|
D A T A W A T C H :
I N S U R A N C E C O V E R A G E
W E B E X C L U S I V E
4 June 2003
Covering The Uninsured:
How Much Would It Cost?
The cost of additional medical
care used by newly
insured Americans would be lower than most people think, this analysis confirms.
by Jack Hadley and John Holahan
ABSTRACT:
To provide benchmarks for evaluating the costs of alternative proposals to
provide insurance coverage for the uninsured, this study presents two sets of
cost estimates derived from medical spending patterns of lower- or middle-income
people with private insurance plans and those of people with public insurance
coverage during 1996–1998. The analysis suggests that the uninsured would
use $33.9–$68.7 billion (in 2001 dollars) in additional medical care if
they were fully insured. An increase in medical spending of this range would
increase total health care spending by 3–6 percent and would raise health
care’s share of GDP by less than one percentage point.
It is well documented that
having insurance increases medical care use.1 Consequently,
a critical question in the ongoing national debate over whether and how to extend
insurance to the uninsured is, “How much more will it cost to insure the
uninsured, over and above what is being spent now for their medical care?”
This question has several components. How much will the increased medical care
used by newly insured people cost? How much will government spending go up,
both to pay for the cost of the additional services used by the uninsured and
to cover “cost transfers” for care that was either subsidized from
private sources or paid for out of pocket by the uninsured? How much will government
spending increase because of “crowding out,” which occurs when people
switch from private insurance to expanded public insurance?
This paper focuses on the first of these questions: the cost of the additional
care that would be used by people who obtain full-year coverage after being
uninsured for either all or part of a year. It builds on a recent analysis of
the current cost of medical care used by the uninsured and its sources of financing.2
It does not address increased government costs that would inevitably occur because
some people would give up their current private coverage to enroll in the new
program, the crowding-out effect. The number of people who would switch would
depend on the specific design of the expanded insurance program.
Our analysis estimates the cost of increased medical care used by the uninsured
under two alternative assumptions: The newly insured’s spending would be
similar to that of either lower- or middle-income people covered by the “average”
private insurance policy, or people covered by the “average” public
insurance policy (primarily Medicaid and the State Children’s Health Insurance
Program, or SCHIP, but also including similar state-funded public insurance
programs).
Prior studies have either (1) estimated the effects of insurance on the use
of specific services and then applied estimates of the cost per service, or
(2) specified the detailed services and benefits of hypothetical insurance plans,
applied actuarial valuations to develop the “cost” of the plan, and
then multiplied by the number of people who would be covered by the new plan.
As an example of the first approach, Stephen Long and Susan Marquis estimated
statistical models of the effect of insurance coverage on two services: ambulatory
care contacts and hospital admissions.3 They estimated
the cost by multiplying the projected increases in the numbers of contacts and
admissions by national estimates of the cost per ambulatory contact and per
hospital admission. According to their estimates, spending for these two services
by the full-year uninsured would increase about 50 percent. In another example,
Pamela Farley Short and colleagues expanded the list of services to ten.4
They estimated a somewhat larger increase in spending, 73 percent.
As examples of the latter approach, Lewin-VHI applied actuarial estimates of
insured people’s per capita spending by age, sex, health status, and income
to corresponding categories of uninsured people to estimate the cost of President
Clinton’s Health Security Act.5 In a related
analysis, the cost implications of a managed competition insurance plan were
estimated by multiplying the number of affected people by the average annual
premium of an efficiently operated health maintenance organization (HMO).6
Conceptual Approach
To simulate the health care spending of the uninsured if they should gain insurance
coverage, we estimated a series of statistical models that relate annual health
care spending to measures of insurance coverage, sociodemographic characteristics,
and health status. We estimated separate models that alternatively combined
a sample of uninsured people with samples of (1) lower- and middle-income people
with private insurance, and (2) people with public insurance. By using nationally
representative samples, this approach assumes that the uninsured would have
coverage similar to the average private or public insurance policy. We adopted
this strategy because the specifics of plan design, which reflect values about
what health insurance should cover and how it should be paid for, are both complex
and controversial. Our goal is to establish benchmarks that can be used to compare
cost estimates for possible future specific proposals for expanding insurance
coverage against the cost of an “average” private or public insurance
plan.
Within this basic goal, we distinguished between private and public insurance
because the two typically have distinctive features. Private insurance generally
incorporates cost sharing through deductibles, coinsurance, and copayments;
offers a range of covered services; and provides access to a broad set of providers
under varying payment rates. Public insurance other than Medicare typically
incorporates very little patient cost sharing and covers a broad range of services
but limits access to a more narrow set of providers who are willing to accept
lower payment rates.
The coefficients of the statistical models estimated using the combined samples
of uninsured and insured people implicitly assume that the effects of sociodemographic
and health characteristics reflect an average of the care-seeking and medical
spending behavior of the uninsured and the comparison insured population (either
private or public). An alternative approach, for example, could estimate a spending
model using only data for privately insured people and then apply the coefficients
from that model to the characteristics of the uninsured. However, this approach
makes the unrealistic assumption that the sociodemographic and health characteristics
of the uninsured have the same effects on their spending that the characteristics
of the privately insured have on their spending. In other words, this approach
ignores possible differences in care-seeking behavior that are attributable
to sociodemographic differences between the uninsured and the privately (or
publicly) insured. Thus, in our simulations, differences in predicted expenditures
between the private and public insurance models are attributable to a combination
of (1) differences in the effects of each type of insurance coverage on medical
spending, and (2) differences in the characteristics of the uninsured relative
to people with full-year private or public insurance.
We excluded higher-income people from the full-year privately insured sample,
to avoid confounding from possible differences in care-seeking behavior resulting
from differences in socioeconomic status between the uninsured and higher-income
people with full-year private coverage. For example, people with higher incomes
might be more likely to use more costly, out-of-network providers than would
lower-income people with the same insurance coverage, and they should in general
be less deterred by cost sharing. Thus, their behavior would not be a good basis
for predicting how the uninsured would respond to having coverage.
We next simulated predicted spending by setting the value of the insurance coverage
variable to full-year coverage (either private or public) and dropping the full-year
insured population from the prediction sample. Thus, the predictions are based
on the characteristics of the uninsured population under the assumptions that
they have coverage for a full year and that the effects (coefficients) of sociodemographic
and health characteristics reflect the average behavior of the uninsured and
of the specific insured sample used to estimate the statistical models.
Study Methods
Data and sample.
All data for this analysis are from the Medical Expenditure Panel Surveys (MEPS)
conducted in 1996, 1997, and 1998. MEPS is a nationally representative sample
of the noninstitutionalized population, which contains detailed information
on annual total charges and payments for health care used, monthly information
on insurance coverage, and detailed demographic and health characteristics.7
The analysis sample excludes people age sixty-five or older and nonelderly people
covered by Medicare, as well as privately insured families with incomes higher
than 400 percent of the federal poverty level.8
People who are in the sample for only a portion of the year (newborns, people
who died, and people who became institutionalized) are included for the part
of the year they were in the sample. Data from the three surveys were pooled
to form a single analysis file. Expenditures were inflated to 2001 dollars using
the annual percentage increase in the National Health Accounts (NHA) of the
Centers for Medicare and Medicaid Services (CMS).9
Aligning MEPS data with
the NHA. MEPS
defines expenditures as explicit payments (as opposed to charges) made for health
care services provided to a specific patient. MEPS does not count provider revenues
from general government appropriations and from programs such as the Medicare
and Medicaid disproportionate-share hospital (DSH) programs, since they are
not payments for specific patients. As a result of these and other definitional
differences, the MEPS estimate of total national health spending is much lower
than the NHA estimates.
To correct for MEPS’s systematic underreporting, we used information from
a detailed comparison of the MEPS and NHA estimates to develop an adjustment
factor to align the MEPS estimates with the NHA.10
We first subtracted the estimates of Medicare spending from both sources, since
they are not direct payments for care received by the uninsured. We also subtracted
from the NHA expenditures attributed to the Department of Defense for military
personnel, revenues from non–patient care activities reported by providers,
and expenditures for long-term nursing home and long-term hospital care, which
are not likely to be included as covered services by a program to extend coverage
to the uninsured. These adjustments reduce the NHA total from $912 billion to
$556.1 billion. The MEPS/NHA adjustment factor we use to inflate the MEPS estimates
is 1.25, the ratio of the revised NHA spending level to the comparable MEPS
spending level.
Simulation model and statistical
estimation. We
used a standard two-part approach to estimate the simulation model, to account
for the fact that a large proportion of people incur no health care expenses.11
The first part uses a logistic model to estimate the probability of having any
spending during the year. The second part estimates the effects of insurance
and other characteristics on spending, given that the person has incurred some
expenses. We estimated separate models for children (under age 19) and adults
(ages 19–64), because we used different measures of health conditions for
children and adults and because health insurance affects children’s and
adults’ medical spending differently. All data were weighted using the
MEPS person weights for both estimating the models’ coefficients and simulating
predicted spending. We used the STATA software program for statistical estimation
and computations.12
To simulate spending for the uninsured under the assumption that they have either
private or public coverage, we assigned the variable measuring the percentage
of time covered by either private or public coverage a value of 1.0, which represents
full-year coverage. We then combined the estimated coefficients from the expenditure
models with the uninsured’s values of the independent variables to predict
both the probability of having any spending and the amount spent (for people
with positive expenditures).13
Independent variables in
the statistical models.
Insurance coverage. The key independent variables in the statistical
models are the percentage of months that a person has private insurance coverage
and the percentage of months that a person has public insurance coverage. Measuring
insurance coverage in this way improves the accuracy of the predictions because
a substantial number of Americans are uninsured for only a portion of a year.14
Being uninsured for only one or two months may have little impact on health
spending, while being uninsured for ten or eleven months increases the likelihood
of forgoing care for financial reasons. Both measures were included in all of
the models because the population of uninsured people includes people who have
some private or public coverage, or both, for a portion of the year. In addition,
a very small proportion of people with full-year coverage had a combination
of private and public coverage over the course of the year.15
Sociodemographic characteristics. The statistical models for adults include
sets of dichotomous variables for sex, age, race and ethnicity, education, family
income relative to poverty, and marital status. The models for children control
for sex, age, race and ethnicity, family income relative to poverty, and parents’
education and marital status. All models also include controls for census region.16
Health characteristics. Although MEPS contains detailed information on
the presence of both acute and chronic conditions, there is some concern that
these may be underreported for the uninsured because they tend to have fewer
contacts with medical care providers. Therefore, for adults we used a combination
of self-reported general health, mental health, and functional status measures,
along with measures of acute and chronic conditions derived from contacts with
providers. We used a smaller and different set of health measures for children
because they have a much lower incidence of specific medical conditions. Finally,
we included a dichotomous indicator of whether the person died or was institutionalized
for some portion of the year.17
Study Results
Differences in population
characteristics by insurance status.
Exhibits
1 (adults) and 2
(children) report the mean values of total spending and population characteristics
of the different samples of people used to estimate the statistical models.
The uninsured sample is divided into people uninsured for the full year and
for part of the year. The part-year uninsured have coverage for about 55 percent
of the year, and most of that coverage is private insurance (tabulations not
shown).
Full-year uninsured adults and children are much less likely than any of the
other groups are to have any expenditures over a year, and they spend much less
per person. Among the full-year uninsured, 58 percent of adults and 64 percent
of children have any expenses, and average spending per person is $1,158 for
adults and $475 for children.18 Among the privately
insured, 85 percent of adults and 86 percent of children had expenses, and average
spending per person was $2,970 for adults and $1,492 for children.
Adults with full-year public insurance coverage have the highest spending per
person, but this largely reflects their much higher incidence of fair and poor
self-reported health status, as indicated by their higher proportions with functional
and activity limitations and specific medical conditions (Exhibit
1). Although there are more full-year-uninsured adults who report being
in fair or poor health than among the full-year privately insured, smaller proportions
of the full-year uninsured report having specific health conditions. This paradox
may be the result of this group’s fewer contacts with health care providers
and subsequent underreporting of diagnosed acute and chronic conditions.
Compared with the privately insured, the uninsured are more likely to be racial
and ethnic minorities, have lower educational attainment, and have lower family
incomes relative to poverty. People covered by public insurance also have very
different sociodemographic characteristics than privately insured people have.
Simulated increases in per
capita and total medical spending.
The complete statistical models are reported elsewhere.19
The insurance coverage variables were positive and statistically significant
(p < .01) in all models, which indicates that having coverage raises
total spending and that people with full-year coverage spend more than those
with part-year insurance coverage spend. Although coefficients vary from model
to model, it is generally the case that racial and ethnic minorities and people
with less education have lower expenditures. Poor health—whether measured
by self-reported health status, activity and functional limitations, or specific
acute or chronic conditions—is associated with greater spending.
Exhibit
3 reports the simulated impact of insurance coverage on medical expenditures
per uninsured person. Baseline figures include out-of-pocket payments, insurance
payments for people with part-year coverage, and identified sources of uncompensated
care (such as from public hospitals and clinics, workers’ compensation,
and local welfare programs), but they do not include uncompensated care paid
for by implicit sources, such as general government payments (appropriations,
grants, Medicare and Medicaid DSH) to private providers, private philanthropy,
or providers’ financial surpluses. Prior research indicates that this would
add about 15 percent to the estimate of baseline per capita spending by people
uninsured any part of the year.20 (The value of
all uncompensated care is accounted for in projecting the aggregate increase
in incremental spending associated with complete insurance coverage.)
Estimates of simulated spending in Exhibit
3 reflect the effects of insurance on increasing both the likelihood of
having any spending and a higher level of spending, given that some spending
occurs. Combining adults and children who are uninsured for at least one month
of the year, the simulations predicted that annual spending nearly doubles,
from $1,383 to $2,676 per person, under the assumptions that the expanded insurance
coverage is like an average private insurance plan and that the effects of other
characteristics (age, sex, health, education, and marital status) reflect an
average of the effects for the uninsured and the privately insured.
Under the alternative assumption that the expanded coverage is similar to the
average public insurance plan, simulated spending per person increases by 53
percent, to $2,121. As expected, the increases for the full-year uninsured are
larger than those for the part-year uninsured. However, the simulated percentage
increases in spending for adults and children are fairly similar, although children’s
level of spending is much lower, both simulated and at baseline.
Exhibit
4 presents estimates of total simulated spending for the populations of
people who would gain coverage under universal insurance. Our baseline estimate
of the amount of medical care used by the uninsured is $98.9 billion, which
includes all uncompensated care (explicitly and implicitly financed), insurance
payments for people with part-year coverage, payments from other identified
sources, and the insured’s out-of-pocket payments.21
Under the assumption that coverage expansion would provide insurance
similar to the average private insurance policy observed for lower- and middle-income
people in the base period, total spending for all people uninsured any part
of the year would increase to $167.6 billion, split almost evenly between the
full-year ($86.7 billion) and the part-year uninsured ($80.9 billion). However,
the increase in total spending, $68.7 billion, is more heavily weighted toward
the full-year uninsured, whose total spending more than doubles, while that
for the part-year uninsured increases less than 40 percent.
Under the assumption that the expanded coverage would be similar to the average
public insurance plan observed at baseline, the simulated total spending is
$132.8 billion, which reflects an increase in total spending that is about half
as large, $33.9 billion, as under the assumption of expanded private insurance
coverage. For people with part-year coverage, the simulated increase is relatively
small, only about 12 percent, reflecting the assumption that existing private
coverage would be replaced by the average public insurance plan.
Simulated increases in total
charges and services used.
The regression-adjusted simulations in Exhibit
3 suggest that spending per uninsured person would be 26 percent higher
under the assumption of “average” private coverage ($2,676 per capita)
than under a plan reflecting “average” public coverage ($2,121 per
capita). As a result, the increase in aggregate spending under a plan that mimicked
average private coverage is twice as large as the simulated increase under the
assumption of public coverage (Exhibit
4).
This observation raises the question of whether the public-private difference
is attributable to differences in service use or to differences in payment rates.
To address this question, we simulated the effects of complete insurance coverage
on total office visits (to any provider and to physicians), total hospital days,
and total charges for all medical care using the same basic approach used to
simulate increased medical spending. Office visits and hospital days are direct
measures of service use. Total charges for all care received are a better measure
of the quantity of care than total payments (expenditures) are, because payments
reflect the effects of contractual allowances, insurer discounts and fee schedules,
and unpaid balances (including charges for uncompensated care). If simulated
total charges, office visits, and hospital days are similar under the alternative
assumptions about the type of coverage, then the differences in simulated expenditures
in Exhibit
4 would be attributable to differences in plans’ payment rates and
policies, not to differences in service use.
The results of these simulations (Exhibit
5) indicate that the difference in simulated total charges for medical care
is relatively small, just over 5 percent greater for private coverage. They
also show that the simulated increases in service use are actually slightly
greater under the assumption of public coverage than under private coverage,
especially for hospital days per 100 people.
The finding that simulated total charges are somewhat higher under the private
coverage assumption but service quantities are higher under the public coverage
assumption could result from the fact that privately insured people use providers
who charge more (that is, more specialists than primary care physicians) or
receive more-intensive care per visit or hospital stay (that is, more diagnostic
or surgical procedures). The key point, however, is that the simulations reported
in Exhibit
5 imply that most of the difference in simulated payments between private
and public insurance plans reported in Exhibit
4 is attributable to differences in payment rates. In other words, they
reflect the fact that Medicaid programs typically pay hospitals, clinics, and
physicians much lower rates than private plans pay.22
Discussion
As shown in our statistical appendix, the estimates are sensitive to specific
assumptions about sample specification, estimation method, variable definitions,
and model specifications.23 These variations could
affect the estimates by ± 10–20 percent (based on variations in
per capita estimates from preliminary analyses). However, the estimates of the
total increases in spending we report are very similar to unpublished estimates
made by researchers at the Agency for Healthcare Research and Quality (AHRQ),
who also used MEPS data but used a number of different methodological assumptions.24
Our estimates of the percentage increase in total health spending are also similar
to projections from earlier studies that simulated the cost of increases in
the use of specific services.25 Thus, in spite
of methodological variations across studies, our estimates are consistent with
the results of other studies that predicted an increase in total health spending
of 3–6 percent associated with expanding insurance coverage to the uninsured.
Limitations.
One methodological limitation is that the underlying statistical models do not
adjust for possible bias attributable to people’s selecting into various
types of coverage or choosing to be uninsured because of their underlying health
characteristics. For example, low-income people with serious health problems
have a strong incentive to seek public insurance coverage, while those who are
healthy may opt to forgo the expense of private coverage. In other words, those
who choose to be uninsured may be likely to use less medical care than we estimate
if there are unobservable factors that influence both their insurance coverage
and their likely use of medical care if insured. The available MEPS public-use
data make it very difficult, if not impossible, to incorporate insurance choice
into the formal model, although this should be a high priority for future research.
However, to the extent that there is bias from this source, it should be to
exaggerate the effect of coverage on the increase in medical spending by the
uninsured.
Another limitation is that our simulations do not make any assumptions about
specific features of potential plans—that is, specific services covered,
cost sharing, and provider payment method—all of which could have a substantial
impact (either positive or negative) on our estimates which are based on “average”
plan characteristics.26 For example, if expanded
private coverage were “catastrophic only” or incorporated substantial
cost sharing, then spending would presumably be less than under an average private
insurance plan because of smaller increases in service use. Moreover, high cost
sharing could make private plan premiums much lower than the costs of public
plans.
It is also important to emphasize that the simulations only provide estimates
of the cost of expanding coverage to those who are uninsured. Plans to expand
coverage will typically entail larger cost increases for government because
some privately insured people will inevitably switch to the government-subsidized
plan. The estimate of increased government spending depends on both the cost
of covering the uninsured (which we estimate) and the cost of crowding out or
displacement. The more “target-efficient” the plan (that is, the better
it addresses the costs of its target population), the lower the amount of costs
that would be transferred from private to public coverage; the less target-efficient,
the greater the transfer costs and the higher the overall costs of a plan. While
target-efficient plans have lower public costs, they raise equity issues—not
providing government subsidies to those with current coverage despite their
being in similar economic circumstances. In the end, the design of plans represents
difficult political judgments.
From our perspective, increased government spending as a result of crowding
out is important because it affects who pays for care, but it does not represent
new resources drawn into the medical care system and does not add inflationary
pressure to the existing delivery system. In fact, depending on the structure
of the expanded coverage, total spending might not be affected very much if
people switch from private to subsidized public coverage.
Two approaches.
Our simulations reflect two “generic” approaches to structuring health
insurance. These estimates provide a benchmark against which to compare specific
proposals. They should not be interpreted to mean that public insurance is necessarily
less expensive than private insurance. In particular, if Medicaid enrollment
were to double, there could be a substantial increase in political pressure
to make provider payments more generous, to induce more providers to treat people
covered by public insurance.
Similarly, although we have referred to the two generic approaches as an “average
private plan” and an “average public plan,” our analysis does
not address how expanded coverage under either approach would be financed. Combinations
of federal income tax credits, income-related premiums, and federal and state
appropriations could be used under either scenario. “Private” and
“public” in our analysis refer primarily to a structure of covered
services, cost-sharing arrangements, and provider payment approaches. Thus,
even though most of the people with public coverage in the baseline analysis
are in fact covered by Medicaid, the extra spending we simulate under expanded
coverage would not necessarily be in the form of an expanded Medicaid program
or a program financed in the same way that Medicaid (and SCHIP) are financed
now.
Effects of coverage expansions.
Even with these caveats, however, the overall impact of expanded coverage on
total health care costs, an increase of $35–$70 billion, is actually relatively
small, accounting for roughly 3–6 percent of total health care spending.
An expansion of this magnitude would increase health spending’s share of
gross domestic product (GDP) by less than one percentage point, from 14.1 percent
of GDP to 14.5–14.9 percent.
Given the growing evidence of the beneficial effects of having insurance on
health, labor-force participation, earnings, and education, the cost of expanding
insurance coverage may be a relatively small or at least a very worthwhile investment
when considered against the benefits of improved health, increased longevity,
and potentially greater national income.27
These benchmark cost estimates should reassure policymakers that the cost of
additional care that would be used by the newly insured, in spite of its large
absolute value, is much lower than the expected average annual revenue loss
of almost $170 billion from federal tax cuts enacted since 2001. Total federal
government costs of actual proposals may very well exceed the magnitudes of
recent and proposed tax cuts.28 However, total
government costs include substantial offsets or transfers of costs that correspond
to savings to employers, workers, state and local governments, and individuals
under the current system of financing health insurance. Our cost estimates suggest
that the magnitude of forgone tax revenues is comparable to the cost of the
additional medical care that would be used by a fully insured population over
the next ten years.
Although we have emphasized the cost of additional medical care used by the
uninsured, our analysis noted that a substantial amount is already being spent
on care received by uninsured people (Exhibit
4).29 Much of this money flows through an elaborate
and often hidden network of grants, indirect payments, and subsidies from a
variety of primarily public sources to medical care providers. A potentially
important implication of a comprehensive rather than incremental approach to
covering all of the uninsured is that the existing public money already being
used to pay for care received by the uninsured will be very difficult to capture
or reallocate if insurance expansion is piecemeal. Providers treating the uninsured
will be loath to relinquish their existing subsidies unless they are assured
that everyone will be insured.
This research was supported by a grant from the Henry J. Kaiser Family Foundation
under the Cost of Not Covering the Uninsured Project. Diane Rowland, Catherine
Hoffman, David Rousseau, and Barbara Lyons of the Kaiser Commission on Medicaid
and the Uninsured as well as the members of this project’s Advisory Group
provided useful comments on earlier drafts. The authors gratefully acknowledge
the contributions of Willard Manning, Sherry Glied, and Emmet Keeler, who commented
on the statistical analysis. Marc Rockmore provided excellent research and programming
assistance.
NOTES
1. For comprehensive summaries of this extensive literature,
see J. Hadley, “Sicker and Poorer,” Medical Care Research and Review
60, no. 2 (Supp., June 2003): 3S–75S; J. Hadley, Sicker and Poorer:
The Consequences of Being Uninsured,May 2002, www.kff.org/content/2002/20020510/4004.pdf
(12 May 2003); Institute of Medicine, Care without Coverage (Washington:
National Academies Press, 2002); and American College of Physicians, No Health
Insurance? It’s Enough to Make You Sick (Philadelphia: ACP, 1999).
2. J. Hadley and J. Holahan, “How Much Medical Care Do
the Uninsured Use, and Who Pays for It?” 12 February 2003, www.healthaffairs.org/WebExclusives/Hadley_Web_Excl_021203.htm
(12 May 2003).
3. S.H. Long and M.S. Marquis, “The Uninsured ‘Access
Gap’ and the Cost of Universal Coverage,” Health Affairs (Spring
II 1994): 211–220.
4. P.F. Short et al., “The Effect of Universal Coverage
on Health Expenditures for the Uninsured,” Medical Care 35, no.
2 (1997): 95–113.
5. Lewin-VHI, The Financial Impact of the Health Security
Act (Reston, Va.: Lewin-VHI, 1993).
6. J.F. Sheils, L.S. Lewin, and R.A. Haught, “Potential
Public Expenditures under Managed Competition,” Health Affairs (Supplement
1993): 229–242.
7. See J.W. Cohen, Design and Methods of the Medical Expenditure
Panel Survey Household Component, MEPS Methodology Report no. 1, Pub. no.
97-0026 (Rockville, Md.: Agency for Healthcare Research and Quality, 1997).
8. About 15 percent of uninsured adults and 8 percent of uninsured
children have family incomes exceeding 400 percent of poverty. We retained these
cases in the analysis in order to have a representative sample of all uninsured
people. We also retained the approximately 2 percent of full-year publicly insured
people with incomes greater than 400 percent of poverty.
9. Centers for Medicare and Medicaid Services, “Table 1:
National Health Expenditures Aggregate and Per Capita Amounts, Percent Distribution,
and Average Annual Percent Growth, by Source of Funds: Selected Calendar Years
1980–2001,” 8 January 2003, cms.hhs.gov/statistics/nhe/historical/t1.asp
(12 May 2003).
10. T.M. Selden et al., “Reconciling Medical Expenditure
Estimates from the MEPS and the NHA, 1996,” Health Care Financing Review
23, no. 1 (2001): 161–178. For details, see J. Hadley and J. Holahan, Who
Pays and How Much? The Cost of Caring for the Uninsured, 12 February 2003,
www.kff.org/content/2003/4088/4088.pdf
(12 May 2003).
11. A. Jones, “Health Econometrics,” in Handbook
of Health Economics, vol. 1a, ed. A.J. Culyer and J.P. Newhouse (Amsterdam:
Elsevier, 2000), 268–344; and W. Manning and J. Mullahy, “Estimating
Log Models: To Transform or Not to Transform?” Journal of Health Economics
20, no. 4 (2001): 461–494.
12. We did not adjust the estimates of the standard errors
for the effects of MEPS’s complex sampling design because we were not interested
in testing any hypotheses about the statistical significance of any of the parameter
estimates. Details regarding the statistical procedures and model coefficients
are provided in a statistical appendix. See J. Hadley and J. Holahan, “Estimates
of the Cost of Covering the Uninsured: Statistical Appendix,” 4 June 2003,
www.kff.org/content/2003/20030604.
13. Letting S* be predicted spending and S be actual spending,
the formula for simulating spending for the uninsured, assuming that they have
full-year coverage, is
S* = Prob[anyexp | Xu]*E(S |
S > 0, Xu), where the first part of the expression is the logistic
model for the probability of having any spending and the second part is expected
spending, given that actual spending was greater than zero. Xu represents
the values of the independent variables for the uninsured, with the insurance
coverage variable set to full-year coverage, either private or public.
14. More than thirty-five million Americans were uninsured
for less than a full year in 2001–2002. Families USA, Going without
Health Insurance, Pub. no. 03-103 (Washington: Families USA, 2003).
15. They were assigned to the private full-year coverage group
if they had private coverage for six months or more.
16. Hadley and Holahan, “Estimates of the Cost of Covering
the Uninsured: Statistical Appendix.”
17. Data are collected only for the noninstitutionalized portion
of the year.
18. These estimates are not adjusted for the value of uncompensated
care that is subsidized by general payments to providers—that is, payments
that are not specifically linked to an individual, identified patient. However,
they are adjusted for the differences between MEPS and the NHA.
19. Hadley and Holahan, “Estimates of the Cost of Covering
the Uninsured: Statistical Appendix.”
20. Hadley and Holahan, “How Much Medical Care Do the
Uninsured Use?”
21. Ibid.
22. S. Norton and S. Zuckerman, “Trends in Medicaid Physician
Fees, 1993–1998,” Health Affairs (July/Aug 2000): 222–232;
and Medicare Payment Advisory Commission, Report to the Congress: Medicare
Payment Policy (Washington: MedPAC, March 2002), 156.
23. Hadley and Holahan, “Estimates of the Cost of Covering
the Uninsured: Statistical Appendix.”
24. E. Miller, J. Banthin, and J. Moeller, “Covering the
Uninsured: Estimates of the Impact on Total Health Expenditures for 2002”
(Unpublished manuscript, AHRQ, 2003).
25. S. Long and M.S. Marquis, Universal Health Insurance
and Uninsured People: Effects on Use and Cost (Washington: U.S. Congress,
Office of Technology Assessment, 5 August 1994); and Short et al., “The
Effect of Universal Coverage.”
26. For an example of the variation in cost estimates associated
with specific proposals, see U.S. Congress, Office of Technology Assessment,
Understanding Estimates of National Health Expenditures under Health Reform,
Pub. no. OTA-H-594 (Washington: U.S. Government Printing Office, May 1994).
27. Hadley, “Sicker and Poorer”; and IOM, Care
without Coverage.
28. “Breaux Proposes Universal Care—American Style,”
Medicine and Health 57, no. 4 (2003): 4; and D. Balz, “Gephardt
Health Plan to Cover All,” Washington Post,24 April 2003.
29. Hadley and Holahan, “How Much Medical Care Do the
Uninsured Use?”
Jack Hadley, an economist, is a principal research associate at the Urban Institute
in Washington, D.C., and a senior fellow at the Center for Studying Health System
Change. John Holahan, also an economist, directs Urban's Health Policy Center.
©2003 Project HOPE–The People-to-People Health Foundation, Inc.
|
