MFIP’s impacts.
The weight of the evidence, although indirect, suggests that
another aspect of public and subsidized housing may be important.
The work disincentive created by the rent rule may have led
to a situation in which many residents in public and subsidized
housing were especially responsive to MFIP’s employment
incentives. The evidence on this issue is only suggestive,
however, highlighting the need for further research on the
interaction between public housing and welfare reform.
MFIP's Impacts
Tables 1
and 2
present MFIP's impacts on employment, earnings, and welfare receipt for
single-parent, long-term recipients in urban areas. These data were presented
in the 1997 interim report. Recall that the impact of the program is measured
as the difference in outcomes between the research groups. The full MFIP
program produced fairly substantial increases in employment and earnings
(shown in column 4 of Table
1). MFIP's financial incentives alone (column 6) also increased employment
rates somewhat but not earnings. MFIP also increased welfare receipt (Table
2). In the last quarter of follow-up, for example,
80.6 percent of the MFIP group received welfare, compared with 76.9 percent
of the AFDC group.
Tables 3 through 6 present impacts for this
sample estimated separately by housing status at random assignment.
Recipients who reported at program entry that they were living
in “public” or “subsidized” housing are
referred to as the assisted housing group, and all other recipients
are referred to as the unassisted housing group.[1]
Using these responses, 40 percent of the recipients are in
assisted housing and 60 percent are not. In addition, the
majority of recipients in assisted housing (80 percent) reported
that they were in subsidized housing, most likely meaning
that they were receiving Section 8 vouchers, rather than living
in a public housing project.
Tables 3
and 4
present employment and earnings impacts for the two groups.
Comparing the impacts of the full program (columns 4) shows
that MFIP's effects on employment and earnings were substantially
larger for those in assisted housing. In quarter 7, for example,
employment rates for those in unassisted housing were 47.1
percent for the MFIP group and 38.5 percent for the AFDC group,
for an impact of 8.7 percentage points (Table
4 ). In contrast, for those in assisted housing, 36.1
percent of the recipients in the AFDC group were employed,
compared with 60.1 percent of the MFIP group, for an impact
of 23.9 percentage points (Table
3 ). The earnings impacts also differ. MFIP's impact on
earnings in the last quarter is a statistically significant
$453 for those in assisted housing ($1,363 for the MFIP group
versus $910 for the AFDC group), compared with a statistically
insignificant $87 for those in the unassisted housing ($910
for the MFIP group versus $823 for the AFDC group). Most of
the quarterly differences in impacts between the two housing
groups and both of the differences for the summary measures
(covering quarters 2 through 7) are statistically significant.
The differences in the effects of the full
program appear to be driven by two factors--the employment
impacts of financial incentives alone and the earnings impacts
of adding mandatory services to the incentives. First, comparing
column 6 in both tables shows that MFIP's financial incentives
alone have virtually no effect on the employment rates of
the unassisted group but large effects for the assisted group.
Among those in assisted housing, for example, employment rates
in quarter 5 were 47.1 percent for the MFIP Incentives Only
group, compared with 30.9 percent for the AFDC group, for
an impact of 16.2 percentage points (Table
3). The corresponding impact for the unassisted group
is a statistically insignificant 2.2 percentages points. Second,
comparing column 8 in both tables shows that the earnings
impacts of adding mandatory services to the incentives are
very different for the two groups. Over the entire follow-up
period, for example, the impact on average earnings of adding
services to incentives was an insignificant $387 for the unassisted
group and a significant $1,507 for the assisted group.[2]
Tables 5
and 6
present impacts on welfare receipt. In contrast to the effects
on employment, MFIP's effects on welfare receipt are fairly
similar for the two groups. The most notable difference between
the two groups is the higher rates of welfare receipt for
those in assisted housing, among both the MFIP and AFDC groups.
For example, among those in unassisted housing (Table
6), 73.7 percent of the AFDC group received welfare in
quarter 7. For those in unassisted housing, 81.6 percent of
the AFDC group received welfare in quarter 7.
Since the employment and earnings impacts
were larger for those in assisted housing, one might have
expected that the increases in welfare receipt would be smaller
for this group. Two factors may account for why they are not.
First, as illustrated in the 1997 interim report, the difference
in benefit levels between MFIP and AFDC occurs only if the
recipient works, so more of the assisted group would be eligible
for MFIP’s higher benefits because more of them went
to work. Related to this, MFIP’s incentives are relatively
greater for part-time work. The fact that MFIP’s incentives
alone increased employment rates for the assisted group but
had no significant effect on earnings (see Table
3) suggests that many of those who went to work were working
part-time, a level at which their MFIP benefits were still
fairly large. Second, although MFIP produced a bigger impact
on average earnings during the follow-up period for the assisted
group ($2,041) compared with the unassisted group ($429),
this amounts to a difference of less than $100 only monthly
basis. Thus, the difference in impacts on welfare dollar amounts
would not be very large.
Based on conversations with officials in Minnesota
and findings from research on assisted housing, this paper
examines several possible explanations for the differences
between groups in the program’s impacts on employment
and earnings. First, recipients in assisted housing may differ
in many ways from those not in assisted housing, and the ways
in which they differ may be related to their ability to respond
to MFIP by getting and keeping a job. For example, the process
of applying for and obtaining housing benefits may require
a certain degree of motivation and persistence. Second, assisted
housing may provide benefits to recipients that aid in their
ability to hold a job, such as residential stability and the
ability to weather changes in income due to temporary job
loss. Third, those in assisted housing, specifically those
receiving vouchers, may live in areas closer to jobs, increasing
their ability to take advantage of MFIP's incentives and services.
Finally, the rent rules for assisted housing alter residents'
incentives to work, which may have implications for the effects
of MFIP's incentives.
One possible explanation to note up front
is that employment rates are lower for the control group in
assisted housing, compared with the control group not in assisted
housing. In quarter 5, for example, 30.9 percent of the assisted
group worked (Table
3), compared with 35.9 percent of the unassisted group
(Table
4). In this case, MFIP might produce bigger impacts for
the assisted group not because of a difference in the effect
of the program per se, but because the control group had lower
employment rates, making gains easier to achieve. However,
this factor probably explains only a small part of the impact
difference, since employment rates for the MFIP groups in
assisted housing are higher at the end of follow-up than those
for the MFIP groups not in assisted housing.
Finally, it is worth noting that at least
one other program has been found to have varying effects by
housing status. As part of the national JOBS evaluation, preliminary
estimates indicate that the program’s impacts at the
Atlanta, Georgia site on employment and earnings were larger
for recipients living in public housing, compared with those
not living in public housing (Riccio 1998). Research is currently
being conducted to examine possible explanations for this
difference.
Analysis
The Department of Housing and Urban Development
(HUD) provides rental assistance to nearly five million low-income
families. Assistance is typically provided either in the form
of residence in a government owned public housing development
or in the form of (Section 8) vouchers or certificates, which
residents use to subsidize rent on housing in the private
market. Over time, the percentage of those receiving assistance
through residence in a public housing project has fallen,
and they now represent less than a third of the assisted caseload
(Kingsley 1998). Like welfare, eligibility for assisted housing
is income-based. Unlike welfare, however, not all households
who apply receive subsidies. Most local housing authorities
have waiting lists, and the wait can range from months to
years.
The benefit provided by assisted housing derives
from the fact that a tenant's rent payment is based on her
income: typically, rent plus utility payments are not to exceed
30 percent of the household's income, where income includes
earnings and welfare payments. In subsidized housing, the
tenant pays 30 percent of her income, and the voucher covers
the rest of the rent up to a fair market value determined
by the local housing authority. In public housing, the government
owns the unit and collects the tenant's portion of the rent.
It is important to note that the MFIP housing
subgroups are defined using self-reported housing status.
Shroder and Martin (1996) use HUD administrative data to test
the accuracy of responses on the American Housing Survey and
find that many families misreport their housing status. For
example, they find that 11 percent of respondents who are
in assisted housing do not report themselves as such, and
20 percent of those who report they are in assisted housing
are not.
The potential for misreporting should be kept
in mind when interpreting the results, although it is not
likely to be a serious problem for our sample. First, our
sample consists of long-term welfare recipients, who arguably
are more aware than non-welfare families of the different
housing programs and the distinctions between them. In informal
interviews with several housing staff in Minnesota, most felt
that recipients would accurately report their housing status.
Second, the differences in impacts for the two groups suggest
that they represent two distinct groups. If a substantial
number of recipients were responding incorrectly, this would
attenuate the differences in impacts.
Characteristics
The first hypothesis
does not stem from research on housing, but from general research
on the employment effects of any type of program, such as
AFDC, that is voluntary. For almost any program in which individuals
choose whether to apply or participate, there are likely to
be important differences in the types of people who enroll
compared with those who do not. This may be especially true
for assisted housing. Applying for a housing voucher and finding
a landlord willing to accept it may require a certain degree
of motivation and persistence that does not exist among recipients
in private housing. The participation decision is also made,
to some extent, by program administrators and landlords, since
they often screen applicants for their desirability as stable
tenants. Thus, if members of the assisted group are more motivated
or job-ready than the unassisted group, then these differences,
both observable and unobservable, may account for MFIP's differential
effects.
Table
7 presents demographic characteristics of the two groups
when they entered the program. A smaller percentage of the
assisted group lives in Hennepin County (73.7 percent versus
84.0 percent). Recipients in assisted housing are somewhat
older than those in private housing, fewer are black, and
fewer have never been married. In addition, those in assisted
housing have older children; 31.7 percent of the assisted
group had children under age 3 when they entered the program,
compared with 39.4 percent of the unassisted group. Being
somewhat older and having older children are both factors
that may make the assisted group more able to work, or more
employable. Another factor in their favor is education. A
higher fraction of those in assisted housing have at least
a high school degree or GED, 71.7 percent of the assisted
group, compared with 62.9 percent of the unassisted group.
There are not big differences in recent employment
history between the two groups, as shown in the rows entitled
“Labor Force Status,” with the exception of average
hours worked.[3] Among those working when they entered
the program, 46.5 percent of those in assisted housing were
working fewer than 20 hours per week, compared with 34.9 percent
of the unassisted group. Those in assisted housing are also
more likely to have been currently enrolled, or enrolled in
the previous year, in education and training activities. For
example, 28.1 percent of the assisted group was enrolled in
education or training at random assignment, compared with
17.8 percent of the unassisted group. This difference in enrollment
is due primarily to the assisted group’s higher participation
in post‑secondary education. Finally, the assisted housing
group had been receiving welfare for a longer period of time
when they entered the program; 59 percent of the assisted
group had received welfare for five or more years, compared
with 50 percent of the unassisted group.
In addition to demographic characteristics,
respondents also filled out a Private Opinion Survey when
they entered the program, providing information on their attitudes
and opinions about welfare and work. Table
8 presents selected responses for the two groups.[4]
The first section of the table presents reasons provided for
not working either part- or full-time. Recipients in assisted
housing are less likely to cite “no way to get there
every day” and more likely to cite “too much to
do during the day” as reasons for not working part-time.
They are also more likely to cite “too many family problems”
as a reason for not working full-time.
In terms of employment expectations, those
in assisted housing seem to be less likely to take a job offered
under different circumstances. As shown under “Client-reported
employment expectations,” fewer of the assisted group
agreed that they would take a job that offered somewhat higher
income than welfare but consisted of work they did not like
or required occasional work at night. For example, only 38.5
percent of recipients in the assisted group would take the
job if it involved work they did not like, compared with 49.4
percent of those in the unassisted group. In addition, those
in assisted housing have somewhat higher reservation wages
(the minimum wage at which they would take a job). Finally,
fewer of those in assisted housing expected to be working
and more expected to still be receiving welfare one year after
entering the program.
The data show that the assisted group is somewhat
older, somewhat more educated, and has older children than
the unassisted group. Although these factors would seem to
suggest that they are more employable, there is no notable
difference in employment, with the exception of more part-time
work among those employed. In fact, the assisted group seems
to be somewhat less work oriented, expressing less of a preference
for work and higher reservation wages. This apparently inconsistent
pattern of differences is discussed in a later section. Here
we consider whether any of the observed differences can account
for MFIP's different impacts. If those in the assisted group
are more employable, for example, they may be more able to
respond to MFIP by going to work.
As indicated in the interim report, all impacts
are regression-adjusted to control for random differences
in background characteristics between the MFIP and AFDC groups.
For our purposes, testing for the effects of differences in
characteristics between the two housing groups requires augmenting
the regression adjustment equation. Specifically, the procedure
involves adding interaction variables to the model to account
for the possibility that the program affected subgroups differently.
For example, if the bigger impacts for the assisted housing
group derive from the fact that they are somewhat more educated,
coupled with the fact that the program had bigger impacts
for more educated recipients, then once we account for this
possibility there should be no differences by assisted housing
status. The regression model was expanded to include interactions
with each of the variables for which there were notable differences
between the groups: county of residence, race, age, education,
the presence of children under age 3, marital status, and
prior welfare receipt.
Table
9 presents the results. The first column of the table
shows the difference in the impacts of the full program
for the assisted and unassisted groups. In other words, if
MFIP’s impact on the percent employed in a given quarter
is 5 percentage points for the unassisted group and 12 percentage
points for the assisted group, then the difference in impacts
between the groups is 7 percentage points. The first column
shows that the impacts are consistently larger for the assisted
group. In quarter 3, for example, MFIP's impact on the percent
employed was 11.7 percentage points bigger for the assisted
group, and this difference is statistically significant. The
second column shows how this difference changes when we control
for the above-mentioned characteristics. Reading across the
rows shows that controlling for differences in characteristics
between the two groups does not change the basic story. For
example, the difference in impacts on the percent employed
during quarters 2 through 7 changes from 12.8 percentage points
to 11.7 percentage points.
[5] The results were similar when we also controlled
for two differences from the Private Opinion Survey measuring
attitudes towards employment (not reported).
The results suggest that the different impacts
are not due the fact that recipients in assisted housing differ
from their unassisted counterparts. This conclusion must be
qualified, however, given that we are not able to control
for differences in “unobservable” characteristics,
which are probably important. In fact, nonexperimental research
on the effects of program participation generally finds that
characteristics observable to the analyst do not capture all
of the differences between those who chose to participate
and those who did not.
Stability
Assisted housing also provides benefits that
may affect a recipient's efforts to find and keep a job. For
example, since rent is tied to income, those in assisted housing
are less likely to face the threat of eviction if they suffer
an unexpected job or income loss. In addition, in an effort
to attract landlords to the program, many voucher programs
require that recipients sign long-term leases. Although there
is no empirical evidence on this issue, it seems reasonable
that residential stability would encourage employment stability.
Assisted housing may also offer stability in a broader sense,
allowing residents to weather unexpected changes in income.
Although it is difficult to measure general
life stability, Table
10 presents data on residential stability. The first few
rows present the number of times recipients reported moving
during the two years prior to program entry. The differences
by housing status are striking. One quarter of recipients
in the unassisted group moved three or more times during the
two-year period, compared with 8 percent of the assisted group.
Recall that housing status is defined as of the baseline survey,
and we have no information on how long recipients had been
in assisted housing prior to random assignment. Thus, the
differences in stability may not be entirely due to assisted
housing.[6]
Information on mobility after random assignment
is also available from the 12- and 36-month surveys. These
data are shown for the control groups only, in order to present
the effects of assisted housing that are separate from any
effects MFIP might have on mobility. As shown in the table,
the post-program differences correspond with the pre-program
differences. Fewer of those in assisted housing at random
assignment moved during the subsequent year. In addition,
the movers seem to have moved for different reasons. For the
assisted group, a slight majority (52.9 percent) of the movers
gave “got improved housing” as the reason for moving,
compared with 34.1 percent of the unassisted group. There
were no notable differences in the percent of both groups
that cited eviction as the reason.
Data from the 36-month survey also show differences
in mobility, although not as dramatic. The 36-month survey
was administered to a subset of the full evaluation sample.
More information about the survey sample will be presented
in the final report.[7]
For both groups, mobility after three years is fairly high;
63.2 percent of the assisted group reported moving, compared
with 76.7 percent of the unassisted group. It is important
to note, however, that only 67 percent of those who reported
living in assisted housing at random assignment were still
living there three years later.
Accounting for differences in mobility is
similar to the method used to account for differences in characteristics
— including these variables in the regression-adjustment
equation. Mobility prior to random assignment was included
in the model for Table
9. As the table showed, the differences in impacts between
the assisted and unassisted groups remained after accounting
for background characteristics and mobility, indicating that
prior mobility does not explain the pattern of impacts.
Accounting for post-random assignment mobility
is more difficult, because MFIP itself may have affected it.
If MFIP affects mobility, and if mobility is associated with
post-random assignment employment and earnings, then including
mobility in the regression-adjustment model will give misleading,
or biased, estimates of the effect of MFIP, for both housing
groups. Separate analyses revealed that both of these conditions
hold. MFIP increased mobility somewhat, and higher earnings
during the follow-up period was associated with an increased
probability of moving. Thus, it is not possible to test whether
mobility after random assignment accounts for MFIP’s
differential effects by housing status.
Other Potential Benefits from Assisted Housing
Because vouchers allow tenants to rent units
that they might not afford otherwise, housing and neighborhood
quality may differ for those in assisted housing. If recipients
in assisted housing have a higher standard of living, they
may be more able to make the transition to work in response
to MFIP’s incentives and services. Much of the research
on housing quality looks at differences between public housing
and voucher housing among low-income families, rather than
the difference between assisted and unassisted housing among
welfare recipients. Nonetheless, it is informative. Newman
and Schnare (1993) find that families in public housing, compared
with families receiving vouchers, were more likely to live
in central city areas and in areas with lower neighborhood
quality (as rated by the families). In addition, the quality
of the housing itself was higher for the voucher families.
A recent study focused specifically on public housing. Currie
and Yelowitz (1997) find that families in public housing projects
suffer from less overcrowding than otherwise similar families
who are not in public housing and that their children are
less likely to have repeated a grade. Thus, public housing
allows families to occupy higher quality units they would
be likely to afford on the private market.
Table
11 presents data on potential benefits from assisted housing
as measured from the 36-month survey. To focus on the potential
effects of housing per se, housing status is defined as of
the three-year mark, rather than at random assignment. Recipients
who were in assisted housing at the time of the survey had
considerably lower monthly housing costs than their unassisted
counterparts ($210 versus $517). Also, the assisted group
appears to be materially better off, according to some survey
measures, although worse off according to others. For example,
22.8 percent of the assisted group reported that they missed
a rent payment within the last 12 months, compared with 35.4
percent of the unassisted group. In contrast, 55.2 percent
of the assisted group reported that they often do not have
enough money to make ends meet at the end of the month, compared
with 45.9 percent of the unassisted group.[8]
Finally, contrary to findings from other research, those in
assisted housing rate their neighborhoods as lower quality
than do those in unassisted housing; 33.3 percent of the assisted
group rated their neighborhoods as excellent or very good,
compared with 39.5 percent of the unassisted group.
Thus, although the pattern is not consistent,
assisted housing appears to provide some benefits that may
be conducive to employment. As with mobility during the follow-up
period, however, we are not able to test whether these differences
account for the program’s pattern of impacts, since they
are correlated with both the program and with employment and
earnings outcomes.
Location
Related to the ability to rent higher quality
housing, assisted housing may provide the ability to move
to suburban areas or areas with more job opportunities. It
is well known that employment opportunities for less‑skilled
workers in inner city areas have deteriorated over the past
two decades as employers have moved to the suburbs. Results
from a recent housing experiment suggest that the ability
to move to the suburbs has positive effects on families. Rosenbaum
(1995) presents findings from the Gautreaux experiment in
Chicago, in which families living in public housing projects
were given vouchers. One group was allowed to use the vouchers
to move to suburban areas and the other could use them to
move to other areas within the city. Parents who moved to
the suburbs, compared with those who moved elsewhere in the
city, had higher employment rates, and their children had
better schooling outcomes.
We are able to examine the effects of location
using address information collected for the 36-month survey
indicating where recipients lived at random assignment.
Figure
1 presents a mapping for the three urban counties of
the residential location of recipients by housing status.
Figure
2 presents this mapping in more detail for Hennepin County.
The boundaries delineated within each county represent census
tracts.[9] Neither
figure suggests that there are large differences in location
between the groups.
To test the effects of location, we match
tract-level data from the 1990 Census to the census tract
in which the recipient lives in order to determine the percent
of families in poverty and percent of adults employed in the
tract. If recipients in assisted housing live in areas closer
to jobs or areas that provide better access to jobs, for example,
through better transportation, this should be reflected in
lower poverty rates and higher employment rates in their immediate
neighborhoods. These two factors are accounted for by including
them in the regression adjustment equation, in the same manner
in which we accounted for mobility and demographic characteristics.
If MFIP’s effects are bigger for the assisted housing
group because they live in areas with better access to jobs
and can more easily respond to MFIP by working, then the assisted-unassisted
differential should diminish once these tract-level characteristics
are accounted for.
The results (not shown) indicate that these
factors do not explain the differential impacts by housing
status. As with the results shown in Table 9, the impact difference
remains after the adjustment. It is worth mentioning, however,
that recipients were more likely to have worked during follow-up
if more of their neighbors were working and that MFIP’s
effects on recipients’ employment were larger in areas
with higher overall employment rates. Thus, although location
does not explain the pattern of impacts by housing status,
it is important.[10]
Work Incentives
The final hypothesis relates to the work disincentive
created by assisted housing and how it may affect recipients’
responsiveness to MFIP’s incentives. Although there has
not been much empirical research on the effects of housing
assistance on labor supply, a recent paper by Gary Painter
(1997) finds that it does reduce labor force participation
rates among low-income single mothers. He estimates, using
nationally representative data, that adding housing benefits
to the entire welfare package (AFDC, Food Stamps and Medicaid)
reduces labor force participation by an additional 34 percent
over the effect of the welfare package alone.
Two aspects of assisted housing create a disincentive
to work. First, rent increases as earnings increase, so the
payoff to work is less for those in assisted housing. Second,
the housing subsidy, or the difference between the fair market
rent and what the tenant actually pays, alters the incentive
to work because the tenant has more income after rent payments
than she would if she were renting a comparable unit on the
private market. This “income effect” reduces the
incentive to work because the recipient of the subsidy can
work less and still maintain the same standard of living.
Figure
3 illustrates by showing monthly net income available
to a single mother with two children at different hours of
work per week. Net income is the sum of earnings, AFDC and
Food Stamps benefits, income from federal and state earned
income tax credits, less rent payments and federal and state
taxes. In the first case (left panel), she does not live in
assisted housing and is assumed to pay $500 per month in rent.
In the second case (right panel), she lives in assisted housing
and pays one third of her income in rent. The fact that net
income is always higher for the recipient in assisted housing
illustrates the potential for an income effect on work hours.
For example, because of the subsidy, the recipient in assisted
housing could work zero hours and have more net income ($547)
than if she were living in private housing ($269). In addition,
she would have higher net income working part-time and living
in assisted housing ($890) than working full-time while living
in unassisted housing ($841).
The work disincentives created by the rent
rule are illustrated by the change in income that results
from working more hours. For example, for a recipient in unassisted
housing, the gain from moving from no work to part-time work
is $439 ($708-$269) and the gain from moving to full-time
work is $572 ($841-$269). In contrast, for a recipient in
assisted housing the gains are smaller: $343 for part-time
work and $397 for full-time work. Thus, the payoff to work
is smaller for those in assisted housing: 22 percent smaller
for part-time work and 30 percent smaller for full-time work.
The calculations shown in the figure do not
account for two aspects of the housing program that may also
affect work incentives. First, the net income figures are
based on the assumption that rent is adjusted monthly in response
to income changes, when it is actually adjusted yearly. In
practice, then, since a resident could work for several months
before her rent is increased to match her higher earnings,
the work disincentives created by assisted housing may be
less than Figure
3 would suggest. However, a second factor suggests that
the disincentives might be greater than those shown in the
figure. In most urban areas, there are lengthy waiting lists
for assisted housing, and interviews with housing staff indicate
that Minnesota is no different. Housing markets in the urban
counties are very tight, and waiting lists range from six
months to two years. The possibility of losing housing benefits
and facing a lengthy waiting list may make some recipients
reluctant to work.[11]
Although federal law requires that residents be given a six-month
grace period after losing benefits, during which they can
be immediately reinstated if they lose their job, some residents
may not be aware of this entitlement.
The second issue surrounding the incentives
of housing is how they interact with MFIP's incentives.
Figure
4 shows monthly net income under MFIP and AFDC, illustrating
that MFIP's incentives are smaller for recipients in assisted
housing. For example, the extra income obtainable under MFIP,
relative to AFDC, from moving to part-time work is $112 ($820-$708)
for those not in assisted housing, compared with $77 ($967-$890)
in assisted housing. The same pattern holds for moving to
full-time work. MFIP’s incentives are smaller for those
in assisted housing because rent payments are based on earnings
plus welfare benefits. Rent will therefore be somewhat higher
for MFIP families since their benefit level is higher.
Assisted housing affects incentives to work
in ways that are more complicated than shown in a simple graph.
Nonetheless, it is probably safe to say that, on net, assisted
housing reduces the incentive to work. This is consistent
with Painter's empirical findings, mentioned earlier, and
with interviews with some housing staff indicating that the
common perception among recipients in assisted housing is
that working is risky and does not pay.
The incentives created by housing are relevant
for the following reason. Because of the work disincentive,
at program entry we might expect the assisted housing group
to contain a group of relatively employable recipients who
are not working but would be had they not been in assisted
housing. Recipients in this group might be especially sensitive
to an incentive like MFIP and might be more able to take advantage
of it by finding a job on their own. This is consistent with
data shown earlier, where the group in assisted housing looked
more employable at program entry, according to certain demographic
characteristics, but did not have higher employment rates
and expressed less of a preference for work. This might explain
the larger employment effects of MFIP's incentives. Working
in the opposite direction is the fact that MFIP's incentives,
relative to AFDC, are smaller for recipients in assisted housing,
suggesting that MFIP's effects should have been smaller for
this group. This is an important point to keep in mind, but
it does not explain the pattern of impacts. In addition, the
difference in MFIP's incentives for the two groups (Figure
4) is considerably smaller than the difference in work
incentives created by assisted housing itself (Figure
3), suggesting that the initial work disincentive effect
of housing may be a more powerful determinant of how individuals
respond to MFIP than the size of MFIP’s incentives.
This hypothesis cannot be tested directly
with the data used for the evaluation. As a rough test, however,
we can examine the pattern of impacts for subgroups. The incentives
hypothesis is that MFIP’s impacts are bigger for the
assisted group because it contains a group of employable recipients
who are not working and who might be especially sensitive
to MFIP’s incentives. Thus, the difference in MFIP's
impacts by housing status should be more pronounced for more
employable recipients. Although it is difficult to define
employability, one aspect of it is education level.
Table
12 presents the difference in impacts between the assisted
and unassisted groups. The left column show this difference
for recipients without a high school diploma or GED, and the
right column shows the difference for those with at least
a diploma or GED. In general, the differences are bigger and
more consistent for the more educated group. [12] For example, among those without
a diploma, MFIP's impact on the percent employed in quarter
6 was 3.2 percentage points higher for the assisted group,
compared with the unassisted group. Among those with a diploma,
MFIP's impact was 15.3 percentage points higher for the assisted
group. Thus, MFIP's impacts for the assisted housing group
are larger for the more employable subgroup, which is consistent
with the work incentives hypothesis. These results are only
suggestive, however, since education level is just one aspect
of employability.
Discussion
This
paper has raised several hypotheses to explain MFIP's larger
impacts for recipients in assisted housing. Some of them were
testable with these data and some were not. While we could
rule out differences in observable characteristics and residential
location across the groups, what we could not test with these
data are the effects of unobservable differences between the
groups and the effects of the general sense of stability and
other potential benefits provided by assisted housing. Each
of these factors may play a role. However, if the assisted
group overall was really more motivated or persistent than
the unassisted group, or if housing provided stability that
was conducive to employment, then those in assisted housing
should have had higher employment rates prior to random assignment.
Yet they did not.
One factor that could explain this pattern is the work
disincentives of assisted housing. Theory suggests, and empirical research
finds, that assisted housing creates a disincentive to work. In this case,
assisted housing might consist of more persistent individuals or it might
provide stability that encourages employment at the same time that it
discourages work because of its rent rules. On net, then, employment rates
would be no higher, and maybe even lower, for the assisted group, and
the assisted group might be particularly likely to respond to MFIP’s
incentives and marketing.
The story behind the differences by housing
status is most likely a complicated one, involving multiple
factors acting in different ways. It remains possible that
the effects of housing may not be due to the benefits of assisted
housing, but to unobservable differences in the types of people
across groups and to the fact that the work disincentive may
have created a situation in which many in the assisted group
were especially responsive to MFIP’s employment incentives.
References
Currie, J., and A. Yelowitz. 1997. "Are
Public Housing Projects Good for Kids?" Cambridge, MA:
NBER Working Paper No. 6305.
Kinglsey, G.T. 1998. "Federal Housing
Assistance and Welfare Reform: Uncharted Territory" Washington,
D.C.: The Urban Institute.
Miller, C., V. Knox, P. Auspos, J. Hunter-Manns,
and A. Orenstein, 1997. Making Welfare Work and Work Pay:
Implementation and 18-Month Impacts of the Family Investment
Program, New York: MDRC.
Newman, S., and A. Schnare. 1993. "Last
in Line: Housing Assistance for Households with Children"
Housing Policy Debate 4(3): 417-55.
Painter, G. 1997. "Low Income Housing
Assistance: Its Impact on Labor Force and Housing Program
Participation" unpublished paper, University of Southern
California, School of Public Administration.
Riccio, J. 1998. "Emerging Findings
from a Study in Progress: The Employment Potential of Welfare
Recipients Living in Public and Assisted Housing" unpublished
paper, New York: MDRC.
Rosenbaum, J. 1995. "Changing the
Geography of Opportunity by Expanding Residential Choice:
Lessons from the Gautreaux Program" Housing Policy Debate
6(1): 231-69.
Shroder, M., and M. Martin. 1996. "New
Results from Administrative Data: Housing the Poor, or, What
They Don't Know Might Hurt Somebody," Paper presented
at the 1996 Mid-Year meeting of the American Real Estate and
Urban Economics Association, May 29, 1996.
 Notes
[1]
The question on the Background Information Form read "What
is your current housing status," and possible responses
were "Public Housing," "Subsidized Housing,"
"Emergency/Temporary Shelter," and "None of
the Above." Among those not in assisted housing, 95 percent
reported "None of the Above."
[2]
Impacts were estimated separately for public versus subsidized
housing recipients. The pattern of results was similar for
both groups, although fewer of the impacts were statistically
significant for those in public housing, owing to small sample
sizes.
[3]
Data on prior employment and earnings from the Unemployment
Insurance administrative records also showed little difference
in prior employment.
[4]
Due to nonresponse, the sample sizes for the Private Opinion
Survey are smaller than for the demographic data. The extent
of nonresponse does not differ between the two housing groups.
[5]
Similar results were obtained for the effects of the financial
incentives alone and of adding mandatory services to the incentives.
The differences reported in Table 9 differ slightly from the
differences obtained from Tables 3 and 4, because the latter
were estimated using split samples, and the former were estimated
using a combined sample.
[6]
Of course, it is also possible that more stable individuals
apply for and get assisted housing, so that lower mobility
is not due to assisted housing per se.
[7]
All analyses using the 36-month sample use appropriate weighting
procedures to adjust for the oversampling of families eligible
for the child outcomes section of the survey, i.e., those
with at least one child whose age falls within a certain range.
[8]
These contradictory findings indicate that these two questions
may not be measuring the same factor, or that housing status
affects perceptions of well-being.
[9]
A census tract is a small subdivision of a county created
to include several thousand people.
[10]
As with mobility, however, it is important to consider that
more employable individuals may choose to live in higher employment/lower
poverty areas, in which case the bigger impacts of the program
in higher employment areas are not due to location per se,
but to differences across individuals.
[11]
The probability of losing all benefits may be low, since it
requires fairly high monthly income. At a fair market rent
of $666, for example, housing benefits become zero if the
recipient's monthly income exceeds $2,220.
[12]
Note that this finding is not inconsistent with the fact that
education level did not explain the pattern of impacts, as
shown in the section on characteristics. That test asked whether
MFIP affected more educated recipients differently and whether
that explained why MFIP affected assisted housing recipients
differently. The test in this section, on the other hand,
asks whether the effects of MFIP for those in assisted housing,
compared with those not in assisted housing, are different
|
