Administration and Policy in Mental Health, Vol. 32, No. 1, September 2004 ( 2004)
UTILIZATION OF PUBLIC MENTAL HEALTH
SERVICES BY ADULTS WITH SERIOUS
MENTAL ILLNESS
Jeffrey A. Buck, Judith L. Teich, Linda Graver,
Don Schroeder, and Dian Zheng
ABSTRACT: Public mental health (MH) services were examined for non-elderly adults
with serious mental illness (SMI) using a database combining information from Medicaid,
MH, and substance abuse agencies in three states. These data show that between 23%
and 39% of those with SMI received MH services only through Medicaid. Relative use of
community versus state hospitals for delivery of psychiatric inpatient care varied across
the three states. However, state hospitals accounted for a large proportion of total inpatient days, due to high mean annual days of care. In two states, Medicaid paid for fewer
psychiatric inpatient days than expected.
KEY WORDS: Medicaid; program data; public mental health services; serious mental illness (SMI); state mental health agency.
Of all the public spending on mental health (MH) services, more than
70% occurs in programs administered at state and local levels (Coffey
et al., 2000). These dollars support services for the uninsured, the indigent, and those with disabilities. Medicaid accounts for slightly more than
half of this spending, with most of the rest consisting of programs
funded directly by state or local governments (Buck, 2003).
Although Medicaid is the primary payer of public MH services at the state
and local levels, information about this spending and its role in the overall
public system is limited. No state Medicaid agency produces regular comprehensive reports on its MH services spending, and federal policy does
not require such accounting. Further, only a few studies have examined
Jeffrey A. Buck, Ph.D., is Associate Director for Organization and Financing, and Judith L. Teich,
M.S.W., is a Social Science Analyst, both at SAMHSA/CMHS in Rockville, MD. Linda Graver is a
Senior Research Manager, Don Schroeder, Ph.D., is a Senior Research Manager, and Dian Zheng is
a Programmer/Analyst, all at The MEDSTAT Group in Santa Barbara, CA.
Address for correspondence: Jeffrey A. Buck, Ph.D., SAMHSA/CMHS, 5600 Fishers Lane, 15-87,
Rockville, MD 20857 . Fax: 301-480-8296. E-mail: jbuck@samhsa.gov.
3
2004 Springer Science+Business Media, Inc.
4
Administration and Policy in Mental Health
MH services within one or more states. Recent studies generally show that
users of MH (and substance abuse [SA] treatment) services account for
7–13% of Medicaid enrollees, and that such services make up 7–11% of
Medicaid expenditures (Buck, Teich, Bae, & Dilonardo, 2001; Larson
et al., 1999; Wright, Smolkin, & Bencio, 1995).
Information about most of the rest of state-based public spending is
also limited. Much of this spending comes from state mental health
authorities (SMHAs). In 1997, SMHAs controlled $16 billion in MH services expenditures, of which 29% came from Medicaid (Lutterman, Hirad, & Poindexter, 1999). However, little detail is available about where
this spending went, other than broad categories of services. More importantly, nothing is available concerning the overlap of this spending with
that from Medicaid.
New data now offer the ability to address some of these problems and
provide a better picture of state-based public MH services spending.
Known as the integrated database (IDB) project, this effort has combined
data from Medicaid and state MH and SA authorities in three states from
1996 through 1998. The resulting data allow analyses of many questions
concerning the individual contribution of programs administered by state
MH and SA agencies, and their relationship to Medicaid-funded MH/SA
services.
Initial results from this project show that, depending on the state,
about a quarter to a half of all persons receiving state-supported MH services receive them solely through Medicaid (Coffey et al., 2001). Between
13% and 22% receive such services both through Medicaid and the
SMHA. Clients were predominantly female, regardless of their source of
support. About three-quarters or more only received outpatient services.
Despite the value of these initial results, many important questions
remain unanswered concerning patterns of utilization and financing of
state-supported MH services. One such question concerns the treatment
of adults with serious mental illness (SMI). This group is generally considered a high-priority population for the receipt of public MH services,
particularly for those services supported through SMHAs. However, many
such individuals qualify for Medicaid coverage due to psychiatric disability. Using 1998 data from the IDB, this study examines the sources of
support for adults with SMI and their utilization of state-supported MH
services, regardless of source.
METHOD
The IDB consists of data for the states of Delaware, Oklahoma, and
Washington. These states were selected based on their interest in the project and the ability of their data systems to link clients across agencies. The
J. A. Buck, J. L. Teich, L. Graver, D. Schroederer, and D. Zheng
5
IDB contains information on MH and SA clients for all behavioral health
and physical health services through Medicaid, or programs supported by
state MH or SA agencies. All persons who received clinical treatment services from programs of state MH and SA authorities are included in the
IDB. For Medicaid data, diagnostic or other indicators of MH or SA treatment were used to identify MH or SA users. In addition to service and cost
information, the database includes information on demographics, client
history, Medicaid eligibility, diagnoses, and providers.
About a quarter to a half of all persons receiving state-supported MH
services receive them solely through Medicaid.
To comply with federal and state requirements governing data sharing
and use, non-disclosure agreements were signed with each state to protect confidential information. The most stringent requirements of any
state were applied to the data for all states. To ensure confidentiality,
after record linkage was performed, all personal identifiers and all personally identifiable data were dropped from the database, and new
encrypted personal identifiers were assigned. Each of the three states represented in the IDB had a different organizational arrangement and benefit structure for administering public behavioral health services. Coffey
et al. (2001) describe the characteristics of each state’s programs and
data, and provide additional detail concerning the creation of the IDB.
Several of these characteristics have implications for the understanding
and interpretation of the findings presented in this paper. Delaware did
not have particular program policies or data limitations that would lead
to unusual results. For Oklahoma, however, Medicaid coverage of inpatient psychiatric care was limited to 12 days a year in 1998. This meant
that the large majority of all inpatient psychiatric care was paid by the
MH authority, even for those with Medicaid eligibility. Finally, Washington’s
data had limitations that affected results in some instances. Information
on outpatient services provided through its managed care program only
allowed the determination that an individual had received a service, but
provided no diagnostic information or service detail (e.g., number of
visits). No diagnostic information was provided on nursing facility
records. While the state MH agency paid for some non-hospital
residential care, individual data were not collected from such facilities.
The chief impact of these limitations was that there was an under-identification of those with SMI due to limited diagnostic information, and an
inability to determine use of nursing facilities or residential care by the
study group.
6
Administration and Policy in Mental Health
Two other issues related to the understanding of results are worth noting. The first concerns the identification and treatment of possible duplicate service records that appear in more than one state agency’s data
files. Counts of individuals were unduplicated, and good information
about dates of services allowed most records for inpatient or residential
settings to be unduplicated. However, outpatient or other service records
often lacked specific dates of service, or other details that could allow
accurate identification of potential duplication. Therefore, the possibility
of double-counting some Medicaid and state agency outpatient services
records cannot be fully eliminated.
The second issue concerns the availability and use of managed care
encounter data within the IDB. All three states provided encounter data
for their Medicaid population. As a result, persons receiving services
through managed care programs were included in the analytic database
if they met the criteria for selection. Nevertheless, managed care encounter data are not always complete, and no separate studies were conducted
to assess the completeness of the encounter data within individual state
agencies’ databases.
Although results for the three states in this analysis are presented side
by side, the intent is not to compare these data across the states. Rather,
it is to provide a more complete picture of how individual states organize
and finance services. State programs for delivering MH/SA programs differ along many dimensions, and these differences make it impossible to
ascribe any particular finding across the states to a specific underlying
factor.
Selection Criteria for SMI Population
The analyses presented here are for a subset of individuals in the IDB
identified as having SMI. For inclusion in this analysis, two types of criteria had to be met. First, an individual must have had one of the following
as his/her most frequent primary diagnosis (numbers in parentheses
indicate ICD-9-CM codes):
•
•
•
•
Schizophrenia (295)
Affective psychoses (296.0, 296.1, 296.4–296.99)
Major depression (296.2, 296.3)
Other psychoses (297, 298, 299)
These diagnoses include all schizophrenic disorders; all affective psychoses, including major depression and bipolar affective disorder; and
other psychoses including paranoid states, other non-organic psychoses,
and psychoses with origins specific to childhood.
J. A. Buck, J. L. Teich, L. Graver, D. Schroederer, and D. Zheng
7
Across the three states, schizophrenia was the most frequent primary diagnosis for between 42% and 56% of the SMI group.
SMI is not identified solely through diagnosis, but also depends on the
existence of concurrent functional limitations. However, administrative
data do not generally provide independent information on such limitations. Accordingly, this was approximated by determining if the individual had one or more of the following indicators of disability or high
service use:
• Medicaid-eligible due to disability (SSI)
• Number of Medicaid outpatient services at the 90th percentile or
greater for the total MH population in the state
• Psychiatric care of 15 days or more a year in a nursing facility or hospital
• Number of state agency community services at the 90th percentile or
greater for the total state agency MH population outpatient records
in the state
Across the three states, for ages 21–64, the proportion of MH service
users in the database who had one of the four SMI diagnoses varied from
21.1% in Washington to 49.7% in Delaware. When the additional criteria
were applied, the proportion of MH service users aged 21–64 in the database represented by the SMI study group ranged from 16.0% to 20.3%.
The majority of persons in the SMI group qualified for Medicaid on the
basis of disability. This proportion ranged from 50.2% to 80.4% across
the states.
Across the three states, schizophrenia was the most frequent primary
diagnosis for between 42% and 56% of the SMI group. Those with a
diagnosis of affective or other psychosis constituted between 21% and
35% of the study group, and those with a diagnosis of major depression
ranged from 22% to 27%.
RESULTS
In all three states, the average age of individuals in the SMI group was
between 42.0 and 42.9 years. As shown in Table 1, the group in each
state was nearly equally divided between males and females. With regard
to ethnicity, the group was predominantly White, ranging from 57.4% in
Delaware to 82.6% in Washington. African–Americans were the next most
frequent category, representing between 6.9% and 36.6% of persons in
8
Administration and Policy in Mental Health
TABLE 1
Demographic Characteristics of SMI Population, 1998 (Percent)
Age 21–64
Gender
Male
Female
Race/ethnicity
White
African–American
Native American
Asian/Pacific Islander
Hispanic
Other
Delaware
Oklahoma
Washington
n = 1,623
n = 8,489
n = 12,265
47.6
52.4
48.3
51.7
50.4
49.6
57.4
36.6
0.2
0.4
3.5
1.9
75.1
16.5
6.2
0.6
1.5
0.0
82.6
6.9
1.9
3.1
2.0
3.6
the SMI group. There was no consistency in the relative proportions of
the remaining groups.
The proportion of the SMI study group receiving MH services solely
supported by the Medicaid agency, solely supported by the state MH
agency, or both, is displayed in Table 2. Individuals appearing in either
the Medicaid or state MH agency categories are those whose recorded
MH services only appear in one or the other agency’s database. Those
who have service records in both databases, whether duplicated or not,
are included in the ‘‘both’’ category. This includes persons who received
Medicaid services administered by the state MH agency. Although some
TABLE 2
Payment Sources of MH Services for SMI Population, 1998
State MH
Medicaid
Agency Onlya
Onlya
(%)
(%)
State
Delaware
Oklahoma
Washingtonb
a
37.8
39.0
22.6
52.1
48.8
3.1
Both Medicaid
Total SMI
and State
Population
MH Agencya
(%)
(Number of Persons)
10.2
12.2
74.3
1,623
8,489
12,265
Persons are only reported once within these three categories.
Clients may be underrepresented in the State MH Agency Only category, as diagnoses are not
reported for outpatient services (a diagnosis is required for selection into the SMI study population).
b
J. A. Buck, J. L. Teich, L. Graver, D. Schroederer, and D. Zheng
9
of the individuals in the ‘‘both’’ category have duplicated services that
could not be clearly attributed to one state agency or the other, the large
majority in this category represent individuals who received separately
identifiable services from each agency.
For 22.6–39.0% of the individuals with SMI, Medicaid was the only
source of payment for MH services. Between 10.2% and 74.3% of the
SMI group received services from both Medicaid and the SMHA. In
Washington, only 3.1% of the SMI group received MH services solely
from the SMHA, while in Delaware and Oklahoma about one-half of the
SMI group received MH services solely funded through the SMHA.
Co-occurring SA is an important health and treatment issue for persons with SMI. The proportion of the study group using any SA treatment service from any payer was identified through diagnosis or type of
service. This included SA services supported by the state SA agency. The
results were fairly consistent across states. Between 9.1% and 14.1%
showed some evidence of treatment for SA, from any payer. Between
5.3% and 11.3% of the SMI group received at least one identifiable SA
service from the state SA agency.
General categories of MH service utilization are shown in Table 3. The
table also displays the total annual mean and median use for all those
having at least one instance of use for each category of service. Due to
differences in coding, reporting, and record keeping, there is a considerable amount of heterogeneity in the outpatient/community category,
with some records representing a single visit and others a period of treatment. Therefore, this category only shows the mean and median numbers of such records. No data are presented in the nursing facility
category for the state of Washington, since those records do not carry
diagnostic information necessary to determine if the stay was for psychiatric reasons.
Overall, about a third of the SMI group had some indication of inpatient psychiatric care in either community or state hospitals (32–38%).
Only 6% or less of psychiatric inpatient users showed receipt of care
over the year in both types of facilities. No consistent pattern was found
in the relative use of community versus state inpatient facilities. In Delaware, a higher proportion of the SMI group received inpatient care in
state facilities. In Oklahoma, the proportion of the SMI group receiving
care in state facilities and community facilities was approximately the
same, while in Washington, a higher proportion of the SMI group
received inpatient care in community hospitals. Median annual days of
care in community hospitals was between 7 and 13 days, suggesting that
these settings are primarily for acute care. This is not the case for state
hospitals, where median annual days of care ranged from 28 to 97
across the three states.
10
Community Hospitala
% of
Annual Days
of Careb
State Hospitala
% of
Annual Days
of Careb
Nursing Facilitya
% of
Annual Days
of Careb
Outpatient/Community Servicesa
% of
SMI Pop Mean Median SMI Pop Mean Median SMI Pop Mean Median SMI Pop
State
Delaware
Oklahoma
Washington
a
8.8
17.9
24.7
9
12
23
7
9
13
30.1
19.7
12.9
151
61
132
79
28
97
3.4
21.9
–
316
113
–
364
29
–
Persons may be reported in more than one category.
Mean and medians are for those having at least 1 day or service visit.
c
Service records include all types of outpatient services, e.g., day treatment, individual therapy, medication, etc.
b
79.6
97.8
96.7
Number of
Service Recordsc
Mean
121
75
81
Median
22
30
36
Administration and Policy in Mental Health
TABLE 3
Utilization of MH Services by SMI Population, 1998
J. A. Buck, J. L. Teich, L. Graver, D. Schroederer, and D. Zheng
11
TABLE 4
Total Days of Inpatient Psychiatric Care for SMI Population, 1998
State
Delaware
Oklahoma
Washington
Community Hospital
% of Inpatient Days
State Hospital
% of Inpatient Days
Total Number
of Days
1.7
15.0
25.6
98.3
84.9
77.2
75,072
120,383
271,764
Note: Percentages may not add to 100 due to overlaps in discharge and admission dates between
community and state hospitals.
Information on utilization of nursing facilities was available for two
states. In Oklahoma, the relative proportion of the SMI group using nursing facilities is similar to those for either type of inpatient hospital care,
and the median number of days was similar to the median for state hospitals. In Delaware, the proportion of the SMI group using nursing facilities was low, but the median number of days was much higher than the
median for state hospitals.
Due to the heterogeneity of outpatient records, it is difficult to make
general observations about the use of such services. In two states, almost
everyone in the study group had some evidence of outpatient psychiatric
service use. In the other state, this proportion was about four-fifths. The
median number of outpatient records per user varied between 22 and
36, indicating that the typical user with SMI had multiple instances of
outpatient care during the year. More information about use of these services will have to await further refinement of the database and subsequent analysis.
Tables 4 and 5 provide additional detail about hospital use indicated
for this population and payment for such use. While Table 3 displays the
TABLE 5
Medicaid Support of Inpatient Psychiatric Care for SMI Population, 1998
% of Inpatient Psychiatric Days Supported by Medicaid
State
Delaware
Oklahoma
Washington
a
Community
Hospital
State Hospital/
Institution
All Inpatient
Psychiatric Daysa
100.0
34.9
33.5
0.2
0.1
0.0
1.9
5.4
8.6
Total hospital days are shown in Table 4.
12
Administration and Policy in Mental Health
number of persons with SMI using state and community hospital psychiatric care, Table 4 shows the distribution of annual psychiatric inpatient
days. The use of this variable provides greater weight for inpatients with
greater annual use and better reflects the actual resources expended for
the inpatient care of this population. The results show that, in all three
states, the large majority of inpatient days are spent in state facilities.
However, the proportion of inpatient days recorded in state hospitals is
noticeably greater than the proportion of the SMI group using such settings. This reflects the longer-term nature of the treatment provided in
these facilities.
The proportion of psychiatric inpatient days paid for by Medicaid is
presented in Table 5. Although most of the SMI group in each state were
eligible for Medicaid, the proportion of all inpatient psychiatric days for
the SMI group that was supported by Medicaid ranged from 1.9% to
8.6% across the three states. For community hospitals, Medicaid’s share
of payment ranged from one-third to 100%, depending on the state. In
contrast, the proportion of inpatient days in state hospitals paid for by
Medicaid was between 0% and 0.2%. Both these results may be explained
by the Medicaid policy excluding payment for psychiatric institutions that
provide services to those between ages 22 and 64 ( ‘‘the Institution for
Mental Diseases [IMD] exclusion’’). Such institutions include state mental hospitals and private psychiatric hospitals, whose services would be
represented in the community hospital category.
Variability in the remainder of the results may be partly explained by
differences in each state’s payment policies and the contributions of private psychiatric facilities (IMDs) to the community hospital category. In
1998, Oklahoma’s Medicaid program only paid for 12 days annually of
inpatient hospital care of any kind, with any additional psychiatric inpatient care covered by the SMHA. Use of private psychiatric hospitals and
under-reporting of Medicaid-covered care in Washington may partly
account for the lower percentages there.
DISCUSSION
State and local governments manage a substantial portion of all SA
and MH treatment dollars. Multiple state agencies often support treatment of the same clients, and multiple data systems collect information
on them. No one state agency has data on the full spectrum of publicly
supported MH and SA treatment services. Creation of an IDB that combines Medicaid, state MH, and state SA agency data for three states represents an effort to address this situation. This paper has presented
J. A. Buck, J. L. Teich, L. Graver, D. Schroederer, and D. Zheng
13
findings for one subset of persons represented in IDB records for 1998,
adults with SMI.
Caution should be exercised in the use and interpretation of results
from this study. While some findings appear to be fairly consistent across
the three states, it is not known to what degree they represent the experience of others. Further, the difficulty of creating the IDB from individual
and heterogeneous data files means that there are areas of missing data
or other anomalies. Nevertheless, the IDB represents the best current
source of information for understanding the characteristics and interaction of state MH services supported by MH and Medicaid agencies.
About a third of the SMI group had some indication of inpatient psychiatric care in either community or state hospitals.
Similarly, for several reasons, some of these results may differ from the
experience of other states, or from what some may believe is known
about public MH services. First, these results are for persons with SMI,
and do not constitute a description of all public MH services utilization
in these states. Second, there is no consensus about how to identify persons with SMI using agency administrative data. The definition used in
this study may vary from other studies, or from definitions used in other
states for program purposes. In particular, the use of service data to identify those with SMI partly biases the findings reported for service utilization. Finally, the inclusion of data for individuals who receive MH
services through Medicaid as well as from SMHAs mean that these results
will likely differ from information derived solely from SMHA sources.
While all of this study’s findings contribute to understanding the treatment of individuals with SMI, several are particularly notable. The most
important is the demonstration that public MH services for persons with
SMI are best characterized as a system shared between Medicaid and MH
authorities. For between 23% and 39% of persons with SMI, Medicaid
was the only payment source for MH services. In one state, three-quarters
of persons with SMI received services funded by both the SMHA and
Medicaid. These findings underscore the importance of efforts to better
incorporate Medicaid data and policies into state MH planning and
administration.
Another issue concerns treatment for co-occurring disorders. One
study shows that between 15% and 27% of persons with SMI have cooccurring SA, depending on the definition of severity (Kessler et al.,
1996). Findings from the current study show that the percentage of persons with SMI having undergone SA treatment is at the lower end of this
range. Further, it appears that the state SA agency plays a major role in
14
Administration and Policy in Mental Health
supporting such treatment. This raises further questions that should be
addressed concerning the effectiveness of screening for SA problems
among this population and the coordination of treatment for them.
State and local governments manage a substantial portion of all SA and
MH treatment dollars.
Finally, results from the current study provide additional insights into
the provision and support of inpatient psychiatric care for this population. Community hospitals appear to be primarily sites for acute psychiatric care, while state hospitals (and sometimes nursing homes) function as
long-term care sites. There is no apparent consistency in the relative use
of these settings across states. In data from two states, the proportion of
community hospital days reimbursed by Medicaid is lower than would be
expected, given that half or more of those in the study group have Medicaid eligibility. This suggests that there still are areas of service delivery
where states may have the ability to shift financing to Medicaid to obtain
federal matching funds.
This study shows the potential importance and value of efforts to integrate administrative data to assess the full range of state-supported MH
services. Future analyses will attempt to learn more about utilization patterns and spending for MH services, as well as specifics about the providers of these services. In addition to providing better information about
state MH services, it is hoped that the IDB project will encourage other
states to explore the utility of combining administrative data to better
understand their own systems.
While illustrating the potential benefits of data integration, this study
also highlights the limitations of the component databases. Specifically,
the databases utilized in this study all were restricted to some degree in
their quality and completeness. In particular, data from Medicaid and
MH agencies systematically differed in their organization, record elements, and specificity, even when recording similar service events. As a
result, it is clear that the full benefits of data integration will not be
entirely realized without corresponding efforts to identify and eliminate
such differences.
REFERENCES
Buck, J. A. (2003). Medicaid, healthcare financing trends, and the future of state-based public mental
health services. Psychiatric Services, 54, 969–975.
Buck, J. A., Teich, J. L., Bae, J., & Dilonardo, J. (2001). Mental health and substance abuse services
in ten state Medicaid programs. Administration and Policy in Mental Health, 28, 181–192.
J. A. Buck, J. L. Teich, L. Graver, D. Schroederer, and D. Zheng
15
Coffey, R. M., Graver, L., Schroeder, D., Busch, J. D., Dilonardo, J., Chalk, M., & Buck, J. A. (2001).
State mental health and substance abuse treatment patterns: Results from the integrated database project for
three states (SAMHSA Publication No. SMA-01-3528). Rockville, MD: Center for Substance Abuse
Treatment and Center for Mental Health Services, Substance Abuse and Mental Health Services
Administration.
Coffey, R. M., Mark, T., King E., Harwood, H., McKusick, D., Genuardi, J., Dilonardo, J., & Buck, J.
(2000). National estimates of expenditures for mental health and substance abuse treatment, 1997 (SAMHSA
Publication No. SMA-00-3499). Rockville, MD: Center for Substance Abuse Treatment and Center for
Mental Health Services, Substance Abuse and Mental Health Services Administration.
Kessler, R. C., Berglund, P. A., Zhao, S., Leaf, P. J., Douzis, A. C., Fruce, M. L., Friedman, R. M.,
Grosser, R. C., Kennedy, C., Narrow, W. E., Kuehnel, T. G., Laska, E. M., Manderscheid, R. W.,
Rosenheck, R. A., Santoni, T. W., & Schneier, M. (1996). The 12-month prevalence and correlates
of serious mental illness (SMI). In R. W. Manderscheid & M. A. Sonnenschein (Eds.), Mental
health, United States, 1996 (SAMHSA Publication No. SMA-96-3098). Rockville, MD: Center for Mental
Health Services, Substance Abuse and Mental Health Services Administration.
Larson, M. J., Farrelly, M. C., Hodgkin, D., Miller, K., Lubalin, J. S., Witt, E., McQuay, L., Simpson, J.,
Pepitone, A., Keme, A., & Manderscheid, R. W. (1999). Payments and use of services for mental
health, alcohol, and other drug abuse disorders: Estimates from Medicare, Medicaid, and private
health plans. In R. W. Manderscheid & M. Henderson (Eds.), Mental health, United States, 1998
(DHHS Publication No. SMA-99-3285). Washington, DC: US Government Printing Office.
Lutterman, T., Hirad, A., & Poindexter, B. (1999). Funding sources and expenditures of state mental health
agencies: Fiscal year 1997. Alexandria, VA: National Association of State Mental Health Program
Directors Research Institute.
Wright, G., Smolkin, S., & Bencio, D. (1995). Medicaid mental health and substance abuse 1992, use and
expenditure estimates for Michigan and California: Final report. Washington, DC: Mathematica Policy
Research.