Skip navigation

Placement of Homeless Persons with Severe Mental Illness in Supportive Housing, University of Pennsylvania, 2002

Download original document:
Brief thumbnail
This text is machine-read, and may contain errors. Check the original document to verify accuracy.
Culhane.qxd

5/28/02

10:26 AM

Page 107

·

Housing Policy Debate
Volume 13, Issue 1
© Fannie Mae Foundation 2002. All Rights Reserved.

107
107

Public Service Reductions Associated with
Placement of Homeless Persons with Severe
Mental Illness in Supportive Housing
Dennis P. Culhane, Stephen Metraux, and Trevor Hadley
University of Pennsylvania

Abstract
This article assesses the impact of public investment in supportive housing for homeless persons with severe mental disabilities. Data on 4,679 people placed in such housing in New York City between 1989 and 1997 were merged with data on the utilization
of public shelters, public and private hospitals, and correctional facilities. A series of
matched controls who were homeless but not placed in housing were similarly tracked.
Regression results reveal that persons placed in supportive housing experience marked
reductions in shelter use, hospitalizations, length of stay per hospitalization, and time
incarcerated. Before placement, homeless people with severe mental illness used about
$40,449 per person per year in services (1999 dollars). Placement was associated with a
reduction in services use of $16,282 per housing unit per year. Annual unit costs are
estimated at $17,277, for a net cost of $995 per unit per year over the first two years.
Keywords: Homelessness; Housing

Introduction
Placing homeless persons with severe mental illness (SMI) into subsidized permanent housing with social service support promises to substantially reduce the demand for shelter among those placed. This
housing provides a more humane alternative to living on the streets
and in shelters, and providers report retention rates in such housing
to be upwards of 70 percent in the first year after placement. However,
little empirical evidence has been gathered to quantify the degree to
which supportive housing supplants shelter use among the formerly
homeless with SMI. Furthermore, it can similarly be assumed that
homeless persons with SMI, once placed in supportive housing, reduce
their use of acute psychiatric and medical services, and are arrested
and incarcerated less often. However, such assumptions are somewhat
more tenuous, and a similar dearth of empirical evidence exists both on
the demand for these services among homeless persons with SMI and
on the impact of supportive housing on this level of demand.
The study reported here examines service use by formerly homeless
persons with SMI before and after being placed into New York/New

Culhane.qxd

5/28/02

108

10:26 AM

Page 108

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

York (NY/NY) housing, a large housing program in New York City
(NYC). Administrative data from public health, psychiatric, criminal
justice, and shelter service providers are used to assess the aggregate
level of service demand, pre- and postintervention, for the study group
and for a matched set of controls. The extent to which reductions in
services are present and attributable to NY/NY housing placement is
assessed, and the degree to which service reductions offset supportive
housing costs is measured.

Background
In 1990, New York State (NYS) and NYC agreed to jointly fund and
develop 3,600 community-based permanent housing units for homeless
persons with SMI under what became known as the New York/New
York Agreement to House the Homeless Mentally Ill (Hevesi 1999;
Kennedy 1995, 1997).1 This initiative was in response to problems with
homelessness and community mental health services that were perceived to have reached crisis proportions in NYC. The NY/NY agreement was designed to target those who were among the most chronic
and difficult to serve among the homeless population and to ease
demands on public shelter and psychiatric treatment services.
The agreement provided housing and psychosocial services in a variety
of configurations collectively known as NY/NY housing. There are two
general models: The first, supportive housing, includes scattered-site
housing with community-based service support and single-room occupancy (SRO) housing (independent housing linked to either communitybased or site-based service support). The second, community residence
facilities, includes community residences, long-term treatment facilities,
and adult homes (Center for Urban Community Services 1995; Lipton
et al. 2000). In general, supportive housing emphasizes “normality”
in housing in terms of separating services from housing arrangements
and giving tenants a choice in their housing arrangements and mental
health service regimens. By contrast, community residences take a more
clinical approach that integrates housing and services delivery by having services available on site and participation mandated by the residence agreement. Supportive housing maintains that such housing is
appropriate for persons with mental illness regardless of the severity of
impairment, while the community residence model places people in
increasingly less restrictive living arrangements as they progress

1

This initiative is also now referred to as NY/NY I, since a second initiative to provide
additional units under a similar state/city–financed structure (NY/NY II) was passed in
1999.

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 109

Public Service Reductions Associated with Supportive Housing

109

through their treatment regimens (Bebout and Harris 1992; Carling
1993).
To be eligible for this housing, tenants must have a diagnosis of SMI
and have been recently homeless in shelters or on the streets. After
going through an application and assessment with the NYC Human
Resources Administration (HRA) to determine NY/NY housing eligibility, the prospective tenant then applies through one of the nonprofit
agencies that administer the actual units funded under the agreement.
Thus, NY/NY eligibility, housing availability, agency eligibility guidelines, and tenant preference all factor into the placements provided
under the agreement.

Literature review
Studies that focus on supportive housing interventions for homeless
persons with mental illness consistently find high rates of retention in
these programs. Lipton, Nutt, and Sabatini (1988) followed 49 homeless persons with mental illness, half of whom were provided program
housing. After one year, they found that 69 percent of the experimental
group were living in permanent housing, as opposed to 30 percent of
the control group. Drake et al. (1997) similarly report improved housing outcomes for a group of dually diagnosed2 homeless persons who
were provided residential treatment, compared with a control group
given standard treatment. Caton et al. (1993) and Murray et al. (1997)
report high rates of housing retention for participants in transitional
and/or continuum-model programs, although none of the studies
included comparable control groups.
In the most comprehensive review of supportive housing studies to
date, Ridgway and Rapp (1998) report that supportive housing for
homeless persons with mental illness reduced homelessness and improved housing stability among program participants. Research from a
McKinney Demonstration Program (Center for Mental Health Services
1994), in which the National Institute of Mental Health sponsored five
supportive housing projects in four cities, found increased rates of stable housing among the experimental groups—formerly homeless persons with mental illness who received supportive housing and case
management services—compared with similar groups of controls who
were provided with standard treatment services (Shern et al. 1997).

2

In this article, “dually diagnosed” refers to comorbid diagnoses of serious mental illness and substance abuse disorder.
Housing Policy Debate

Culhane.qxd

5/28/02

110

10:26 AM

Page 110

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

In the San Diego McKinney project, Hurlburt, Wood, and Hough (1996)
report on 362 persons who were homeless and severely mentally ill and
who were randomly assigned to four groups that varied Section 8 rental
subsidies and case management services. They found that Section 8
manipulation had a dramatic impact on subsequent housing stability,
but that enhanced case management manipulation had no significant
impact. Only 30 percent of the study participants who did not receive a
rent subsidy achieved stable independent living, compared with 57 percent of those who did.
Both Goldfinger et al. (1999) and Dickey et al. (1996) report on an
initiative that provided two types of housing in Boston for homeless
persons with mental illness: independent apartments with communitybased services and so-called evolving consumer households, where the
tenants lived communally and with gradually diminishing levels of staff
assistance. Three-quarters of the total subjects were stably housed at
the end of the 18-month follow-up period. The subjects in the group
homes had fewer days homeless than the supported housing group, but
otherwise no significant differences in housing outcomes or services use
were found (Dickey et al. 1996; Goldfinger et al. 1999).
Three studies have looked at the housing provided by the NY/NY agreement. Lipton et al. (2000) found that after one, two, and five years,
75 percent, 64 percent, and 50 percent of the almost 3,000 persons
placed had remained in the program across all types of NY/NY housing
configurations. Tsemberis (1999) and Tsemberis and Eichenberg (2000)
have also found high rates of housing retention by NY/NY recipients,
but in addition, found that tenants of one supported scattered-site
housing program affiliated with NY/NY had a substantially higher
retention rate after five years (88 percent) than other NY/NY supportive housing programs (55 percent).
High tenant-retention rates among housing interventions regardless
of the particular configuration of services and housing have been a
common finding in the housing programs examined so far. However,
these similar outcomes belie the disparate costs involved in the two
approaches. Community residence models, with their incorporation of
site-based staff and services that work exclusively with the tenants,
have substantially higher associated service costs than supportive housing models; these decouple residency and services to a greater extent,
and their tenants make greater use of existing community services.
Persons with SMI could also be expected to reduce their use of hospital
services following a housing placement, because persons who are receiving services would be in a better position to engage in regular outpatient regimens that could prevent the need for hospitalization.
Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 111

Public Service Reductions Associated with Supportive Housing

111

Furthermore, if they are hospitalized, access to housing and support
could reduce the length of stay in a hospital. In a study of public hospital records in NYC, Salit et al. (1998) found homelessness to be associated with substantial excess stays and costs per hospital stay. Lewis and
Lurigio (1994), in a study of state hospital patients living in Chicago,
found that when poor persons with mental illness seek psychiatric hospitalization, they often do so more as a short-term housing arrangement than for psychiatric reasons. In a review of the literature,
Rosenheck (2000) found that enhancing services, either housing or case
management, can generate reductions in the use of inpatient mental
health services, especially among heavy hospital users. These reductions, however, may be at least partially offset by increased use of outpatient services (Averyt and Kamis-Gould 2000) and support services
such as case management, which are needed to effect such inpatient
reductions (Rosenheck 2000).
This leads to the question of whether providing a service such as supportive housing is cost-effective in reducing homelessness among persons with mental illness. Rosenheck (2000) found that for all but the
heaviest service users, enhanced interventions cost more than the savings they generate. However, studies of reductions in service utilization
and associated cost savings have typically focused on one type of service
and on a single service system and have not integrated multiple providers and multiple systems. Integrating costs accrued by homeless
persons across multiple providers in such systems as shelters, mental
health services, medical care, and criminal justice would allow for a
more comprehensive assessment of the cost of homelessness from
which to estimate savings. By tracking people across multiple systems,
the estimated public expenses associated with homelessness would
likely increase, as would the estimated reductions in service use following the receipt of targeted housing. Thus, greater cost-effectiveness
may be demonstrated.

Data and methods
Data sources
The data used in this study come from administrative databases maintained by eight different agencies. These databases are collected in
computerized management information systems and track service utilization over time. As such, they represent comprehensive banks of
data on users, both their characteristics and their patterns of use.
Because these databases contain client identifiers, they can be linked
across systems to identify how services received through one system
may affect services in others. Administrative databases are the only
Housing Policy Debate

Culhane.qxd

5/28/02

112

10:26 AM

Page 112

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

practical means of obtaining information on a large number of homeless persons over an extended period of time and with accurate data on
service consumption across multiple systems (Culhane and Metraux
1997).
Databases used for this analysis come from the following sources:
1. NYC HRA, with records for 4,679 persons either placed in housing
developed under the NY/NY agreement or deemed eligible for
NY/NY housing and placed in community-based housing. The database includes demographic and identifying information, as well as
the date and type of housing for placements through 1997. No
information on the duration of discrete NY/NY placements is
available.
2. NYC Department of Homeless Services (DHS), with records for all
shelter users and shelter use since 1986 for its single-adult shelter
network.
3. NYS Office of Mental Health (OMH), with a database of lifetime
records of inpatient stays in the state psychiatric hospital system
for anyone who was an inpatient in a state hospital from 1990
through 1996.
4. NYS Department of Health, Office of Medicaid Management (hereafter referred to as Medicaid), with records of Medicaid-reimbursed
inpatient and outpatient health care claims for persons with records
of shelter use and/or NY/NY housing placements for the years 1993
through 1997.
5. NYC Health and Hospitals Corporation (HHC), with records of
inpatient stays in municipal hospitals between 1989 and 1996 for all
persons with a DHS shelter record.
6. U.S. Department of Veterans Affairs (VA), with records of inpatient
stays in the VA hospital system between 1992 and 1999 for all persons with records of DHS shelter utilization or NY/NY placement.
7. NYS Department of Correctional Services (NYSDOCS), with a database on state prison utilization for persons with a NY/NY housing
placement, and a set of control observations selected from the DHS
shelter system. Data used in this study were from 1988 through
April 15, 1997.
8. NYC Department of Corrections (NYCDOC), with a database on
NYC jail utilization for persons with a NY/NY housing placement,
Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 113

Public Service Reductions Associated with Supportive Housing

113

and a set of control observations selected from the DHS shelter system. Data used in this study were from 1988 to 1999.
These databases were merged by matching five common identifiers:
first and last names, sex, date of birth, and Social Security number.
Segments of the first four identifiers were combined to create a unique
identifier that was used to match cases across databases. Also, Social
Security numbers (when available) were used to provide additional
matches when the other identifiers were missing or contained erroneous data.3

Matched control groups
By comparing each individual’s history of service use in the two-year
periods immediately before and after his or her NY/NY placement, it is
possible to estimate the changes in service use for persons with NY/NY
placements across these seven service systems. In addition, each person
placed in NY/NY housing was matched to an individual control observation with similar characteristics to assess service use in the absence of
a supportive housing placement. Because of the difficulty in consistently pairing case (NY/NY) and control observations with similar preintervention service use patterns across the seven service systems,
different control groups were used to analyze different service systems.
Appendix A contains a more detailed overview of the sampling frames
for the respective control groups.
To construct the matched-pair control group for each analysis, the following criteria were used to select observations for the control groups
based on similarities with specific control observations:
1. Demographics. Gender, race (black/nonblack), and age. Ages of
those in the control pool are within five years of the case.
2. Indicators of mental illness and substance abuse. For the DHS
control group, these indicators are based on data from DHS records
and reflect the assessments of DHS social service staff and selfdisclosure by shelter users at the intake interview (no standardized
criteria for determining mental illness or substance abuse problems
are used). For the OMH, HHC, Medicaid, and VA control groups,
these indicators are based on Diagnostic and Statistical Manual
of Mental Disorders, 4th Edition, diagnoses that accompany
hospitalizations.
3

Further details on this process are available from the authors.
Housing Policy Debate

Culhane.qxd

5/28/02

114

10:26 AM

Page 114

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

3. Similar service use for the two-year period up to the NY/NY placement date, based on the number of stays and days spent in service
facilities (shelter, hospital, or prison) during the two-year intervention period, and the length of time between the last service use and
NY/NY placement.
This case-control matching process has two parts. First, matching on
elements of the first two groups of criteria (demographics and diagnosis
indicators) limits the numbers of potential matches between each case
observation and the pool of controls. Then, for each match, using the
case observation’s NY/NY placement date (which varies for each observation) as a surrogate intervention date for each potential control
observation means that the control observation with the most similar
pattern of preintervention service use (i.e., days and episodes for the
two-year preintervention period) is selected to pair up with each case.
For each case-control pair, the case observation’s NY/NY placement
date represents the intervention point that separates pre- and postintervention periods for the case and control observations.

Analysis methods
Each analysis follows a parallel set of procedures. First, descriptive statistics that facilitate comparisons of raw pre- and postintervention service use among the aggregate group with NY/NY housing placements are
provided. Second, descriptive statistics on pre- and postintervention
service use are provided for the case and control groups, with paired
comparison t tests used to assess whether the differences between and
within groups across intervention periods are statistically significant.
Following these two analyses, the effect of a NY/NY housing placement
on the reduction in postintervention service use, measured in days, is
estimated with multivariate least squares regression models, using a
generalized estimating equations (GEE) methodology. With its use of
maximum likelihood estimation and an iterative generalized least
squares algorithm, GEE can accommodate nonindependent observations such as matched pairs. Such a data structure normally violates
the assumptions associated with ordinary least squares (OLS) regression, but this approach corrects the attenuated standard error values
that would otherwise result from using OLS regression (Allison 1999).4
In each regression model, the dependent variable is the difference, for
each observation, in the number of days accrued in each of the service
systems across the pre- to postintervention periods. The covariate of

4

Regression analyses were computed using GENMOD with the REPEATED option in
SAS statistical software, version 8.1.

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 115

Public Service Reductions Associated with Supportive Housing

115

primary interest estimates the effect of getting a NY/NY placement,
with all other factors held constant. Along with this NY/NY covariate,
the control variables in the models are as follows:
1. The variables used to match the control groups
2. When applicable, a set of dichotomous variables to control for the
year of NY/NY placement
3. A set of measures for preperiod service use, including service
episodes, service days consumed, and (when available) cost of
services5
4. Measures of preperiod DHS shelter use

Results
System-specific effects
Use of DHS shelter services. Almost three-quarters of those placed in
NY/NY housing have some record of having stayed in a DHS shelter at
some point between 1987 and 1999.6 These 3,365 persons with DHS
and NY/NY records were matched with control observations from the
NYC DHS single-adult shelter system, and the resulting dataset, with
3,338 matched pairs, provides the basis for a case-control comparison of
shelter usage for two-year periods before and after the NY/NY intervention point. A total of 27 case observations (0.8 percent) could not be
matched with a control observation.
The descriptive results (table 1) show a dramatic 85.6 percent pre/post
placement decline in the mean number of shelter days used by persons
with NY/NY placements, from 137.0 days per placement to 19.8 days
5

Measures of preperiod use of services control for two phenomena: first, the prerequisite that higher degrees of preintervention services must have higher differences in
levels of pre/post use of services and second, the effects of prior use of services on the
likelihood of engaging in subsequent use of services. These anticipated effects run
counter to each other, since the former would associate higher preperiod use of services with greater reductions in postintervention use, and the latter would associate
preperiod use of services with lower or negative differences in pre/post use of services.
While this could lead to difficulty in interpreting the services use coefficients, it should
control for these effects when considering the effects of NY/NY placement.

6

Of the 4,679 persons with NY/NY housing placements, 3,365 (71.9 percent) also had
DHS shelter records, even though all of them, by NY/NY eligibility criteria, must have
been homeless before housing placement. Of those without shelter records, some may
have used shelters not covered by the DHS database (approximately 20 percent of all
NYC shelter beds), some may have used shelters outside NYC, and some may have
stayed exclusively in nonshelter (street) arrangements during their periods of
homelessness.
Housing Policy Debate

Culhane.qxd

5/28/02

116

10:26 AM

Page 116

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

per placement. In the case-control comparison, the unadjusted case
group decline (85.6 percent) is consistent with that of the entire NY/NY
group and far outpaces the 6.4 percent unadjusted decline experienced
by the controls. The pre/post declines for both groups are statistically
significant (at p < 0.01) using paired-comparison t tests. In the preintervention period, the mean per placement number of shelter days used
among the cases is heavier (p < 0.0001) among the case group, but this
relationship inverts in the postintervention period (p < 0.0001) as the
controls become, on average, the heavier shelter users.
In the regression model shown on table 2, the NY/NY placement is still
associated with a 115.3-day reduction in shelter days used from the preto the postintervention period (95 percent confidence interval [CI],
107.7 to 123.0 days) after controlling for other factors, especially heavy
shelter use.7 Taking this 115.3-day reduction and averaging it out over
all 4,679 persons with NY/NY placements yields an estimated reduction
per NY/NY placement of 82.9 shelter days (95 percent CI, 77.4 to
88.5 days) over the two-year period.8 This represents an adjusted reduction of 60.5 percent, compared with the average preintervention shelter
usage by the NY/NY placements (137.0 from table 1).
Use of OMH inpatient state psychiatric hospital services. Of the 2,396
persons receiving a NY/NY placement from 1992 to 1994, 897 (37.4 percent) had some record of an inpatient OMH state hospital stay. Of this
subgroup, 630 observations also had a record of DHS shelter use and
were matched with DHS controls. These 630 case observations provided the basis for the OMH case-control analysis, with 570 (90.5 percent) of these observations matched with control observations selected
from DHS shelter users.

7

Heavy shelter use is interpreted as a combination of two covariates in the model,
“shelter days accrued” and “any shelter use.” These coefficients have opposite signs,
meaning that the value of preintervention shelter days is associated with decreased
reductions in the number of postintervention shelter days with few preintervention
shelter days accrued, and increased reductions with many preintervention shelter days
accrued (see footnote 4). Thus, if an observation had five preintervention shelter days,
the combined coefficient of the coefficients shelter days accrued and any shelter use
(0.75  5 – 34.48) would be a postperiod increase of 30.73 shelter days used (all else
held equal), while a preintervention stay of 100 days (0.75  100 – 34.48) would lead to
a combined coefficient associated with a decrease of 40.52 shelter days used. Equilibrium here would be at 46 days.

8

This average is computed by dividing estimated aggregate reduction in shelter days
attributed to NY/NY (115.3  3,365) by the total number of NY/NY placements (4,679).

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 117

Public Service Reductions Associated with Supportive Housing

117

Table 1. Shelter Days Consumed by Persons in NY/NY Housing and Controls
in the Two-Year Periods before and after the NY/NY Intervention
NY/NY
NY/NY
Controls
(Total 1989–97) (Matched Pairs) (Matched Pairs)
N
Total service users

4,679
3,365

3,338
3,338

3,338
3,338

Pre-NY/NY intervention (two years)
Total persons with shelter records 2,786 (59.5%)
Total days sheltered
641,171
Mean days (all persons)
137.0
Mean days (shelter users)
230.2

2,750 (82.4%)
636,319
190.6
231.4

2,265 (67.9%)
544,700
130.9
240.5

776 (23.2%)
91,751
27.5
118.2

1,754 (51.4%)
408,883
122.5
233.0

Post-NY/NY intervention (two years)
Total persons with shelter records
Total days sheltered
Mean days (all persons)
Mean days (shelter users)

782 (16.7%)
92,421
19.8
118.2

Note: Between the NY/NY and control groups, paired-comparison t tests assessing difference yield,
for preintervention differences in shelter days, t = 27.3 (3,337 df and p < 0.0001), and for postintervention differences, t = –26.2 (3,337 df and p < 0.0001).
Using paired-comparison t tests, pre/post differences yield, within the NY/NY group, t = 46.04
(3,337 df and p < 0.0001), and within the control group, t = 2.6 (3,337 df and p < 0.01).

Table 2. Regression Model Estimating Effects on Changes in Shelter Days
Used in the Two-Year Periods before and after the NY/NY Intervention
(N = 3,338 Matched Pairs)

Covariate
Intercept
Received NY/NY placement
Shelter days accrued in two-year
preintervention period
Any shelter use in two-year preintervention period
NY/NY placement in 1996–97
NY/NY placement in 1994–95
NY/NY placement in 1992–93
NY/NY placement before 1992
Age at NY/NY placement
Male
Black race
DHS mental illness indicator
DHS drug use indicator

Parameter
Estimate
(Days Saved)

Lower
(95%) CI

Upper
(95%) CI

–44.13***
115.33***
0.75***

–61.35
107.66
0.72

–26.92
123.01
0.78

–34.48***

–41.77

–27.19

–24.83***
–9.54
3.70

–36.32
–19.97
–6.35
Reference Category
–0.12
–0.42
–13.35***
–20.26
1.97
–4.72
–8.14*
–15.06
0.02
–7.04

–13.34
0.89
13.76
0.19
–6.44
8.67
–1.21
7.09

*p < 0.05. **p < 0.01. ***p < 0.001.

Housing Policy Debate

Culhane.qxd

5/28/02

118

10:26 AM

Page 118

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Descriptive results of state hospital use, shown in table 3, show large
reductions in pre/post state hospital use, measured in days, among the
total NY/NY group (59.9 percent reduction), as well as among the more
restricted case group (57.0 percent reduction). Looking at the case-control comparison, the preintervention state hospital use, measured in
days, is (by design) very similar (i.e., with statistically nonsignificant
differences). By contrast, the difference in postintervention state hospital use is both substantial and statistically significant (p < 0.0001).
Comparing within groups, the NY/NY group shows significant pre/post
reductions in state hospital use (p < 0.0001), while the reductions in
state hospital days used are nonsignificant for the control group. Further, for the NY/NY group, far fewer persons experienced postintervention hospital episodes than in the control group, and the mean number
of hospital days per hospitalized person also declined after the intervention. While persons hospitalized also declined in the latter time period
for the control group, the average number of days hospitalized
increased substantially for the control group.
Table 3. OMH State Hospital Days Consumed by Persons in NY/NY
Housing and Controls in the Two-Year Periods
before and after the NY/NY Intervention
NY/NY
NY/NY
Controls
(Total 1992–94) (Matched Pairs) (Matched Pairs)
N
Total service users

2,396
897

570
570

570
570

Pre NY/NY intervention (two years)
Total persons hospitalized
634 (26.4%)
Total days hospitalized
137,215
Mean days (all persons)
57.3
Mean days (hospital users)
216.4

406
78,250
137.3
192.7

406
78,940
138.5
194.4

Post NY/NY intervention (two years)
Total persons hospitalized
Total days hospitalized
Mean days (all persons)
Mean days (hospital users)

240
33,623
59.0
140.1

335
74,869
131.4
223.5

353
55,070
23.0
156.0

Note: Between the NY/NY and control groups, paired-comparison t tests assessing difference yield,
among preintervention state hospital days, t = –1.8 (569 df and p = 0.07), and for postintervention differences, t = –7.7 (569 df and p < 0.0001).
Using paired-comparison t tests, pre/post differences yield, within the NY/NY group, t = 9.3
(569 df and p < 0.0001), and within the control group, t = –1.8 (569 df and p = 0.37).

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 119

Public Service Reductions Associated with Supportive Housing

119

A multivariate regression model (table 4) shows that, holding other
factors constant, a NY/NY placement is associated with a statistically
significant estimated reduction of 75.3 days (95 percent CI, 55.7 to
95.0 days). Averaging this adjusted reduction for the case group across
all the 2,396 NY/NY placements from 1992 to 1994 yields an estimated
reduction of 28.2 days per NY/NY placement (95 percent CI, 20.8 to
35.6 days).9 Compared with the 57.3 days of mean preintervention state
hospital use by the NY/NY group (table 3), this reflects a 49.2 percent
adjusted reduction.
Table 4. Regression Model Estimating Effects on Changes in State Hospital
Days Used in the Two-Year Periods before and after the NY/NY Intervention
(N = 570 Matched Pairs)

Covariate
Intercept
Received NY/NY placement
Days between last preintervention OMH
exit and NY/NY placement (gap)a
No preintervention period OMH inpatient
record
Hospital days in preintervention period
Hospital stays in preintervention period
Shelter days in preintervention period
NY/NY placement in 1992
NY/NY placement in 1993
NY/NY placement in 1994
Age at NY/NY placement
Male
Black race
295 diagnosis (schizophrenia)
296 diagnosis (affective disorders)
Drug/Alcohol dependency diagnosis

Parameter
Estimate
(Days Saved)

Lower
(95%) CI

Upper
(95%) CI

–101.40**
75.33***
0.16***

–165.69
55.66
0.11

–37.12
95.00
0.21

–70.49***

–99.90

–41.09

0.75***
1.76
–0.03

0.66
–15.19
–0.09
Reference Category
2.51
–17.51
12.08
–9.98
0.42
–0.60
–8.20
–26.98
3.90
–15.15
–46.44***
–63.61
–35.17***
–55.93
–7.39
–25.84

0.83
18.70
0.03
22.53
34.15
1.45
10.59
22.94
–29.26
–14.41
11.07

a

For those with no preintervention OMH inpatient record, the gap is set at 731 days.
*p < 0.05. **p < 0.01. ***p < 0.001.

9

This is estimated by multiplying 75.3 by the 897 persons with both NY/NY placements and OMH records and then dividing by the 2,396 NY/NY placements in 1992–94.
This assumes that the pre/post reduction in state hospital use is the same for NY/NY
placements with and without shelter records. Comparisons of these two subgroups
(Metraux, Culhane, and Hadley 2000) indicate that use of OMH inpatient services is in
fact somewhat higher among nonshelter users in the preintervention period and that
this group has higher pre/post intervention period reductions compared with counterparts with DHS shelter records. Thus, this 75.3-day reduction extrapolation for nonshelter users is likely to be a conservative estimate.

Housing Policy Debate

Culhane.qxd

5/28/02

120

10:26 AM

Page 120

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Use of NYC public hospitals (HHC). HHC granted access to inpatient
hospital records from 1989 to 1996 for all persons with a history of
DHS shelter use. These parameters limit the analysis to the 1,984 persons who had a NY/NY placement between 1991 and 1994 and who also
had a DHS shelter record. Of these, 855 (43.1 percent) had at least one
record of inpatient hospitalization through HHC that was not reimbursed through Medicaid,10 and these observations were matched with
controls selected from persons who had both a DHS shelter record and
at least one HHC hospitalization record. The resulting case-control
group, consisting of 791 matched pairs (92.5 percent of those with HHC
records), are used for further analysis on NY/NY housing placements
and their impact on hospitalizations.11
HHC hospital utilization, summarized in table 5, shows another substantial unadjusted pre/post placement decline for the NY/NY group.
Among all NY/NY placements (first column), total users decline
68.6 percent from the pre- to the postintervention period, while hospital days consumed declines even more sharply at 79.9 percent. Looking
at the case-control groups (second and third columns), for the cases
there is a similar pre/post decline in persons hospitalized, 68.9 percent,
and again a larger decline in days consumed, 78.0 percent. Comparatively for the controls, both the pre/post declines in persons hospitalized and days consumed, 49.5 percent and 53.4 percent, respectively,
are substantial but considerably lower than the declines for case observations. While the cases have a significantly higher preintervention
number of hospital days used (p < 0.01), their number of postintervention hospital days used is significantly lower than that of the control
group (p < 0.0001).
Regression model results (table 6) show that, after controlling for differences in the included covariates, NY/NY placement is associated with a
greater pre/post differential of 8.1 days (95 percent CI, 4.6 to 11.6 days).
Averaging this 8.1-day reduction over the 1,984 observations results in

10
Hospitalizations that are included in both the HHC and Medicaid datasets (i.e., a
Medicaid-reimbursed inpatient stay occurring in an HHC hospital) are omitted from
the HHC analysis and included in the subsequent Medicaid analysis.
11
A separate analysis of hospital stays finds that over three-quarters of the hospitalizations fall into nine Diagnosis Related Groups (DRGs), all of which correspond to treatment for either mental health or substance abuse issues. The Psychosis DRG (430)
alone accounts for over half of all hospitalizations by persons receiving NY/NY placements, during both the pre- and postintervention periods (Metraux, Culhane, and
Hadley 2000). DRG is a categorization system for hospital stays that are medically
related with respect to diagnosis and treatment and that are statistically similar in
length of stay.

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 121

Public Service Reductions Associated with Supportive Housing

121

a reduction of 3.5 days per placement (95 percent CI, 2.0 to 5.0 days).12
Compared with the 16.5 days of mean preintervention HHC use by the
NY/NY group (table 5), this reflects a 21.2 percent adjusted reduction.
Table 5. HHC Hospital Days (non-Medicaid) Consumed by Persons in
NY/NY Housing and Controls in the Two-Year Periods
before and after the NY/NY Intervention
NY/NY
NY/NY
Controls
(Total 1991–94) (Matched Pairs) (Matched Pairs)
N (NY/NY placements with shelter
record)
Inpatient (non-Medicaid) HHC
record: 1989–96

1,984

791

791

855

791

791

Pre-NY/NY intervention (two years)
Total persons hospitalized
549 (27.7%)
Total days hospitalized
32,823
Mean days (all persons)
16.5
Mean days (hospital users)
59.8

515 (65.1%)
27,014
34.2
52.5

515 (65.1%)
26,456
33.4
51.4

Post-NY/NY intervention (two years)
Total persons hospitalized
Total days hospitalized
Mean days (all persons)
Mean days (hospital users)

160 (20.2%)
5,937
7.5
37.1

260 (32.9%)
12,330
15.6
47.4

175 (8.8%)
6,610
3.3
37.8

Note: Between the NY/NY and control groups, paired-comparison t tests assessing difference yield,
for preintervention HHC hospital days, t = –2.6 (790 df and p < 0.01), and for postintervention
differences, t = 5.0 (790 df and p < 0.0001).
Using paired-comparison t tests, pre/post differences yield, within the NY/NY group, t = 15.2
(790 df and p < 0.0001), and within the control group, t = 9.6 (790 df and p < 0.0001).

12
This is estimated by multiplying 8.1 by the 855 persons with NY/NY placements,
DHS records, and HHC records (from which the control group was selected) and then
dividing by the 1,984 persons with NY/NY placements and DHS records. For this analysis, it is assumed that NY/NY placements without DHS shelter records have HHC hospital use patterns that are the same as those of the persons with DHS records used in
this case-control analysis.

Housing Policy Debate

Culhane.qxd

5/28/02

122

10:26 AM

Page 122

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Table 6. Regression Model Estimating Effects on Changes in HHC Hospital
Days (non-Medicaid) Used in the Two-Year Periods before and after
the NY/NY Intervention (N = 791 Matched Pairs)

Covariate
Intercept
Received NY/NY placement
Days between last preintervention HHC
exit and NY/NY placement (gap)a
No preintervention period HHC inpatient
record
Hospital days in preintervention period
Hospital stays in preintervention period
Shelter days in preintervention period
NY/NY placement in 1991
NY/NY placement in 1992
NY/NY placement in 1993
NY/NY placement in 1994
Age at NY/NY placement
Male
Black race
295 diagnosis (schizophrenia)
296 diagnosis (affective disorders)
Drug/Alcohol dependency diagnosis

Parameter
Estimate
(Days Saved)

Lower
(95%) CI

Upper
(95%) CI

–12.67*
8.05***
0.03***

–23.57
4.55
0.01

–1.77
11.55
0.04

–15.31***

–20.39

–10.23

0.94***
–1.34
0.00

0.87
–3.59
–0.01
Reference Category
7.48*
1.78
3.17
–1.87
5.73*
0.58
–0.11
–0.28
2.47
–1.06
2.02
–1.25
–7.28***
–10.44
–8.27***
–12.57
–6.19***
–9.39

1.01
0.90
0.01
13.17
8.21
10.87
0.07
6.00
5.29
–4.11
–3.96
–2.99

a

For those with no preintervention HHC inpatient record, the gap is set at 731 days.
*p < 0.05. **p < 0.01. ***p < 0.001.

Use of Medicaid-reimbursed inpatient and outpatient services. This
analysis looks at claims records, both inpatient and outpatient, for
medical and psychiatric health services that were eligible for reimbursement under the NYS Medicaid program.13 Medicaid data were
available for 1993 through 1997. To provide full two-year pre- and
postintervention periods of claims records, only the cohort placed in
NY/NY housing in 1995 and a set of matched controls were included in
this analysis.

13
The Medicaid inpatient claims data include hospital stays that are duplicated in the
HHC database but are not used for the HHC analysis. Over three-quarters of the
health care services provided in both outpatient and inpatient settings involved a primary diagnosis of either mental illness or substance abuse. The inpatient claims, which
allow up to seven diagnoses, showed at least one diagnosis involving mental illness or
substance abuse 92 percent of the time. More details can be found in Metraux, Culhane, and Hadley (2001a).

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 123

Public Service Reductions Associated with Supportive Housing

123

Inpatient services
As has been the pattern, unadjusted inpatient service use reimbursed
by Medicaid (table 7) drops substantially in the post-NY/NY placement
period. Of the 733 persons who were in the 1995 NY/NY cohort, 502
(68.5 percent) had a Medicaid claims record from 1993 to 1997. The
percentage of this cohort using inpatient services dropped 22.4 percent
between the pre- and postperiods, while the number of inpatient days
consumed dropped a more drastic 39.9 percent. The cost of services,
also included in the Medicaid data, drops proportionately as well.
Table 7. Inpatient Hospital Days Reimbursed by Medicaid for Persons
in NY/NY Housing and Controls in the Two-Year Periods
before and after the NY/NY Intervention
NY/NY
(Total 1995)
N
Medicaid service users

NY/NY
Controls
(Matched Pairs) (Matched Pairs)

733
502

457
457

457
457

406 (55.4%)
25,892
35.3
63.8
$12,538,656
$17,106
$30,883

372 (81.4%)
21,157
46.3
56.9
$10,525,629
$23,032
$28,295

372 (81.4%)
19,210
42.0
51.6
$10,025,685
$21,938
$26,951

Post-NY/NY intervention (two years)
Total persons hospitalized
316 (43.1%)
Total days hospitalized
15,558
Mean days (all persons)
21.2
Mean days (hospital users)
49.2
Total amount billed to Medicaid
$8,070,885
Mean amount billed (all persons)
$11,011
Mean amount billed (hospital
$25,541
users)

280 (61.3%)
13,542
29.6
36.4
$7,109,844
$15,558
$19,112

313 (68.5%)
19,137
41.9
51.4
$10,738,287
$23,497
$28,866

Pre-NY/NY intervention (two years)
Total persons hospitalized
Total days hospitalized
Mean days (all persons)
Mean days (hospital users)
Total amount billed to Medicaid
Mean amount billed (all persons)
Mean amount billed (hospital
users)

Note: For the number of inpatient days (non-HHC) reimbursed by Medicaid:
Between the NY/NY and control groups, paired-comparison t tests assessing difference yield, for
the preintervention period, t = –4.8 (456 df and p < 0.0001), and for postintervention differences,
t = 3.7 (456 df and p < 0.001).
Using paired-comparison t tests, pre/post differences yield t = 6.0 (456 df and p < 0.0001) within
the NY/NY group and t = 0.05 (456 df and p = 0.96) within the control group.
For the billing of inpatient days (non-HHC) reimbursed by Medicaid:
Between the NY/NY and control groups, paired-comparison t tests assessing difference yield, for
the preintervention period, t = –1.3 (456 df and p = 0.20), and for the postintervention period,
t = 4.5 (456 df and p < 0.0001).
Using paired-comparison t tests, pre/post differences yield, within the NY/NY group, t = 5.1
(456 df and p < 0.0001), and within the control group, t = –0.84 (456 df and p < 0.40).

Housing Policy Debate

Culhane.qxd

5/28/02

124

10:26 AM

Page 124

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Also in table 7, the case group’s pre/post drop in service use is in contrast to virtually no change in the number of days used and costs
accrued by the controls over this time. Compared with the controls, the
cases have a significantly higher number of days consumed in the preintervention period (p < 0.0001), but nonsignificant cost differences
and a significantly lower number of postintervention days consumed
and costs accrued in the postintervention period (p < 0.001).
Separate regression models are presented in table 8 for pre/post
changes in days and costs. Controlling for all other factors in the
model, a NY/NY placement is still significantly associated with pre/post
reductions of 12.6 days (95 percent CI, 6.2 to 18.9 days) and $7,983
(95 percent CI, $4,608 to $11,358). Averaged over the total number of
NY/NY placements in 1995, this leads to an estimated reduction of
8.6 days (95 percent CI, 4.2 to 13.0 days) and $5,467 (95 percent CI,
$3,156 to $7,779).14 These adjusted reductions reflect 24.4 percent and
31.9 percent declines from the mean preintervention levels of inpatient
days used and costs accrued, respectively, by the overall group of
NY/NY placements studied here.

Outpatient services
In contrast to the reductions in inpatient services that have been documented so far, table 9 shows that the number of outpatient visits and the
costs increase, by 95.2 percent and 114.1 percent, respectively, for the
1995 NY/NY cohort. Looking at the case-control group, the same group
as was used for the inpatient analysis, the significant and substantial
increase among the cases is matched by a modest, nonsignificant pre/post
increase in the number of outpatient visits by the control group. These
pre/post changes in visits and costs, when adjusted through a multivariate model, yield an increase of 68.9 visits (95 percent CI, 47.1 to 90.6)
and $5,612 (95 percent CI, $3,871 to $7,352) associated with NY/NY
placement. Averaging this over all 733 NY/NY placements in 1995 results
in increases of 47.2 visits (95 percent CI, 32.3 to 62.1 visits) and $3,843
(95 percent CI, $2,651 to $5,035). These adjusted amounts reflect proportional increases of 75.9 percent and 81.5 percent over the mean number preintervention outpatient visits consumed and costs accrued,
respectively.

14
This is estimated by multiplying the reductions in days and costs associated with
NY/NY in the case group, 12.6 and $7,983, respectively, by the 502 persons with NY/NY
placements and Medicaid inpatient records (from which the control group was selected)
and then dividing by the 733 persons with NY/NY placements in 1995.

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 125

Public Service Reductions Associated with Supportive Housing

125

Table 8. Regression Model Estimating Effects on Changes in MedicaidReimbursed Inpatient Hospital Days Used and Related Costs in the Two-Year
Periods before and after the NY/NY Intervention (N = 457 Matched Pairs)

Covariate
Intercept
Received NY/NY
placement
Gap—Hospital to NY/NY
interventiona
No preintervention
Medicaid record
Medicaid days
(preintervention)
Medicaid stays
(preintervention)
Amount billed to Medicaid
(preintervention)
Shelter days
(preintervention)
Age
Male
Black race
295 diagnosis
(schizophrenia)
296 diagnosis
(affective disorder)
Chemical dependency
diagnosis

Reduction in
Stays (Days)
95% CI
Parameter
Estimate
Lower Upper

Cost
Reduction ($)
95% CI
Parameter
Estimate
Lower Upper

5.67 –
12.56 ***

14.99
6.19

26.32
18.93

–807
–11,986
7,983 ***
4,608

10,373
11,358

0.01
–27.33 ***

–0.01
–39.95

0.04
–14.71

11.6 *
0.2
–15,063 *** –21,916

22.9
–8,210

0.84 ***

0.69

1.00

–10.4

–1.95

–5.74

1.85

–999.6

0.00

0.00

0.00

0.8 ***

0.01

–0.01

0.02

3.0

–0.01
–4.50
–8.28 *
–21.60 ***

–0.43
–11.44
–15.30
–29.24

–21.07 ***
–13.74 ***

–95.0

74.2

–3,213.5 1,214.3
0.7

1.0

–6.3

12.2

0.41
2.44
–1.26
–13.96

27.6
–188.9
–1,543
–5,400
–4,853 *
–8,680
–7,947 *** –12,177

244.1
2,314
–1,026
–3,717

–29.24

–12.91

–11,524 *** –16,152

–6,895

–20.15

–7.33

–9,979 *** –13,402

–6,556

a

In records where there is no preintervention period Medicaid inpatient stay, a value of 731 days is
inserted.
*p < 0.05. **p < 0.01. ***p < 0.001.

Use of VA hospitals. This analysis examines inpatient VA hospital data
from 1992 through 1999 across pre- and postintervention periods for
the 2,496 persons with NY/NY placements in the years 1994–97 and,
when applicable, the controls matched to individual observations.15

15
Judging from the information available, approximately 20 percent of the persons
receiving NY/NY placement claim veteran status. These proportions are somewhat
higher for men (27 percent) and very small for women (2 percent). This suggests that
approximately one fifth of the persons with NY/NY placements are eligible for VA services and that a smaller number will likely actually use these services. The DRGs for
hospitalizations for persons with NY/NY placements show that over 75 percent of the
stays involved treatment for mental illness or substance abuse or both (Metraux et al.
2000).

Housing Policy Debate

Culhane.qxd

5/28/02

126

10:26 AM

Page 126

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Table 9. Outpatient Visits Reimbursed by Medicaid for Persons
in NY/NY Housing and Controls in the Two-Year Periods
before and after the NY/NY Intervention
NY/NY
(Total 1995)
N
Medicaid service users
Pre-NY/NY intervention (two years)
Total persons with outpatient
visits
Total outpatient visits
Mean visits (all persons)
Mean visits (hospital users)
Total amount billed to Medicaid
Mean amount billed (all persons)
Mean amount billed
(hospital users)

NY/NY
Controls
(Matched Pairs) (Matched Pairs)

733
502

457
457

457
457

461 (62.9%)

419 (91.7%)

410 (89.7%)

45,615
62.2
98.9
$3,448,239
$4,704
$7,480

42,623
93.3
101.7
$3,246,487
$7,104
$7,748

37,323
81.7
91.0
$2,796,755
$6,120
$6,821

440 (96.3%)

374 (81.8%)

80,913
177.1
183.9
$6,587,614
$14,415
$14,972

40,109
87.8
107.2
$3,218,494
$7,043
$8,606

Post-NY/NY Intervention (two years)
Total persons with outpatient
483 (65.9%)
visits
Total outpatient visits
89,042
Mean visits (all persons)
121.5
Mean visits (hospital users)
184.4
Total amount billed to Medicaid
$7,382,207
Mean amount billed (all persons)
$10,071
Mean amount billed
$16,013
(hospital users)

Note: For the number of outpatient visits (non-HHC) reimbursed by Medicaid:
Between the NY/NY and control groups, paired-comparison t tests assessing difference yield, for
preintervention differences, t = 1.5 (456 df and p = 0.15), and for postintervention differences,
t = 7.7 (456 df and p < 0.0001).
Using paired-comparison t tests, pre/post differences yield, within the NY/NY group, t = –8.2
(456 df and p < 0.0001), and within the control group, t = –0.9 (456 df and p = 0.35).
For the billing of outpatient visits (non-HHC) reimbursed by Medicaid:
Between the NY/NY and control groups, paired-comparison t tests assessing difference yield, for
preintervention differences, t = 1.4 (456 df and p = 0.15), and for postintervention differences,
t = 7.6 (456 df and p < 0.0001).
Using paired-comparison t tests, pre/post differences yield, within the NY/NY group, t = –9.2
(456 df and p < 0.0001), and within the control group, t = –1.7 (456 df and p = 0.10).

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 127

Public Service Reductions Associated with Supportive Housing

127

Table 10. Regression Model Estimating Effects on Changes in Visits and
Costs related to Medicaid Outpatient Use in the Two-Year Periods before and
after the NY/NY Placement (N = 457 Matched Pairs)

Covariate
Intercept
Received NY/NY
placement
Gap—Visit to NY/NY
interventiona
No preintervention
Medicaid record
Medicaid visits
(preintervention)
Amount billed to Medicaid
(preintervention)
Shelter days
(preintervention)
Age
Male
Black race
295 diagnosis
(schizophrenia)
296 diagnosis
(affective disorders)
Chemical dependency
diagnosis

Reduction in
Visits
95% CI
Parameter
Estimate
Lower Upper
–0.97
–68.88 ***

Cost
Reduction ($)
95% CI
Parameter
Estimate
Lower Upper

–66.20
–90.62

64.25
–47.14

–4,130
–5,612 ***

–9,394
–7,352

1,134
–3,871

0.11 **

0.04

0.18

10.02 ***

5.00

15.04

–64.36 **

–110.25

–18.47

–8,808

–2,047

0.53 ***

0.26

0.81

–26.99

19.61

0.00

0.00

0.00

0.30

0.87

–0.05

–0.14

0.05

–9.33

3.53

–1.24
32.64 *
0.79
–25.96

–2.67
6.86
–22.25
–52.75

0.19
58.42
23.82
0.82

–10.99
2,637 *
545
–2,820 **

–118.89
525
–1,296
–4,867

96.92
4,749
2,385
–774

–38.09 **

–66.53

–9.64

–2,641 *

–4,725

–557

6.83

–22.38

36.04

1,009

–1,306

3,324

–5,427 **
–3.69
0.58 ***
–2.90

a

In records where there is no preintervention period Medicaid inpatient stay, a value of 730 days is
inserted.
*p < 0.05. **p < 0.01. ***p < 0.001.

Among the NY/NY placements, 323 (12.9 percent) had some record
of VA inpatient hospitalization between 1992 and 1999. Of these,
255 (10.2 percent) had records of hospitalization, and these observations, whether or not they had shelter records, were matched with persons who had DHS shelter records (but not necessarily in the two-year
preintervention period). Of the 323 observations with VA records, 294
(91 percent) were matched with control observations. Table 11 shows
that both NY/NY and control groups had (by design) virtually identical
numbers of days of preintervention VA inpatient hospital use and that
this use declined significantly for both cases and controls in the postintervention period. However, the decline was substantially greater
among the cases than among the controls, leading to statistically significant postintervention case-control differences (p < 0.001).

Housing Policy Debate

Culhane.qxd

5/28/02

128

10:26 AM

Page 128

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Table 11. VA Inpatient Days Consumed by Persons in NY/NY Housing and
Controls in the Two-Year Periods before and after the NY/NY Intervention
NY/NY–DHS
NY/NY
Controls
(Total 1994–97) (Matched Pairs) (Matched Pairs)
N
Total service users

2,496
323

294
294

294
294

Persons with preintervention
VA hospitalization
Total preintervention hospital
days
Mean preintervention hospital
days (all persons)
Mean preintervention hospital
days (hospital users)

255 (10.2%)

229 (77.9%)

229 (77.9%)

19,578

15,332

15,130

7.8

52.1

51.5

76.8

67.0

66.1

Persons with postintervention
VA hospitalization
Total postintervention hospital
days
Mean postintervention hospital
days (total)
Mean postintervention hospital
days (hospital users)

169 (6.8%)

153 (52.0%)

180 (61.2%)

8,053

7,651

12,289

3.2

26.0

41.8

47.7

50.0

68.3

Note: Between the NY/NY and control groups, paired-comparison t tests assessing difference yield,
for preintervention state hospital days, t = –0.8 (293 df and p = 0.41), and for postintervention
differences, t = 3.7 (293 df and p < 0.001).
Using paired-comparison t tests, pre/post differences yield, within the NY/NY group, t = 6.9
(293 df and p < 0.0001), and within the control group, t = 2.3 (293 df and p < 0.05).

The regression model results on table 12 show a significant 14.4-day
reduction in VA hospital use (95 percent CI, 5.6 to 23.1 days), all other
factors held equal. When averaged over all 2,496 NY/NY placements
made during the years 1994–97, this effect becomes considerably more
diluted, resulting in an estimate of 1.9 days saved (95 percent CI, 0.7 to
3.0 days) per NY/NY placement. This represents an adjusted 24.4 percent decrease in mean preintervention VA hospital days used attributable to the effect of a NY/NY placement.16
Incarceration in NYSDOCS prisons. The last type of institution
included in this study consists of incarceration facilities: state prisons
and city jails. For state prisons, data include incarcerations up to April
15, 1997, so all NY/NY placements made before April 15, 1995, are
included in this analysis. Because of the availability of records, the
16
This was estimated by multiplying 14.4 by the 323 persons with NY/NY placements
and Medicaid inpatient (non HHC) records (from which the control group was selected)
and then dividing by the 2,496 persons with NY/NY placements and Medicaid inpatient
records.

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 129

Public Service Reductions Associated with Supportive Housing

129

Table 12. Regression Model Estimating Effects on Changes in VA Hospital
Days Used in the Two-Year Periods before and after the NY/NY Intervention
(N = 294 Matched Pairs)

Covariate
Intercept
Received NY/NY placement
Gap—VA to interventiona
Pre-NY/NY shelter record
VA days (preintervention)
VA stays (preintervention)
Pre-NY/NY shelter days
Placement in 1994
Placement in 1995
Placement in 1996
Placement in 1997
Age
Male
Black race
295 diagnosis (schizophrenia)
296 diagnosis (affective disorders)
Chemical dependency diagnosis

Parameter
Estimate
(Days Saved)

Lower
(95%) CI

34.63*
14.37**
0.03
–18.31
0.77***
0.02
0.01

0.54
5.60
0.00
–37.40
0.62
–4.89
–0.01
Reference Category
–1.45
–12.47
9.73
–0.67
9.26
–3.15
–0.62*
–1.11
–19.15*
–34.97
–4.14
–12.60
–15.40**
–25.96
–26.78***
–37.45
–10.93*
–19.43

Upper
(95%) CI
68.72
23.14
0.06
0.78
0.91
4.93
0.04
9.57
20.12
21.68
–0.13
–3.34
4.31
–4.84
–16.12
–2.44

a

For those with no preintervention VA inpatient record, the gap is set at 731 days.
*p < 0.05. **p < 0.01. ***p < 0.001.

case-control group used in the DHS shelter analysis is again used here
to assess differences in pre- and postintervention days spent incarcerated. Thus, the case group includes only those persons with DHS shelter records, and 44 pairs were omitted from the analysis because the
control observation was incarcerated at the intervention point and thus
would have biased the pre/post analyses.17
Table 13 shows that low proportions of observations in either group
have records of incarceration. Despite this, the NY/NY placements
show substantial reductions in the use of state prisons. In the casecontrol comparison, the state prison utilization rate for the two groups
is very similar (and statistically nonsignificant) in the preintervention
period, suggesting that, for the purposes of this analysis, the groups are
comparable. In the postintervention period, the NY/NY group shows a
substantial, statistically significant reduction in the number of days
incarcerated (p < 0.0001), while the control group fails to show any statistically significant change in the number of persons incarcerated or in
the total number of days the group was incarcerated.
17
In these 44 pairs, the case observation did not necessarily have an incarceration
record.

Housing Policy Debate

Culhane.qxd

5/28/02

130

10:26 AM

Page 130

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Table 13. NYSDOCS Prison Days Used by Persons in NY/NY Housing and
Controls in the Two-Year Periods before and after the NY/NY Intervention
NY/NY–DHS
NY/NY
Controls
(Total up to 4/1/95) (Matched Pairs) (Matched Pairs)
N
Total service users
Pre-NY/NY intervention period
Persons incarcerated
Time incarcerated (days)
Time incarcerated (days per
total persons)
Time incarcerated (days per
person incarcerated)
Post-NY/NY intervention period
Persons incarcerated
Time incarcerated (days)
Time incarcerated (days per
total persons)
Time incarcerated (days per
person incarcerated)

3,196
109

2,285
94

2,285
136

87 (2.7%)
29,569
9.3

75 (3.3%)
25,490
11.2

74 (3.2%)
25,241
11.0

339.9

339.9

341.1

36 (1.1%)
7,818
2.4

32 (1.4%)
6,938
3.0

78 (3.4%)
26,236
11.5

217.1

216.8

336.4

Note: Between the NY/NY and control groups, paired-comparison t tests assessing difference yield,
for preintervention state incarceration days, t = –0.05 (2,294 df and p = 0.96), and for postintervention differences, t = 5.2 (2,294 df and p < 0.0001).
Using paired-comparison t tests, pre/post differences yield, within the NY/NY group, t = 5.2
(2,294 df and p < 0.0001), and within the control group, t = –0.2 (2,294 df and p = 0.83).

After multiple regression was used to control for various factors in
table 14, it was determined that having a NY/NY placement is associated with a reduction of 7.9 days (95 percent CI, 4.8 to 11.0 days). This
estimate differs from those in the previous models in that placements
in the case and control groups were included regardless of whether
they had a state prison record. Since case-control data are unavailable
for the incarceration of persons without a DHS shelter record, this
adjusted reduction is used, without further adjustment, as the per
placement reduction in prison use attributable to NY/NY. Taking this
reduction estimate as a proportion of average preintervention prison
use, 7.9 days represents an 84.8 percent reduction in the mean preintervention days spent incarcerated by the NY/NY group.

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 131

Public Service Reductions Associated with Supportive Housing

131

Table 14. Regression Model Estimating Effects on Changes in NYSDOCS
Incarceration Days in the Two-Year Periods before and after
the NY/NY Intervention (N = 2,285)

Covariate
Intercept
Received NY/NY placement
Any incarceration (preintervention)
Days incarcerated (preintervention)
DHS days (preintervention)
Age at NY/NY intervention
Male
Black
Mental illness indicator
Drug/Alcohol dependency indicator
NY/NY placement in 1990
NY/NY placement in 1991
NY/NY placement in 1992
NY/NY placement in 1993
NY/NY placement in/after 1994

Parameter
Estimate
(Days Saved)
–24.56***
7.89***
–48.98**
1.04***
0.01**
0.35***
–3.68**
1.88
2.43
0.47
–3.35
–0.80
–0.71
–0.31

Lower
(95%) CI

Upper
(95%) CI

–33.75
4.81
–84.54
0.95
0.00
0.21
–6.46
–1.39
–0.80
–2.81
Reference Category
–9.96
–6.24
–5.83
–5.01

–15.36
10.97
–13.41
1.12
0.01
0.50
–0.89
5.15
5.66
3.75
3.25
4.64
4.41
4.40

*p < 0.05. **p < 0.01. ***p < 0.001.

Incarceration in NYCDOC jails. Analysis of incarceration data from
NYCDOC augments the NYSDOCS analysis, which does not cover any
incarceration episodes in county or municipal corrections facilities. This
analysis of NYCDOC incarceration records for Riker’s Island and other
local jail facilities follows the same case-control group that was examined in the NYSDOCS and DHS analyses.18
As shown in table 15, the number of persons incarcerated, as well as
the time spent in jail, declined significantly for the NY/NY group
between the pre- and postintervention periods. Persons spending time
in jail represented 12.0 percent of the total NY/NY group in the preintervention period, but only 8.2 percent of this group in the postintervention period. The total number of days incarcerated fell 39.8 percent
after the housing placements. The average number of persons and days
18
The data set used here is larger than the one in the NYSDOCS analysis because
information was available through 1999 (see appendix A). There are no duplicate incarceration records between this and the NYSDOCS analyses, although some of the
records examined here immediately precede state prison records used in the other
analysis. Of the 1,590 offenses related to the study group over this time, by far the
most frequently occurring types involved possession or sale of drugs (27.2 percent),
offenses related to assault (12.8 percent), theft (11.6 percent), and larceny (8.9 percent).
Some 39 percent of these offenses were charged as felonies (Metraux, Culhane, and
Hadley 2001b).

Housing Policy Debate

Culhane.qxd

5/28/02

132

10:26 AM

Page 132

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Table 15. NYCDOC Jail Days Used by Persons in NY/NY Housing and Controls
in the Two-Year Periods before and after NY/NY Intervention
NY/NY–DHS
(1989–97)
N
Total service users
Pre-NY/NY intervention period
Persons jailed
Time in jail (days)
Time in jail (days per total
persons)
Time in jail (days per person
jailed)
Post-NY/NY intervention period
Persons jailed
Time in jail (days)
Time in jail (days per total
persons)
Time in jail (days per person
jailed)

NY/NY
Controls
(Matched Pairs) (Matched Pairs)

4,679
766

3,284
607

3,284
716

563 (12.0%)
46,574
10.0

441 (13.4%)
36,165
11.0

480 (14.7%)
41,481
12.6

82.7

82.0

86.4

383 (8.2%)
28,027
6.0

308 (9.4%)
21,711
6.6

457 (13.9%)
37,828
11.5

73.2

70.4

82.8

Note: Between the NY/NY and control groups, paired-comparison t tests assessing difference yield,
for preintervention city jail incarceration days, t = 1.4 (3,283 df and p < 0.17), and for postintervention differences, t = 4.8 (3,283 df and p < 0.0001).
Using paired-comparison t tests, pre/post differences yield, within the NY/NY group, t = 4.8
(3,283 df and p < 0.0001), and within the control group, t = 1.1 (3,283 df and p < 0.29).

incarcerated fell such that not only did fewer persons get jailed after
their housing placements, but, for those incarcerated, the average time
spent behind bars also fell.19 These pre/post dynamics are not replicated
in the control group. While the NY/NY and control groups are comparable in their use of jails in the preintervention period, the magnitude of
the reduction per placement realized by the NY/NY group (4.4 days) is
statistically significant (p < 0.0001), as opposed to the smaller and
nonstatistically significant pre/post reduction for the control group
(1.1 days).
When multiple regression is used to control for other factors in table 16
and everything else is held constant, NY/NY placement is associated
with a 3.8-day reduction per placement (95 percent CI, 1.8 to 5.8 days).
As in the NYSDOCS analysis, the estimated regression model includes
all of the DHS case-control observations, regardless of whether they
have a record of jail use, and 3.8 days represents the number of days
19

A total of 54 matched pairs were omitted because the control observation was incarcerated during the intervention date.
Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 133

Public Service Reductions Associated with Supportive Housing

133

Table 16. Regression Model Estimating Effects on Changes in NYC Jail Days
in the Two-Year Periods before and after the NY/NY Intervention
(N = 3,284 Matched Pairs)

Covariate
Intercept
Received NY/NY placement
Any jail (preintervention)
Days jailed (preintervention)
Any shelter use (preintervention)
DHS days (preintervention)
Age at NY/NY intervention
Male
Black
Mental illness indicator
Drug/Alcohol dependency indicator
NY/NY placement in 1990
NY/NY placement in 1991
NY/NY placement in 1992
NY/NY placement in 1993
NY/NY placement in 1994
NY/NY placement in 1995
NY/NY placement in 1996 or 1997

Parameter
Estimate
(Days Saved)
–18.21***
3.81***
–22.32***
0.93***
1.80
0.01***
0.25***
–4.41***
–0.03
–1.64
–0.33
3.05
4.61*
2.93
0.39
–0.12
2.49

Lower
(95%) CI

Upper
(95%) CI

–24.17
1.79
–28.61
0.88
–1.11
0.00
0.17
–6.48
–2.09
–3.73
–2.61
Reference Category
–1.46
0.28
–1.60
–4.12
–4.84
–1.89

–12.26
5.84
–16.03
0.98
4.71
0.01
0.33
–2.35
2.04
0.45
1.94
7.56
8.94
7.46
4.91
4.59
6.87

*p < 0 .05. ** p< 0 .01. ***p < 0 .001.

per placement (without further adjustment) attributed to a NY/NY
placement. This represents a 38.0 percent decrease in the mean preintervention period number of incarceration days used by the case group.

Cumulative system effects
The results of the system-specific analyses have thus far been presented separately by agency over the two-year postplacement period.
For purposes of facilitating an interpretation of the cumulative effects
of the intervention within and across systems, summary results are
provided in tables 17 and 18.
Table 17 estimates the costs of service utilization by intervention group
members in the two years before their housing placement by multiplying service days used by the average per diem service cost (in 1999 dollars). These costs are then annualized by dividing by two. The results
show that, per placement per year, the total mean cost of service utilization for the two-year pre-NY/NY placement period was $40,451.
The bulk of those expenditures occurred in health services (86 percent,
Housing Policy Debate

Culhane.qxd

5/28/02

134

10:26 AM

Page 134

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Table 17. Summary of Mean Two-Year Pre-NY/NY Intervention Period
Service Use across Seven Service Providers

Service Provider
DHS
OMH
HHC
Medicaid (inpatient)
Medicaid (outpatient stays)
VA
Department of Corrections
(state)
Department of Corrections
(city)
Total

Mean Days Used
(Two Years
Pre-NY/NY)

Cost
Annualized
Per Diem* (Two Years)*
Cost*

137.0
57.3
16.5
35.3
62.2
7.8
9.3

$68
$437
$755
$657
$84
$467
$79

$9,316
$25,040
$12,458
$23,192
$5,225
$3,643
$735

$4,658
$12,520
$6,229
$11,596
$2,613
$1,822
$368

10.0

$129

$1,290

$645

$80,899

$40,451

*In 1999 dollars. Totals reflect rounding.

or $34,778) and in emergency shelter services (11 percent, or $4,658).
Criminal justice services (incarceration costs only) accounted for only
3 percent, or $1,012 per year.
Table 18 summarizes estimates of the cost reductions in service use,
based on pre/post placement comparisons and as adjusted by the casecontrol regression analyses. Cost savings are again imputed based on
estimated per diem costs by service system in 1999 dollars. Results indicate that placement in NY/NY housing is associated with a $12,146 net
reduction in health, corrections, and shelter service use annually per
person over each of the first two years of the intervention. Half of those
cost reductions are associated with reduced use of state psychiatric inpatient services, and another quarter (23 percent) are associated with
reduced use of emergency shelter services. Half of the remaining quarter in net savings is associated with reduced use of NYC public hospitals
(10.9 percent of the total) and VA hospitals (3.7 percent of the total).
Reductions in costs associated with Medicaid inpatient services outweigh, by $843 (6.9 percent of the total cost reductions), the increases
in Medicaid outpatient services. Taken together, about 95 percent of
the cost reductions are associated with reductions in health and shelter
services. The criminal justice system costs account for the remaining
4.5 percent of the total cost reductions associated with a supportive
housing placement.

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 135

Public Service Reductions Associated with Supportive Housing

135

Table 18. Summary of Estimated Cost Reductions Associated with Reductions
in Service Use Attributable to NY/NY Housing, by Type

Service Provider
DHS
OMH
HHC
Medicaid (inpatient)
Medicaid (outpatient
visits)
VA
Department of
Corrections (state)
Department of
Corrections (city)
Total

Days Saved
(Two Years
Pre/Post)

Cost
Reductions
95% CI

Per
Diem*

Cost
Annualized
Reductions*
Cost
(Two Years) Reductions*

82.9
28.2
3.5
8.6

77.4–88.5
20.8–35.6
2.0–5.0
4.2–13.0

$68
$437
$755
$657

$5,637
$12,323
$2,643
$5,650

$2,819
$6,162
$1,322
$2,825

–47.2
1.9

–62.1 to –32.3
0.7–3.0

$84
$467

–$3,965
$887

–$1,983
$444

7.9

4.8–11.0

$79

$624

$312

3.8

1.8–5.8

$129

$490
$24,289

$245
$12,146

*In 1999 dollars. Totals reflect rounding.

Do reductions in service use offset the costs of
supportive housing?
One of the primary purposes of the previous analyses was to determine
whether reductions in service use attributable to a housing placement
offset the costs of the intervention. To compare the costs of the intervention with the reduced service system costs, both sets of costs must
be computed in comparable terms. In the previous cross-system analysis, the reductions in service use were calculated in terms of annualized
average cost reductions per placement in the two-year period after
housing placement. Alternatively, housing cost figures, given the annual budgeting process by which they are calculated by city and state
officials (the methodology for deriving the housing costs is provided in
appendix B), are measured in annual costs per housing unit. Each
measure has its usefulness, the former for service system planners who
need to take account of client turnover and project costs for a pool of
placements and the former for housing planners, who need to project
costs based on fully occupied units of housing, irrespective of
turnover.20
20
Because these measures are not directly comparable, they must be converted, taking
into account client turnover, to produce annualized cost and cost-reduction estimates.
Because tenant-level data on length of housing tenure for each placement were not
available for this analysis, aggregated data on longevity of placement in NY/NY housing, presented in Lipton (1996), are used. These data indicate that NY/NY tenants stay
in housing, on average, for 17.9 months of the two-year postintervention period. The
annualized length of tenure is therefore 8.95 months, or 0.746 a year. The inverse of

Housing Policy Debate

Culhane.qxd

5/28/02

136

10:26 AM

Page 136

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Per placement per year. Table 19 shows the conversion of annual costs
per housing unit into annualized costs per placement, both overall and
as broken down by the two housing models featured in NY/NY.21 The
annualized cost per placement, averaged over the two-year postplacement period and derived by multiplying the annual costs per housing
unit by the average annual length of tenure (0.746 years ), is $13,570.
This can then be compared with the adjusted service cost reductions
reported in the last column of table 18, which show an annualized
reduction of $12,146 per placement. The result is a net annual cost of
$1,425 associated with a NY/NY placement. It is noteworthy that the
supportive housing units, which comprise two-thirds of the units developed, have an annualized cost per placement of $12,889 and therefore
operated at a lower average annualized net cost of $744 per placement.
Table 19. Estimated Annual Costs per Unit and Annualized Costs
per Placement of NY/NY Housing, by Housing Type
Housing Type
Community residence (mean)
Supportive housing (mean)
Weighted mean

Number
of Units

Annualized
per Unit Cost*

Annualized per
Placement*

1,384
2,231
3,615

$19,662
$17,277
$18,190

$14,668
$12,889
$13,570

*In 1999 dollars.

Per housing unit per year. Alternatively (and inversely), one can convert the annualized service utilization reductions, reported in terms
of placements (table 18) into annualized reductions per housing unit
(table 20). These reductions are then expressed in terms of average
annualized service cost reductions per housing unit by multiplying the
annualized per placement service reductions by the annualized number
of tenants per housing unit (1.34 per year). This procedure yields
turnover-adjusted cost reductions per housing unit per year of $16,281,
imputing an assumption of year-round housing occupancy. This figure
can be compared with the estimated cost per housing unit per year,

this number, 1.34, produces the annualized average number of tenants per housing
unit. These numbers are used to compute both annualized per placement costs from
the annual housing unit costs in appendix B and annualized per housing unit service
cost reductions from the per placement service use reductions in table 18. These two
computations reflect inverse procedures and are equally valid approaches for comparing
the service cost savings and housing costs associated with the intervention (all figures
are in 1999 dollars).
21

See appendix B for the specific housing programs and also table B.8.

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 137

Public Service Reductions Associated with Supportive Housing

137

as shown in table 19 (and as computed in appendix B), which also
assumes year-round occupancy. Comparing the average annual cost
of a housing unit ($18,190) with the comparable measure for the service utilization reductions yields a net cost of $1,908 per unit per year.
Again, the supportive housing units, which comprise two-thirds of the
units developed, operate at a lower net cost of $995 per housing unit
per year. The net cost attributable to year-round occupancy of NY/NY
housing can be calculated by multiplying the annualized per unit net
cost by 3,615 (the total number of housing units developed), yielding a
net annual cost of $6,897,420 per year.
Finally, multiplying the turnover-adjusted cost reductions by 3,615 for
each category of service produces an estimate of the annual cost reductions (or increases) accruing to each service type attributable to a yearround housing placement, as shown in the last column of table 20.
These figures provide useful information about the impact of the intervention on aggregate service use annually by service type and demonstrate that annual service use was reduced by $58.9 million. This
compares to the annual cost of the NY/NY intervention (including operating, service, and debt service costs) of approximately $65.8 million.
Table 20. Annualized Cost Reductions per Placement and per Housing Unit,
and Total NY/NY Housing Units (N = 3,615), by Service Type

Service Provider
DHS
OMH
HHC
Medicaid (inpatient)
Medicaid (outpatient stays)
VA
Department of
Corrections (state)
Department of
Corrections (city)
Total

Annualized
Cost Reductions
per Placement*

Annualized Cost
Reductions per
Housing Unit*

Total Cost
Reduction by
NY/NY Units*

$2,819
$6,162
$1,321
$2,825
–$1,982
$444
$312

$3,779
$8,260
$1,771
$3,787
–$2,657
$595
$418

$13,660,436
$29,860,094
$6,401,361
$13,689,511
–$9,604,464
$2,151,555
$1,511,903

$245

$328

$1,187,232

$12,146

$16,281

$58,857,628

*In 1999 dollars. Totals reflect rounding.

Discussion
The placement of homeless people with SMI in supportive housing is,
as expected, associated with substantial reductions in homelessness.
Not only do homeless people with severe mental disabilities who are
placed in housing have marked reductions in shelter use, they also
Housing Policy Debate

Culhane.qxd

5/28/02

138

10:26 AM

Page 138

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

experience marked reductions in their use of hospital and correctional
facilities. Although other studies tracking the placement of homeless
people in housing have found comparable housing retention rates, only
a few, more limited analyses have assessed reductions in collateral services (Averyt and Kamis-Gould 2000; Proscio 2000). By contrast, this
study provides a uniquely broad and more comprehensive test, employing a case-control study design and examining the impact of a comparatively large number of housing placements on seven major publicly
financed service systems.
It is important to note that this study was able to quantify for the first
time in the published literature the extent of service use by homeless
people with SMI before a housing placement. Results show that such
persons are extensive users of publicly funded services, particularly
inpatient health services, and that they accumulate an average of
$40,449 per year in health, corrections, and shelter system costs. While
the costs of services before housing placement are comparably high for
the cases and controls, due to the matching criteria of this study, it is
not clear whether these costs can be generalized to all homeless people
with SMI, and they certainly cannot be generalized to homeless people
irrespective of their mental health. Nevertheless, in light of this high
cost for such a significant number of persons, the importance of the
effect found here—that the supportive housing intervention significantly reduces these costs—is further reinforced.
Strictly on the basis of the direct cost reductions measured here and
compared with the annual cost of the housing, the NY/NY initiative
was a sound investment of public resources. The $6.9 million net
annual cost, or $1,908 per housing unit per year, represents approximately 10 percent of the annual overall cost of providing this housing.
However, supportive housing units, which were the more common type
developed under this initiative and which better represent the trend in
housing development for people with mental illness, operated at a more
modest per unit cost of $995 per year, or 5 percent of the overall housing unit cost. In other words, 95 percent of the costs of the supportive
housing (operating, service, and debt service costs) are compensated for
by reductions in collateral service attributable to the housing placement. This modest cost is particularly striking, given the magnitude
of the initiative, which required an original capital investment of
$200 million and which costs $65.8 million annually (including service,
operating, and debt service costs).
It should be noted that the service reductions measured here represent
a conservative assessment of the impact of the initiative on service use
and costs. First, by limiting the analysis to the impact on service use in

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 139

Public Service Reductions Associated with Supportive Housing

139

the first two years postintervention, the study has included the stabilization period associated with entry into housing. As in other service
interventions for people with SMI, service use often increases temporarily following placement, because tenants’ unmet health and psychiatric needs are more likely to be identified and treated once they
receive regular, periodic case management services (Pollio et al. 2000).
If this were the case here, one would expect service use to decline and
stabilize over time, producing net cost savings in successive years.
However, this possibility must be balanced against the possibility that
people may be engaged in services more intensively before a housing
placement, in part to prepare them for it. This area deserves further
study.
This study did not include all direct or indirect costs associated with
service use by the homeless persons eventually placed in housing.
Street outreach services, soup kitchens, and services provided by dropin centers were not included. Health services funded by the federal
Health Care for the Homeless program were not included. Other clinical and social services provided at shelters that are funded by grants
from the Department of Housing and Urban Development’s (HUD’s)
McKinney Act programs were also not included, and neither were
the costs of uncompensated care provided by private hospitals. Not
included as well are the social costs of homelessness, which are far
more difficult to enumerate or to associate with individual persons.
They include the costs of crime to crime victims, to the courts, and to
the police, and the private and public costs of accommodating homelessness (or not) in public spaces.
Finally, many of the potential benefits of the housing initiative were
not measured here. Residents of supported housing are more likely to
secure voluntary or paid employment (HUD 1994) and to experience
an improved quality of life. Investments in supported housing have also
been shown to be associated with improved neighborhood quality and
property values (Arthur Andersen LLP et al. 2000). Lastly, the social
value of reduced homelessness and of providing greater social protection for those who are disabled, while not possible to translate into economic terms, constitutes an important if less tangible benefit to society.
Were all such costs and benefits included, these unmeasured costs of
homelessness and benefits of the housing intervention would have
increased its already significant net benefit (and potential cost savings).
Although this study was limited to one locality and cannot be generalized to all urban areas, the results have important public policy implications. Research suggests that on any given day, as many as 112,000
single adults with SMI are homeless in the United States and that as

Housing Policy Debate

Culhane.qxd

5/28/02

140

10:26 AM

Page 140

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

many as 280,000 single adults are chronically homeless.22 If such persons, or even significant proportions of them, are extensive users of
acute care health services, public shelters, and criminal justice systems,
then the results of this study suggest that an aggressive investment
in supportive housing is warranted. While such housing may not be
appropriate or effective for every person who is homeless and mentally
ill, enough would likely benefit that their placement in housing could
significantly offset the costs of a targeted initiative, such as was demonstrated here. In effect, the results presented here indicate that policy
makers could substantially reduce homelessness for a large, visible
segment of the homeless population—often thought to be stubbornly
beyond the reach of the social welfare safety net—at a very modest cost
to the public.
However, while reductions in service use may nearly cover the costs of
supportive housing intervention in the aggregate (assuming that the
results given here can be generalized beyond NYC), it remains a major
public policy challenge to shift funds from one set of purposes (health,
jails, prisons) to another (housing or housing support services). Different
levels of government pay for different activities, and some will have to
do so regardless of whether homeless people are using them (jails and
prisons, for example). Moreover, legislative committees with responsibility for housing cannot appropriate funds from health committees for
housing or housing support purposes, regardless of the savings in health
costs that might justify the expenditure. So, the challenge facing proponents of a national strategy to increase the supply of supportive housing
will be to determine how costs can be paid in one area (for housing or
housing support services), when the bulk of the savings from the intervention will accrue elsewhere (state mental health services, Medicaid,
etc.). In New York, a complex package of federal, state, and city resources was required to pay for the operating, service, and debt service
costs of the NY/NY initiative (see appendix B). Similarly, a national
strategy will require the participation of various levels of government
and multiple agencies within each level of government.
22
The Burt et al. (2001) analysis of a 1996 federal survey of homeless persons suggests
that as many as 840,000 people were homeless at one point in time in the United States
that year. One-third of those were people in families, leaving approximately 560,000
single adults. A meta-analysis of epidemiological research estimates that approximately
20 percent of homeless adults without children have SMI (Lehman and Cordray 1993),
yielding an estimated 112,000 persons with SMI as homeless for that study period.
Longitudinal research in two large U.S. cities (Philadelphia and New York) finds that
people who are chronic shelter users, with or without a mental disability, represent
approximately 50 percent of the daily shelter-using population (Kuhn and Culhane
1998) or an estimated 280,000 of the adults during the Burt et al. one-day study period.
Given the differing sampling frames underlying their derivation, these figures must be
understood as gross estimates only.

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 141

Public Service Reductions Associated with Supportive Housing

141

Operating costs
A substantial hurdle that must be overcome in developing and sustaining permanent supportive housing is bridging the gap between the
costs of operating the housing and the extremely low incomes of prospective tenants. Supportive housing providers typically address this
gap through a direct housing subsidy to the tenant or housing unit
and/or income supports to the tenants. The NY/NY initiative drew on
both strategies to cover the operating costs, relying on a combination of
federal Section 8 subsidies, supplemental security income payments by
the state, and some direct state support, resulting in an average per
unit subsidy of $4,600 (see appendix B, derived from table B.3).
Historically, HUD has been the primary source of housing subsidies.
An especially potent source of operating subsidies for supportive housing serving homeless persons, including those with SMI, has been the
McKinney-Vento Homeless Assistance Act, which authorized operating
subsidies in various forms under its three major programs: Section 8
Moderate SRO Rehabilitation, the Supportive Housing Program, and
Shelter Plus Care. Federal primacy and initiative in the provision of
operating subsidies are likely necessary if supportive housing for homeless persons with SMI is to be taken to scale. Even in relatively wealthy
states, there is little evidence of an inclination to displace or even significantly add to the federal role in this regard (Twombley et al. 2001).
Although the federal investment in incremental housing subsidies
slowed to a trickle in the mid-1990s (DeParle 1996), significant opportunities in this area may be on the horizon. In enacting HUD’s fiscal
year (FY) 2001 budget, Congress explicitly stated its goal that “HUD
and local providers increase the supply of permanent supportive housing for chronically homeless, chronically ill people over time until the
need is met (estimated 150,000 units)” (U.S. Senate 2000, 52–53). To
that end, Congress maintained its recent requirement that 30 percent
of McKinney-Vento funds (about $300 million per year based on recent
annual appropriations) be targeted to permanent housing for homeless
persons with disabilities. If this funding level is maintained, then this
investment alone could result in subsidizing 96,000 new units of supportive housing over the next 10 years.23

23
This figure was derived from an estimated cost of $6,100 per unit (based on HUD’s
FY 1999 estimates for the Shelter Plus Care program), assuming five-year terms and
an annual inflation adjustment of 2 percent. Some 9,643 subsidies per year would
result. Over a decade, this would translate into 96,433 incremental subsidies, assuming
that any subsidies expiring during this period were renewed from another source.

Housing Policy Debate

Culhane.qxd

5/28/02

142

10:26 AM

Page 142

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

The FY 2001 VA–HUD appropriations bill also authorized substantial
changes in the statute that allow local housing authorities to convert
tenant-based Section 8 vouchers into project-based subsidies linked to
specific units in order to spur new development in tight housing markets or for special populations (Public Law 106–377, Sec. 232. 114
Stat. 1441A at 31). The changes streamlined a previously underused
tool for housing development24 and also increased the ceiling on such
“project-basing” of vouchers to 20 percent of the total tenant-based
portfolio (the previous limit was 15 percent). Nationally, this could
translate into over 300,000 potential project-based operating subsidies.25 Even in the absence of incremental vouchers, the new projectbasing statutory authority can be of significant use in adding affordable units to serve special populations, like homeless persons with SMI,
who often cannot access housing in a tight rental market even with a
tenant-based subsidy (Sard 2001).26
Thus, opportunities already exist to finance operating subsidies for
permanent supportive housing. Of course, other issues remain to be
resolved, including local resistance to the siting of such housing, the
capacity of states and localities to develop it, and the ongoing financial
burden of renewing operating subsidies (though this final factor is an
issue in all federally subsidized affordable housing).

Capital and debt service
With respect to capital and debt service costs, the NY/NY initiative
used a combination of city and state bonds, valued at nearly $200 million, and, secondarily, federal tax credits, valued at approximately
$5 million,27 to fund the capital costs for acquisition, development, and

24

The NYC Housing Authority has been a notable exception among local housing agencies in its willingness in recent years to stimulate the development of new housing by
project-basing a portion of its tenant-based Section 8 portfolio. The new statutory provisions could encourage other housing authorities to follow its example.
25
While Congress has appropriated between 50,000 and 100,000 incremental vouchers
each year for the past half decade, the Bush budget proposes only 34,000 this year.
26

Evidence suggests that homeless households, even without SMI, struggle to use housing subsidies in the private market. See “Judge Orders City” (2001) (noting that the
city-funded rent subsidy program intended to house 460 homeless families had to date
housed only 11 because private landlords were reluctant to accept them as tenants).
27
Approximately $25 million in federal tax credit expenditures were involved in the
NY/NY initiative, but only 20 percent of these funds supported capital costs; 80 percent
was used to fund operating reserves (see appendix B).

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 143

Public Service Reductions Associated with Supportive Housing

143

rehabilitation. The average debt service cost per unit per year for the
NY/NY initiative is approximately $4,900 (see table B.7).
Several factors will affect whether sufficient capital investment/debt
service can be obtained to develop supportive housing at the scale to
meet the need among homeless person with SMI. Nationally, competition with other low-income housing programs for federal tax credits and
competition with other state and local purposes for bond funds will pose
a challenge to state and local leaders who must balance demands for
housing for the homeless with other public interests. It remains to be
seen whether the federal or state governments are willing to establish
a priority for supportive housing for the homeless in this competitive
process or will allocate new dollars, given the potential for offsetting
cost savings. The availability of capital/debt service funding is also likely
to vary significantly across different regions of the country.
To the extent that policy makers perceive existing affordable housing
programs, including the Low-Income Housing Tax Credit and HOME
programs, as insufficient to produce capital for housing that extremely
low income households (below 30 percent of the area median income
[AMI]) can afford, supportive housing providers would clearly benefit
from a new production program targeted to that population.28 In the
106th Congress, senators from both major parties introduced bills
directed to this purpose,29 and such a program was nearly enacted as
part of the FY 2001 HUD appropriations bill. It remains to be seen
whether the momentum will carry over into this congressional session
or whether the new administration will identify such an initiative as a
priority for HUD.

Supportive services
The challenge of identifying funding for the services that must accompany this housing may well prove greater than finding the resources for
the housing. In the case of the NY/NY initiative, the NYS OMH paid
for nearly all of the services associated with this housing, at an average

28

Data from the 1999 American Housing Survey indicate that there are more than 5
million fewer housing units affordable and available (i.e., not occupied by a household
of higher income) to households earning below 30 percent of the AMI than there are
such households. Put another way, there are 7.7 million households with incomes below
30 percent of the AMI and 4.9 million units affordable to such households, but 2.6 million of those units are already occupied by households with incomes above 30 percent of
the AMI—so only 2.3 million of the affordable units are actually available. (Dolbeare,
C., unpublished data. Available from the author.)
Housing Policy Debate

Culhane.qxd

5/28/02

144

10:26 AM

Page 144

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

cost of approximately $9,100 per unit per year.30 Results from this
study indicate that the expenditures by the state OMH also provided
ample returns, given that the plurality of cost reductions attributable
to the intervention were reductions in OMH inpatient hospital costs
($8,260 per housing unit, see table 20). Other inpatient health services
paid by Medicaid, public hospitals, and the VA experienced combined
cost reductions as well ($6,153 per housing unit per year, see table 20).
However, only Medicaid paid for some of the housing support services
(at a net savings of $1,130 per housing unit per year).
From the perspective of developing a national strategy, the question
for proponents will be how to motivate other states or health service
payers (and potential savers) to make the same commitment as the
NYS OMH and, secondarily, Medicaid did under this initiative. One
possible mechanism is to make housing support services (or more of
them) reimbursable by Medicaid. Unfortunately, some of these services,
such as intensive case management or community treatment teams, are
already reimbursable by Medicaid at the states’ option, and many
states do not avail themselves of it. Another option is to increase funds
or create a set-aside for housing support service funds in the federal
Mental Health Block Grant. However, states have successfully opposed
federal mandates on block grant funds and may oppose such a mandate
for this program. A third alternative would be to create a new program
at the federal level that would provide matching funds from the
Department of Health and Human Services (DHHS) for funds committed by HUD through its set-aside in the McKinney-Vento Act for permanent, supportive housing. In this case, services could be specifically
targeted to housing for homeless people with mental disabilities.31
Whatever the specific mechanism chosen, the provision of support
services will be a necessary component to any national strategy to
address the housing problems of homeless people with SMI.

29
S.2997, National Affordable Housing Trust Fund Act of 2000 (introduced by Senator
John Kerry, D-MA), available at <http://www.thomas.loc.gov>; S 3033, The Housing
Needs Act of 2000 (introduced by Senator Christopher Bond, R-MO), also available at
<http://www.thomas.loc.gov>.
30

For this calculation, tenant contributions, where applicable, were deducted from
average state service contract amounts to produce a net service cost (see appendix B).

31
“The 106th Congress Wrap-up” contains language of agreement reached in the
appropriations process but dropped at the later stages.

Fannie Mae Foundation

Culhane.qxd

5/28/02

10:26 AM

Page 145

Public Service Reductions Associated with Supportive Housing

145

Conclusion
In sum, acquiring the resources for supportive housing will require
local, state, and federal leadership in all three areas that comprise the
essential elements of this intervention: operating subsidies, capital/debt
service, and supportive services. The federal government, through new
programs, matching funds, and set-asides within existing programs,
can provide the incentives that engage states’ interest. However, only
executive leadership at the state level can compel various agencies to
work together for the common, multijurisdictional purpose of developing permanent supportive housing for homeless persons with SMI. Our
research has demonstrated the compensatory cost reductions of a supportive housing initiative, but only political will and leadership can act
on such findings to guide the next initiative through the intergovernmental and interagency maze.
Of course, there are some caveats to this study—it is post hoc and quasiexperimental. Therefore, the extent to which cases and controls are
truly comparable could not be addressed fully by random assignment.
Comparability problems were reduced by matching cases and controls
according to a variety of available demographic, service, and diagnostic
criteria, and by statistically correcting for differences that may have
remained. However, the extent to which unmeasured differences
between the study groups may persist cannot be fully ascertained, nor
can the possibility of a selection effect in the study sample be eliminated. Whether housing providers select for heavier service users or for
less severe cases could not be determined, although every effort was
made to produce results generalizable to the population of homeless persons with SMI from which the intervention group was drawn. Despite
this limitation, it is also clear that there exists a relatively large pool of
homeless persons with SMI for whom this housing is effective in achieving housing stability and providing offsetting reductions in the use of
collateral services.
There are also caveats on the use of administrative data. Given the
large volume of data entered into the databases of the service systems
studied here and a level of quality control on data entry that is not as
stringent as is usual for scientific studies, these administrative sources
can be prone to missing data, keystroke errors, and erroneous information. Although missing data did not present a problem in these analyses, it is more difficult to ascertain the quality of the data along the
other dimensions. Despite these potential problems, the only source
that can inform a study of service use covering a large study group over
an extended period of time, as is done here, is administrative data.

Housing Policy Debate

Culhane.qxd

5/28/02

146

10:26 AM

Page 146

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Future research should specify the effects of the various housing types
on patterns of service use. Causes of attrition from supportive housing
and the housing transitions of those who exit supportive housing also
deserve careful attention. For although most people remain stably
housed two years after placement, a third of the clients exit this housing and represent a substantial group that should be further studied.
Future research could also benefit from replicating this study method,
in that integrated administrative records provide a wealth of information on the utilization patterns and costs of a population that has
otherwise proven costly and difficult to track and study. In particular,
applying this method to studying patterns of homelessness and service
use among the majority of homeless persons who do not have an SMI
could likewise prove informative as to the potential efficacy of the various policies and intervention strategies that would target them. Further fruitful areas of study could examine the same group studied here
but follow their service use over longer pre- and postintervention periods (as additional data become available) and/or through service systems not covered in this article.

Fannie Mae Foundation

1989–96

1992–99

HHC
(nonMedicaid)

VA

1987–99

1993–97

Medicaid

NYCDOC

1990–96

OMH

1989–97

1989–
4/15/95

1994–97

1991–94

1995

1992–94

1989–97

4,679

3,196

2,496

2,396

733

2,396

4,679

Total
NY/NY
Placements

3,284

2,285

294

791

457

570

3,338

Matched
Pairs

NY/NY placements (cases) omitted include those
without a DHS shelter record (N = 1,341)
whose matched control observation was
incarcerated on the placement date (N = 54)

NY/NY placements (cases) omitted include those
without a DHS shelter record (N = 911)
whose matched control observation was incarcerated
on the placement date (N = 44)

NY/NY placements (cases) omitted include those
without a VA hospital inpatient record (N = 2,173)
without an appropriate control match (N = 29)

NY/NY placements (cases) omitted include those
without a DHS shelter record (N = 412)
with a DHS record but without an HHC inpatient
record (N = 1,920)
without an appropriate control match (N = 64)

NY/NY placements (cases) omitted include those
without a Medicaid claim record (N = 231)
without an appropriate control match (N = 45)

NY/NY placements (cases) omitted include those
without a state hospital inpatient record (N = 1,499)
with a state hospital record but without a DHS
shelter record (N = 267)
without an appropriate control match (N = 60)

NY/NY placements (cases) without a DHS
shelter record are omitted (N = 1,341)

Restrictions

10:31 AM

1987–
4/15/97

1987–99

DHS

Intervention
Years
Selected

5/28/02

NYSDOCS

Time
Frame

Service
Provider

Table A.1. Selection Factors for Constructing the Matched-Pair Case-Control Groups Used in the Analyses

Culhane_app.qxd
Page 147

Public Service Reductions Associated with Supportive Housing
147

Appendix A

Summary of Control Group Selection across Seven Public
Services Systems

Housing Policy Debate

Culhane_app.qxd

148

5/28/02

10:31 AM

Page 148

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Appendix B
Estimating direct federal, state, and city expenditures on the
NY/NY supportive housing initiative
Introduction
In 1990, NYS and NYC agreed to collaborate on what became known as
the NY/NY supportive housing initiative. The agreement committed
the state and the city to jointly fund construction, operating, and social
service costs for 3,600 community-based housing units in NYC for persons who were severely mentally ill and homeless. While much of the
funding for this program comes from the state and the city of New
York, as well as from the federal government, an array of private nonprofit organizations administer these housing units. This appendix provides estimates of the federal, state, and city outlays for the NY/NY
initiative, in the aggregate, per year, per housing unit, and per housing
placement. This compilation and disaggregation of costs is intended to
provide a quantitative benchmark for evaluating returns from the program, as measured by the reductions in the use of collateral services
that were detailed in the article.

Data and methods
Data on the distribution of housing units and both the total and the
per unit operating, debt, and service costs were constructed from
budget documents and in consultation with state and city administrators involved in financing and administering the programs.32 Inflation
adjustments and unit-cost calculations were also verified by personal
communication with program administrators. All figures are reported
in 1999 dollars. Unless otherwise noted, all cost estimates assume full,
year-round occupancy of housing units. Actual costs will differ for specific sites and specific service contracts.
32

Final confirmation of data on construction costs and debt service costs was obtained
by personal communication with Michael Newman of the NYS OMH (March 5, 2001).
Figures for the tenant contributions deducted from OMH-developed sites were confirmed in consultation with Christopher Roblin of the NYS OMH (April 4, 2001, personal communication). The value of federal Section 8 subsidies and the service cost
estimates for the city-developed sites were confirmed by personal communication with
Peter Bittle of the NYC Department of Mental Health (April 4, 2001). Information on
the financing of city Department of Housing Preservation and Development (HPD)–
developed units was obtained by personal communication with Timothy O’Hanlon of
the NYC HPD (March 5, 2001). Information on the city HRA-developed units was
obtained from David Mittelman of the NYC DHS (March 25, 2001).
Fannie Mae Foundation

Culhane_app.qxd

5/28/02

10:31 AM

Page 149

Public Service Reductions Associated with Supportive Housing

149

Housing and service configurations
The housing developed under the NY/NY initiative encompasses a
variety of housing and service configurations and consists of several
distinct models that combine housing with rehabilitative or support
services. The residential continuum includes, on the one end, supportive housing models that provide private, individual apartment-style
living accommodations with varied levels of service support. In the supportive housing model, the intensity of services provided depends on
need as expressed by the tenant and can change as the tenant’s needs
change. Tenants’ housing tenure is based on a lease arrangement and
is not contingent on a prescriptive service plan. On the other end of the
continuum are community residence living arrangements that provide
more intense, structured regimens of supportive services. Community
residential programs are supervised, and housing is part of a structured
treatment plan. Both supportive housing and community residence
approaches are tied into bodies of research literature that provide theoretical rationales and evaluations for each approach. More relevant to
the purposes of this inquiry is the fact that the two models are associated with different sources and amounts of funding.
Table B.1 shows the number of housing units, grouped by different
housing and service configurations, that are funded under NY/NY.
Under the community residences subheading, community residence/
SROs provide extended-stay housing in SRO living units with on-site
services for those whose self-maintenance and socialization skills are
minimal. Community residences are single-site facilities with either private or shared bedrooms, meals provided, access to on-site rehabilitative
services, and 24-hour staff coverage. Community residences often target
special populations, such as people with co-occurring mental health and
substance abuse problems, and they seek to eventually place residents in
less service-intensive, more permanent housing arrangements.
The supportive housing grouping includes various state- and cityadministered programs that range from scattered-site, individual apartments to clustered apartments or SRO units in a single development.
In both cases, services are available on a periodic or as-needed basis.
For capital (state) supportive housing, construction was initiated by
NY/NY funding, while rental (state) supportive housing used existing
units in the private rental market. For city-funded supportive housing,
the Department of Housing Preservation and Development (HPD)
administers NYC–HPD housing units, the HRA administers NYC–HRA
units, the NYC–NYS units are administered by the city with state capital funding, and the three NYC rental units use existing, private apartments that receive state funding for services.

Housing Policy Debate

Culhane_app.qxd

150

5/28/02

10:31 AM

Page 150

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Table B.1. NY/NY–Funded Housing: Number of Units under
Different Housing/Service Configurations
Housing
Community residences
Community residences/SROs
Community residences
Supportive housing
Capital (state)
Rental (state)
Capital (NYC–HPD)
Capital (NYC–HRA)
Capital (NYC–NYS)
Rental (NYC)
Total units

Units
713
671
285
520
1,087
258
78
3
3,615

Operating and service costs
Because of variations in housing layout and services offered, different
housing configurations have varying levels of operating and service
costs. Table B.2 shows the levels of service cost for the different types
of housing. Service costs represent funding for social and related services provided to NY/NY tenants—services that are provided on-site,
brokered by case management staff, or arranged with outside providers.
Depending on the unit type, federal, state, and city funding can pay for
these services. The community residence units have higher average
service costs because of the greater intensity of the services provided.33
For the supportive housing subheading, however, while tenants receive
less intense services than their counterparts in the community residence housing, the differences in per unit funding reflect different levels and sources of money but not necessarily different intensities of
service provision.
Operating costs, shown in table B.3, reflect costs needed for building
upkeep, apartment maintenance, utilities, and so on and are most often
provided in the form of rental subsidies to supplement the rent tenants

33

Average service costs for community residence units are $19,200 per unit. However,
because Medicaid pays for the services in 128 of these units, at an average annual cost
of $17,478 per unit, and because those costs are accounted for in the services utilization
analysis as client outpatient costs paid by Medicaid, they have been removed here. This
reduces the average service cost per community residence unit to $15,865, from nonMedicaid sources.

Fannie Mae Foundation

Culhane_app.qxd

5/28/02

10:31 AM

Page 151

Public Service Reductions Associated with Supportive Housing

151

pay. In federally subsidized units (Section 8), all of which are operated
by the city, tenants pay 30 percent of their income, here assumed to be
the income for a person receiving SSI and living alone ($617/month),
toward rent. Thirty percent of this amount ($185) has been deducted
from the maximum fair market rent for an SRO allowed in NYC
($500), and the difference is calculated as the federal Section 8 contribution ($3,800).
Table B.2. Service Costs for NY/NY Housing Type, by Funding Source
Housing
Community residences
Community
residences/SROs
Community residences3
Supportive housing
Capital (state)
Rental (state)
Capital (NYC–HPD)
Capital (NYC–HRA)
Capital (NYC–NYS)
Rental (NYC)

Units

State
Funding*

City
Funding*

Per Unit
Subtotal*

713
671

$10,500
$15,865

$0
$0

$10,500
$15,865

285
520
1,087
258
78
3

$8,400
$4,800
$9,400
$9,400
$9,400
$9,400

$0
$0
$900
$900
$900
$900

$8,400
$4,800
$10,300
$10,300
$10,300
$10,300

*In 1999 dollars.

Table B.3. Operating Costs for NY/NY Housing Type, by Funding Source

Housing
Community residences
Community
residences/SROs
Community residences
Supportive housing
Capital (state)
Rental (state)
Capital (NYC–HPD)
Capital (NYC–HRA)
Capital (NYC–NYS)
Rental (NYC)

Units

Federal
(Section 8)
Funding*

State
Funding*

City
Funding*

Per Unit
Subtotal*

713
671

$0
$0

$5,700
$4,200

$0
$0

$5,700
$4,200

285
520
1,087
258
78
3

$0
$0
$3,800
$3,800
$3,800
$3,800

$5,700
$5,000
$0
$0
$0
$0

$0
$0
$0
$0
$0
$0

$5,700
$5,000
$3,800
$3,800
$3,800
$3,800

*In 1999 dollars.

Housing Policy Debate

Culhane_app.qxd

152

5/28/02

10:31 AM

Page 152

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Table B.4 combines the net unit operating and service costs and, in
applicable categories, deducts tenant contributions from this cost.
In the case of many NYS OMH-administered units—all community
residence/SROs and Capital (state) units—the Section 8 standard for
tenant rent contribution (one-third of income) is applied. However, tenant rent contributions are deducted from state operating and service
contracts after providers report tenant rent collections. Thus, a separate column in table B.4, Less Tenant Contribution, shows the impact
of deducting tenant contributions from the operating and service cost
contracts of these state-administered units (again, assuming full, yearround occupancy, by persons receiving SSI income for an adult living
alone). The deduction for community residences is higher ($6,000
annually) because additional funds are deducted from tenant income to
cover the costs of board. The Total Cost column reflects the per unit
costs multiplied by the number of units for the particular type of housing, less tenant contributions, where applicable.

Capital costs
Some 11 percent of the NY/NY units consist of existing, private rental
housing, and the rest are located in buildings that have been specifically constructed or rehabilitated under the auspices of the NY/NY
program. In the latter case, the city or the state provided the capital.
Table B.5 provides the capital costs, both per unit and total, broken
down by different subcategories of NY/NY housing. All of the capital
costs per unit are budgeted at $70,000, except for HPD-administered
units, which are set at $50,000 per unit. HPD unit costs were lower for
several reasons: HPD began its property acquisition process earlier
than the state did, and real estate values were relatively more
depressed then; HPD also acquired some properties at essentially no
cost (properties it owned through tax foreclosure—in rem buildings);
and HPD’s developments involved much larger buildings than the
state’s projects, so development costs per unit were lower (Timothy
O’Hanlon, personal communication).34

34
Approximately half (N = 508) of the HPD units received revenue from the sale of
federal tax credits. For purposes of calculating the per unit per year cost, the tax credits
are assumed to pay out over a 15-year period, amounting to a $3,333 per unit per year
cost. Although the cost of the tax credits is not included in the tabulation of debt service costs here, given that only 20 percent of the tax credit revenue was used for capital
support, it is figured into the final costs per unit by source, in table B.7, combined with
the debt service amounts.

Fannie Mae Foundation

*In 1999 dollars.

3,615

285
520
1,087
258
78
3

$5,700
$5,000
$3,800
$3,800
$3,800
$3,800

$5,700
$4,200
$8,400
$4,800
$10,300
$10,300
$10,300
$10,300

$10,500
$15,865
$2,200
$0
$0
$0
$0
$0

$2,200
$6,000

Less
Tenant
Contribution*

$11,900
$9,800
$14,100
$14,100
$14,100
$14,100

$14,000
$14,065

Per Unit
Cost*

$48,013,715

$3,391,500
$5,096,000
$15,326,700
$3,637,800
$1,099,800
$42,300

$9,982,000
$9,437,615

Total
Cost*

10:31 AM

Total

Supportive housing
Capital (state)
Rental (state)
Capital (NYC–HPD)
Capital (NYC–HRA)
Capital (NYC–NYS)
Rental (NYC)

713
671

Units

Service
per Unit
Subtotal*

5/28/02

Community residences
Community residences /SROs
Community residences

Housing

Operating
per Unit
Subtotal*

Table B.4. Combined Service and Operating Costs, Less Tenant Contributions, for NY/NY Housing

Culhane_app.qxd
Page 153

Public Service Reductions Associated with Supportive Housing
153

Housing Policy Debate

Fannie Mae Foundation

*In 1999 dollars.

3,615

$70,000
$0
$0
$0
$70,000
$0

$70,000
$70,000
$0
$0
$50,000
$70,000
$0
$0

$0
$0

City

$70,000
$0
$50,000
$70,000
$70,000
$0

$70,000
$70,000

Total

$122,290,000

$19,950,000
$0
$0
$0
$5,460,000
$0

$49,910,000
$46,970,000

State

$0
$0

$72,410,000

$0
$0
$54,350,000
$18,060,000
$0
$0

City

$194,700,000

$19,950,000
$0
$54,350,000
$18,060,000
$5,460,000
$0

$49,910,000
$46,970,000

Total

10:31 AM

Total

285
520
1,087
258
78
3

713
671

Community residence
Community
residences /SROs
Community residences

State

5/28/02

Supportive housing
Capital (state)
Rental (state)
Capital (NYC–HPD)
Capital (NYC–HRA)
Capital (NYC–NYS)
Rental (NYC)

Units

Combined Costs*

154

Housing

Per Unit Costs*

Table B.5. Capital Costs Allocated to Various Types of NY/NY–funded Housing

Culhane_app.qxd
Page 154

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Culhane_app.qxd

5/28/02

10:31 AM

Page 155

Public Service Reductions Associated with Supportive Housing

155

NYC and NYS both financed capital costs through a separate series of
bond issues. On the basis of available information, on $81 million of the
$130.5 million in state capital costs, the aggregated interest rate on
state bond issues was 6.339 percent. Similarly, interest rates on bonds
issued by the city were between 6 percent and 7 percent; the higher
rate was the basis for these calculations.35 A 25-year amortization
schedule and these interest rates were used to estimate annual debt
service costs and allocate them across the different types of housing
based on the number of units and the capital costs (see table B.6). Both
the city and state incur these debt service costs on behalf of the housing provider as part of their assistance to NY/NY.

Calculating average total costs per NY/NY housing unit
and per housing placement
The debt service, social service, and operating costs are combined and
averaged across housing types in table B.7 to come up with more complete cost estimates. Each of these estimates, which aggregate two or
more subtypes of housing, represents a weighted mean as determined
by the number of housing units and specific costs associated with each
subtype. Federal tax credit costs have been added, along with the debt
service costs in this table, although only 20 percent of federal tax credit
revenue was applied to capital expenses. The remaining 80 percent was
used to fund operating reserves. Tax credit costs are assumed to pay
out over the first 15 years of the project but are assumed as a constant
annual cost here (see also footnote 34).
The total combined cost per unit per year, for all NY/NY units, is estimated at $18,190.36 Breaking down this estimate, community residence
housing costs more per unit than supportive housing ($19,662 versus
$17,277, respectively). Social services and operating costs account for
73 percent of the total estimated per unit cost, and NYS provides, on
average, 78 percent of the estimated per unit cost.

35

Information on the state’s NY/NY bond issues was provided through personal communication with Michael Newman at the NYS OMH; information on the interest rates
for bond issues was provided by the city.
36

This number excludes Medicaid-paid services for the 128 community residence units
mentioned earlier.

Housing Policy Debate

Fannie Mae Foundation

*In 1999 dollars.

3,615

$5,630
$0
$0
$0
$5,630
$0

$5,630
$5,630
$0
$0
$4,293
$5,997
$0
$0

$0
$0

City

$5,630
$0
$4,293
$5,997
$5,630
$0

$5,630
$5,630

Total

$9,835,373

$1,604,550
$0
$0
$0
$439,117
$0

$4,013,976
$3,777,730

State

$0
$0

$6,213,540

$0
$0
$4,666,369
$1,547,171
$0
$0

City

$16,048,913

$1,604,550
$0
$4,666,369
$1,547,171
$439,117
$0

$4,013,976
$3,777,730

Total

10:31 AM

Total

285
520
1,087
258
78
3

713
671

Community residences
Community
residences/SROs
Community residences

State

5/28/02

Supportive Housing
Capital (state)
Rental (state)
Capital (NYC–HPD)
Capital (NYC–HRA)
Capital (NYC–NYS)
Rental (NYC)

Units

Total Costs*

156

Housing

Per Unit Costs*

Table B.6. Estimated Debt Service on State and City Bond Issues to Fund Capital Costs for NY/NY Housing

Culhane_app.qxd
Page 156

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Culhane_app.qxd

5/28/02

10:31 AM

Page 157

Public Service Reductions Associated with Supportive Housing

157

Table B.7. Estimated Unit Costs to Federal, State and City Sources for NY/NY
Housing Averaged over Housing Types
Net Service and Operating Costsa
Housing Type
Community residence
Supportive housing
Total

Units

Federal

State

City

Subtotal

1,384
2,231
3,615

$0
$2,429
$1,499

$14,032
$9,813
$11,428

$0
$575
$355

$14,032
$12,817
$13,282

City

Subtotal

$0
$2,785
$1,719

$5,630
$4,460
$4,908

City

Subtotal

$0
$3,360
$2,074

$19,662
$17,277
$18,190

Debt Service Costsa
Units
Community residence
Supportive housing
Total

Federal

1,384
2,231
3,615

State

$0
$759b
$468

$5,630
$916
$2,721

Total Costsa

Community residence
Supportive housing
Total

Units

Federal

State

1,384
2,231
3,615

$0
$3,188
$1,967

$19,662
$10,729
$14,149

a

In 1999 dollars.
This per unit cost reflects the cost associated with the federal tax credit, paid over the first
15 years of each project. Overall, 20 percent went for debt service, and the remainder went for
operating reserves, to be used in the event that Section 8 support stopped on units after the initial
five-year commitment.

b

To facilitate comparisons of the housing costs reported here and reductions in the use of collateral services reported in the article, housing
costs have been converted into annualized costs per placement in
table B.8. The annualized per unit costs (from table B.7) were converted into per diem costs by multiplying them by 0.746, the average
annualized length of tenure over the first two years following placement (Lipton 1996). This yields an annualized per placement cost of
$13,570. (This approach is consistent with the one used in the service
utilization analysis.)
Table B.8. Estimated per Annum, per Diem, and
per Placement per Year Costs
Housing

Units

Annual per
Unit Cost

Community residence (mean)
Supportive housing (mean)
Weighted mean

1,384
2,231
3,615

$19,662
$17,277
$18,190

Annualized per
Placement Cost
$14,668
$12,889
$13,570

Housing Policy Debate

Culhane_app.qxd

5/28/02

158

10:31 AM

Page 158

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Alternatively, one could also compare housing costs and service system
cost reductions by converting the per placement per year service utilization reductions reported in the article into per housing unit per year
numbers, by applying the same set of assumptions. As shown in
table B.9, the utilization reductions by service type, adjusted for casecontrol differences, can be expressed in terms of the annualized cost
reductions per placement (also shown in the last column of table 18).
These annualized per placement reductions can be expressed in terms
of annualized cost reductions per unit by adjusting for the housing
retention rate (multiplying the annualized per placement reductions by
1.34—the annualized number of tenants per average housing unit). This
procedure yields turnover adjusted and annualized cost reductions
attributable to the full-year housing placement of $16,281 per unit per
year. This figure can then be compared with the estimated costs of the
housing units presented in this appendix, which have already been calculated in terms of the per unit per year costs and (with the exception
of table B.8) have also assumed year-round occupancy.
Table B.9. Annualized Cost Reductions Adjusted for Housing Turnover:
Per Unit and for Total NY/NY Housing Units (N = 3,615), by Type
Service Provider
DHS
OMH
HHC
Medicaid (inpatient)
Medicaid (outpatient stays)
VA
Department of Corrections (state)
Department of Corrections (city)
Total

Annualized
Cost Reductions*
$2,819
$6,162
$1,321
$2,825
–$1,982
$444
$312
$245
$12,146

Annualized Turnover/
Adjusted Cost Reductions*
$3,779
$8,260
$1,771
$3,787
–$2,657
$595
$418
$328
$16,281

*In 1999 dollars.

Conclusion
This article has presented estimates of costs borne by NYC, NYS, and
the federal government for the construction, operation, and service
provision associated with NY/NY housing. City and state program
administrators served as sources for the cost information, and costs
are assessed in 1999 dollars. It must also be reiterated that these cost
estimates are not comprehensive, but rather reflect the assistance provided by the federal, state, and city governments to nonprofit housing
and service providers to administer this housing. Additional funding

Fannie Mae Foundation

Culhane_app.qxd

5/28/02

10:31 AM

Page 159

Public Service Reductions Associated with Supportive Housing

159

may come from the nonprofit agencies themselves and from tenant rent
contributions that are otherwise not deducted here.37
NY/NY housing reflects diverse housing and service configurations,
which correspond to a wide range of expenses. In the process of combining the different types of housing, average costs also become less
representative of individual types of housing. While the average figures
are useful for comparing NY/NY housing costs with potentially offsetting cost reductions brought on by reductions in the use of collateral
services such as psychiatric hospitals, other public and private hospitals, homeless shelters, and corrections programs, future research
should be refined by breaking down cost calculations by specific programs and by specific tenancy histories.

Authors
Dennis P. Culhane is Associate Professor of Social Welfare Policy at the Center for Mental Health Policy and Services Research at the University of Pennsylvania. Stephen
Metraux is Senior Research Coordinator at the Center for Mental Health Policy and
Services Research at the University of Pennsylvania. Trevor Hadley is Professor of Psychology in Psychiatry at the University of Pennsylvania.
The authors gratefully acknowledge the generous assistance of the following individuals and their agencies or organizations (or former agencies/organizations) in obtaining
data access, providing data, reviewing preliminary results, and/or securing funding for
this effort: Gail Clott, Jill Berry, and Susan Wiviott (NYC DHS); Leon Cosler, Peter
Gallagher, Dellie Glaser, and Thomas Fanning (NYS Department of Health); Bruce
Fredrick, Susan Jacobsen, and Steven Greenstein (NYS Department of Correctional
Services); Eric Sorenson (NYC Department of Corrections); Sharon Salit, Laray Brown,
and Ava Quint (NYC HHC); Robert Rosenheck (U.S. Department of Veterans Affairs);
Barry Brauth, Michael Newman, Christopher Roblin (NYS OMH); Timothy O’Hanlon
(NYC Department of Housing Preservation and Development); Peter Bittle (NYC
Department of Mental Health); Frank Lipton (NYC HRA); Brad Race and Robert King
(NYS Executive Offices); Sharon Salit, David Gould, and Lenore Glickhouse (United
Hospital Fund of New York); Tracy Rutnik, Stephanie Jennings, James Carr, Steven
Hornburg, and Lawrence Smalls (the Fannie Mae Foundation); Sandra Newman (Johns
Hopkins University); and Julie Sandorf, Constance Tempel, Ted Weerts, Roger Clay,
Cynthia Stuart, Heidie Joo, James Krauskopf, Jonathan Harwitz, Ted Houghton, and
Richard Ravitch (the Corporation for Supportive Housing).
This research was sponsored by the United Hospital Fund of New York, the Conrad
Hilton Foundation, the Rhodebeck Charitable Trust, and the Corporation for Supportive Housing.

37
Medicaid-paid services to tenants, delivered in 128 of the community residence units,
are excluded.

Housing Policy Debate

Culhane_app.qxd

160

5/28/02

10:31 AM

Page 160

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

References
Allison, Paul D. 1999. Logistic Regression Using the SAS System: Theory and Application. Cary, NC: SAS Institute.
Arthur Andersen LLP, University of Pennsylvania Center for Mental Health Policy and
Services Research, Kay E. Sherwood, and TWR Consulting. 2000. The Connecticut Supportive Housing Demonstration Program Evaluation Report. New Haven, CT: Corporation for Supportive Housing.
Averyt, June M., and Edna Kamis-Gould. 2000. The Effect of Supportive Housing on
Tenants, Including Utilization and Costs of Services. Unpublished paper. University of
Pennsylvania, Center for Mental Health Policy and Services Research.
Bebout, Richard R., and Maxine Harris. 1992. In Search of Pumpkin Shells: Residential
Programming for the Homeless Mentally Ill. In Treating the Homeless Mentally Ill: A
Task Force Report of the American Psychiatric Association, ed. H. Richard Lamb, Leona
L. Bachrach, and Frederick I. Kass, 159–82. Washington, DC: American Psychiatric
Association.
Burt, Martha R., Laudan Y. Aron, and Edgar Lee. 2001. Helping America’s Homeless:
Emergency Shelter or Affordable Housing? Washington, DC: Urban Institute Press.
Carling, Paul J. 1993. Housing and Supports for Persons with Mental Illness: Emerging
Approaches to Research and Practice. Hospital and Community Psychiatry 44(5):
439–49.
Caton, Carol L., Richard J. Wyatt, Alan Felix, Jeffrey Grunberg, and Boanerges
Dominguez. 1993. Follow-up of Chronically Homeless Mentally Ill Men. American Journal of Psychiatry 150:1639–42.
Center for Mental Health Services. 1994. Making a Difference: Interim Status Report of
the McKinney Demonstration Program for Homeless Adults with Serious Mental Illness.
Washington, DC.
Center for Urban Community Services. 1995. A Guide to Supportive Housing Models
for People with Mental Illness and Other Special Needs. New York.
Culhane, Dennis P., and Stephen Metraux. 1997. Where to from Here: A Policy
Research Agenda Based on the Analysis of Administrative Data. In Understanding
Homelessness: New Policy and Research Perspectives, ed. Dennis P. Culhane and Steven
P. Hornburg, 341–60. Washington, DC: Fannie Mae Foundation.
DeParle, Jason. 1996. The Year That Housing Died. New York Times Magazine, October 20, p. 52.
Dickey, Barbara, Olinda Gonzalez, Eric Latimer, Karen Powers, Russell K. Schutt, and
Stephen M. Goldfinger. 1996. Use of Mental Health Services by Formerly Homeless
Adults Residing in Group and Independent Housing. Psychiatric Services 47(2):152–58.
Drake, Robert E., Nancy A. Yovetich, Richard R. Bebout, Maxine Harris, and Gregory J.
McHugo. 1997. Integrated Treatment for Dually Diagnosed Homeless Adults. Journal
of Nervous and Mental Disease 185(5):298–305.

Fannie Mae Foundation

Culhane_app.qxd

5/28/02

10:31 AM

Page 161

Public Service Reductions Associated with Supportive Housing

161

Goldfinger, Stephen M., Russell K. Schutt, George S. Tolomiczenko, Larry J. Seidman,
Walter E. Penk, Winston Turner, and Brina Caplan. 1999. Housing Placement and Subsequent Days Homeless among Formerly Homeless Adults with Mental Illness. Psychiatric Services 50:674–79.
Hevesi, Dennis. 1999. Building Homes for the Single Homeless. New York Times, April
28, Sec. 11, p. 1.
Hurlburt, Michael S., Patricia A. Wood, and Richard L. Hough. 1996. Providing Independent Housing for the Mentally Ill Homeless: A Novel Approach to Evaluating LongTerm Housing Patterns. Journal of Community Psychology 24:291–310.
Judge Orders City to Shelter Two Suffering Homeless Families. 2001. New York Times,
April 12, p. B2.
Kennedy, Randall. 1997. Doors That Offer Hope May Shut: A Program That Has
Housed the Mentally Ill Lapses. New York Times, October 4, p. B1.
Kennedy, Shawn G. 1995. New Look for S.R.O.s: Decent Housing. New York Times,
March 28, p. B1.
Kuhn, Randall S., and Dennis P. Culhane. 1998. Applying Cluster Analysis to Test a
Typology of Homelessness by Pattern of Shelter Utilization: Results from the Analysis
of Administrative Data. American Journal of Community Psychology 26(2):207–32.
Lehman, Anthony F., and David S. Cordray. 1993. Prevalence of Alcohol, Drug and
Mental Disorders among the Homeless: One More Time. Contemporary Drug Problems
20:355–83.
Lewis, Dan A., and Arthur J. Lurigio. 1994. The State Mental Patient and Urban Life.
Springfield, IL: Charles C. Thomas.
Lipton, Frank R. 1996. Summary Client Placement Report of the New York–New York
Agreement to House Homeless Mentally Ill Individuals. New York: New York City
Human Resources Administration, Office of Health and Mental Health Services.
Lipton, Frank R., Suzanne Nutt, and Albert Sabatini. 1988. Housing the Homeless
Mentally Ill: A Longitudinal Study of a Treatment Approach. Hospital and Community
Psychiatry 39(1):40–45.
Lipton, Frank R., Carol Siegel, Anthony Hannigan, Judith Samuels, and Sherryl Baker.
2000. Tenure in Supportive Housing for Homeless Persons with Severe Mental Illness.
Psychiatric Services 51(4):479–86.
Metraux, Stephen, Dennis P. Culhane, and Trevor R. Hadley. 2000. Supportive Housing
and Inpatient Hospital Use in the New York City Health and Hospitals Corporation
among Homeless Persons with Severe Mental Illness: Assessing the Impact of “New
York/New York” on the Use of Public Hospitals. Unpublished paper. University of Pennsylvania, Center for Mental Health Policy and Services Research.
Metraux, Stephen, Dennis P. Culhane, and Trevor R. Hadley. 2001a. The Impact of
NY/NY Supportive Housing Placements on Medicaid-Funded Health Care. Unpublished
paper. University of Pennsylvania, Center for Mental Health Policy and Services
Research.

Housing Policy Debate

Culhane_app.qxd

162

5/28/02

10:31 AM

Page 162

Dennis P. Culhane, Stephen Metraux, and Trevor Hadley

Metraux, Stephen, Dennis P. Culhane, and Trevor R. Hadley. 2001b. NY/NY Housing
and the Use of New York City Jail Services: An Analysis Merging Administrative Data.
Unpublished paper. University of Pennsylvania, Center for Mental Health Policy and
Services Research.
Metraux, Stephen, Robert Rosenheck, Dennis P. Culhane, and Trevor R. Hadley (2000).
The Impact of NY/NY Supportive Housing Placements on Inpatient Use of Veterans
Administration Hospitals. Unpublished paper. University of Pennsylvania, Center for
Mental Health Policy and Services Research.
Murray, Ruth, Marjorie Baier, Carol North, Mary Lato, and Cary Eskew. 1997. One-Year
Status of Homeless Mentally Ill Clients Who Completed a Transitional Residential Program. Community Mental Health Journal 33(1):43–50.
Pollio, David E., Edward L. Spitznagel, Carol S. North, Sanna Thompson, and Douglas
A. Foster. 2000. Service Use over Time and Achievement of Stable Housing in a Mentally Ill Homeless Population. Psychiatric Services 51(12):1536–43.
Proscio, Tony. 2000. Supportive Housing and Its Impact on the Public Health Crisis of
Homelessness. New York: Corporation for Supportive Housing.
Ridgway, Priscilla, and Charles A. Rapp. 1998. The Active Ingredients of Effective Supported Housing: A Research Synthesis. Lawrence, KS: University of Kansas, School of
Social Welfare.
Rosenheck, Robert. 2000. Cost-Effectiveness of Services for Mentally Ill Homeless People: The Application of Research to Policy and Practice. American Journal of Psychiatry
157(10):1563–70.
Salit, Sharon A., Evelyn M. Kuhn, Arthur J. Hartz, Jade M. Vu, and Andrew L. Mosso.
1998. Hospitalization Costs Associated with Homelessness in New York City. New England Journal of Medicine 338(24):1734–40.
Sard, Barbara. 2001. Revision of the Project-Based Voucher Statute. Washington, DC:
Center on Budget and Policy Priorities.
Shern, David L., Chip J. Felton, Richard L. Hough, Anthony F. Lehman, Stephen M.
Goldfinger, Elie Valencia, Deborah Dennis, Roger Straw, and Patricia A. Wood. 1997.
Housing Outcomes for Homeless Adults with Mental Illness: Results from the SecondRound McKinney Program. Psychiatric Services 48(2):239–41.
Tsemberis, Sam. 1999. From Streets to Homes: An Innovative Approach to Supported
Housing for Homeless Adults with Psychiatric Disabilities. Journal of Community Psychology 27(2):225–41.
Tsemberis, Sam, and Rhonda F. Eichenberg. 2000. Pathways to Housing: Supported
Housing for Street-Dwelling Homeless Individuals with Psychiatric Disabilities. Psychiatric Services 51(4):487–93.
Twombley, Jennifer G., Harry Lawson, Carla H. Smith, and Sheila Crowley. 2001. A
Report on State-Funded Rental Assistance Programs: A Patchwork of Small Measures.
Washington, DC: National Low-Income Housing Coalition.

Fannie Mae Foundation

Culhane_app.qxd

5/28/02

10:31 AM

Page 163

Public Service Reductions Associated with Supportive Housing

163

U.S. Department of Housing and Urban Development, Office of Policy Development
and Research. 1994. National Evaluation of the Supportive Housing Demonstration
Program. Washington, DC.
U.S. Senate. 2000. Departments of Veterans Affairs and Housing and Urban Development, and Independent Agencies Appropriations Bill, 2001. Senate Report 106 410.
World Wide Web page <http://frwebgate.access.gpo.gov/cgi-bin/getdoc.cgi?dbname=
106_cong_reports&dcid=fr:sr410.106pdf> (accessed February 23, 2001).

Housing Policy Debate

Culhane_app.qxd

5/28/02

10:31 AM

Page 164

 

 

PLN Subscribe Now Ad
Advertise here
Disciplinary Self-Help Litigation Manual Side