Demand response of mental health services to cost sharing under managed care.

Abstract

Higher demand-side cost sharing on mental health services than on general health services has been justified in economic terms because the demand response for mental health services has been found to be higher under traditional indemnity plans, and the welfare loss associated with insurance is higher while the risk spreading benefits were similar. The empirical studies of demand response for mental health services under fee-for-service health care delivery systems provide the supporting evidence. With the ascendance of managed care, the context in which demand-side cost sharing is imposed today differs from the context in which most of the empirical literature rests due to the presence of managed care. The economics of parity under managed care needs to be under re-examination.
This study measures demand response of mental health services to cost-sharing under managed health care and compares it to demand response under traditional indemnity plans.
The 1996 Medical Expenditure Panel Survey (MEPS) data are used because this is the only year in which sufficient detail is available on coverage and forms of insurance in order to make the desired comparison. To address the selection problem, we focus on employees (and their dependents) who are privately insured and who have no choice of health plan. Couples with more than one insurance plan are also excluded from the analysis. We use logit models to analyze the effect of prices on the probability of any ambulatory mental health uses. We compare the estimated demand response to demand-side cost sharing between managed care plans and non-managed care plans by examining how demand prices affect the likelihood of seeking mental health services.
In the range observed, deductibles have no significant impact on the likelihood of utilization for either indemnity or managed care plans. The coinsurance rate has a significant negative effect on seeking mental health services under indemnity plans. The effect of the coinsurance rate on demand under managed care plans is significantly smaller than that under indemnity plans and not significantly different from zero. Managed care itself decreases rates of utilization.
Results in this study are consistent with the findings from the literature on mental health parity. The evidence suggests that mental health utilization is controlled by management under managed care and not primarily by out of pocket prices paid by consumers. Limitations include the small number of HMO enrollees and the current method can not entirely eliminate a concern about selection bias. IMPLICATION FOR HEALTH POLICY: Efficiency argument against parity of benefits for mental health care may not apply to managed care settings. At the same time, parity may accomplish less than mental health parity advocacy groups expect under managed care in terms of increasing access. IMPLICATION FOR FURTHER RESEARCH: Managed care continues to evolve, take many forms, and uses a number of rationing devices. It is important to conduct studies to isolate the effects of the components of managed care on utilization among different patient groups.

Full text

Page 1

DEMAND RESPONSE TO COST SHARING UNDER MANAGED
HEALTH CARE
CHUNLING LU, RICHARD G. FRANK and THOMAS G. MCGUIRE*
We use 1996 Medical Expenditure Panel Survey to examine the demand response
of ambulatory medical services to demand-side cost sharing under managed health
care and find that demand response to a coinsurance rate is less than that under con-
ventional plans. To address the selection problem, only respondents who were offered
a single insurance plan from private establishments are studied. Our results show that
managed care, relying on different approaches to ration, has blunted demand
response. This suggests that in a managed care context, increasing demand-side cost
sharing will reduce costs to plans mainly by shifting costs to enrollees. (JEL I11)
I.INTRODUCTION
Demand response is one of the most vener-
able topics in health economics. Zeckhauser
(1970) related demand response to optimal
(second-best) health insurance showing that
a greater demand response implied less cover-
age. That is, when cost-sharing provisions are
the only instruments for controlling use of
care, control of moral hazard requires reduced
coverage. Zeckhauser’s work set out the theo-
retical foundation for the largest economics
research project of the 1970s and 1980s, the
RAND Health Insurance Experiment (New-
house and the Insurance Experiment Group,
1993). With the ascendance of managed care
in 1990s, empirical research on demand
response declined. This was due in part to
great reliance on provider payment incentives
and administrative processes to control use of
health care. In this way, controlling costs
could be done without shifting financial risk
to consumers, seemingly rendering demand-
side cost sharing—the portion of the bill the
patient pays—obsolete.
Ma and McGuire (1997) generalized Zeck-
hauser’s model to both supply-side cost shar-
ing which imposes financial penalties on
doctorsforusingreferred
demand-side cost sharing, demonstrating the
superiority of supply-side instrumentsfor con-
trolling quantity while allowing for expansion
of insurance coverage to protect against risk.
In line with this idea, consumers’ out-of-
pocket burden for health-care cost fell in real
terms during the 1990s and fell greatly in rela-
tion to total health-care costs. According to
Health United States published by National
Center for Health Statistics (2003), in 1990,
the inflation adjusted (year 2000 dollars) per-
sonal out-of-pocket expenditures were $224.1
billion. By 2000, they had decreased to $194.1
billion. Out-of-pocket expenditures declined
as a percent of total personal health-care
expenditures from 22.5% in 1990 to 17.1%
in 2000. Since the mid-1990s, the percentage
of workers enrolled in some kind of managed
care plans (MC) has increased from 77% in
1996 to 93% in 2006 (The Kaiser Family
Foundation and Health Research and Educa-
tion Trust 1998, 2006).
Coincidentwiththisgrowthinmanagedcare
enrollmentwastheso-called‘‘backlash’’against
services and
*Theauthorsgratefullyacknowledgefinancialsupport
from NIMH training grant MH19733-09/10 and from
the Agency for Healthcare Research and Quality (PO1
HS10803). The authors thank participants of American
Society of Health Economists Conference for their
comments.
Lu: Senior Research Associate, Harvard Global Health
Initiative, Harvard University. Phone (617) 495 4866,
Fax (617) 495 8231, E-mail chunling_lu@harvard.edu.
Frank: Professor, Health Care Policy Department, Har-
vard Medical School. Phone (617) 432 0178, Fax
(617) 432 1219, E-mail frank@hcp.med.harvard.edu.
McGuire: Professor, Harvard Care Policy Department,
Harvard Medical School. Phone (617) 432 3536, Fax
(617)4322905,E-mailmcguire@hcp.med.harvard.edu.
ABBREVIATIONS
GLM: Generalized Linear Models
HC: Household Component
HMO: Health Maintenance Organization
MEPS: Medical Expenditure Panel Survey
MC: Managed Care Plan
PPO: Preferred Provider Organization
Contemporary Economic Policy (ISSN 1074-3529)
Vol. 27, No. 1, January 2009, 1–15
Online Early publication September 2, 2008
doi:10.1111/j.1465-7287.2008.00124.x
? 2008 Western Economic Association International
1

Page 2

managed care’s rationing methods. Providers
resentedbeingtoldwhatserviceswereappropri-
ate and patients disliked the bureaucracy of
seeking permission from specific types of care.
The negative reaction to restricted choice of
provider and stringent approval processes man-
ifests in the market and through the political
process. Private and public policy makers have
loosened administrative controls of health-care
utilization and are returning to reliance on
demand-side cost sharing to ration health-care
services.Nevertheless,
arrangementsarebeingimplementedinthecon-
text of a health-care delivery system that still
relies heavily on payments arrangements that
reward provider for doing less administrative
mechanisms and capacity constraints to ration
care. In considering policies such as high-
deductible health plans that aim to use demand
prices to ration services, it once again becomes
important to understand demand response.
However, because the basic institutions that
insure and pay for health care have changed
since the 1980s, new research is necessary to
quantify the magnitude of demand response
within managed health-care plans (Health
Maintenance Organizations [HMOs], Preferred
Provider Organizations [PPOs], and Point of
Service plans).
Expectations abouttheability ofnewforms
of cost-sharing arrangements to save money
and steer care toward more efficient types of
care depends critically on demand response.
The Congressional Budget Office and some
actuaries continue to rely on the results of
the RAND Health Insurance Experiment to
project response to high-deductible health
plans or tovarious cost-sharingarrangements.
This article seeks to provide estimates of
demand response to cost-sharing arrange-
ments within the context of recent institutions
that pay for and manage health care for the
vast majority (93%) of privately insured
Americans (Employer Health Benefits 2006).
In this article, we use the 1996 Medical
Expenditure Panel Survey (MEPS) data to
study the effect of demand-side cost sharing
in managed care. MEPS data include years
from 1996 to 2003, but only 1996 MEPS pro-
videsinformationoninsurancecoveragepolicy.
The main challenge when using observational
data to study demand response is selection:
individuals choose different insurance plans
according totheir personal characteristics, such
as health status and anticipated health-care
thesecost-sharing
demand. In the presence of risk selection, inter-
preting the differences of medical expenditures
by different plan designs as demand response is
difficult.Mostpreviousstudiesofmanagedcare
effects try to remove the effects of self-selection
by multivariate controls (Glied, 2000).
We focus on employees (and their depend-
ents) who are privately insured and who have
no choice of health plan, a strategy for mini-
mizing the adverse selection problem at least
with respect to choice of insurance plans. Cou-
ples with more than one insurance plan are
excluded from the analysis. We conduct vari-
ous tests to investigate whether or not this
methodiseffective,andtheresultsprovideevi-
dence that our strategy substantially decreases
concern about selection bias.
Our main results are that the level of
a deductible has no significant impact on uti-
lization of ambulatory services in the ranges
observed; demand response to the coinsurance
rate under MCs is less than that in traditional
indemnity plans, and in particular, raising
coinsurance rate has smaller effects on the
level of expenditures in MCs. The evidence
suggests that increasing demand-side cost
sharing in managed care may reduce plan
costs primarily by shifting costs from plans
or employers to consumers.
This article has five sections. Section II
explainswhyweexpectdemand-sidecostshar-
ing to affect use in MCs less than that in tra-
ditional indemnity plans and presents the
estimation method. Section III describes the
data, study population, sample, and variables
used in the analysis. Section IV contains the
results. Section V discusses our findings and
their implications for design of health insur-
ance coverage under managed care.
II. METHOD
The role of demand-side cost sharing in
controlling expenditures is different between
traditional indemnity plans and MCs. Under
an indemnity plan, demand-side cost sharing
is the only plan policy affecting utilization,
whereas in MCs, both demand-side cost shar-
ing and managed care mechanisms play roles
in determining individuals’ expenditures. In
the presence of the additional rationing mech-
anisms used in managed care, we expect a con-
sumer to be less responsive to changes in
demand-side cost sharing than under indem-
nity plans.
2CONTEMPORARY ECONOMIC POLICY

Page 3

Figure 1
between demand-side cost sharing and medi-
cal expenditures in traditional insurance and
in the presence of managed care. A demand
curve is shown along with two levels of
demand-side cost sharing represented by coin-
surance rates c1 and c2. When quantity is
determined solely by the level of cost sharing,
as it would be in traditional health insurance,
a decrease in the cost sharing from c1to c2
increases use from q1to q2. (Cost is not shown
in the diagram, but in general, we would
expect marginal cost to exceed coinsurance
rates.) Analysis of a deductible would be set
up differently, but the conclusion we reach
below would be the same.
MCs rely on other ways to control utiliza-
tion in addition to demand-side cost sharing,
such as selecting, organizing or paying pro-
viders, use of capitation, and monitoring ser-
vice utilization (Glied, 2000). Using MEPS
data, we find that in 1996, about 79% of
HMO plans had exclusive providers and 92%
of HMO plans had some gatekeeper mecha-
nisms. The level of expenditure in MCs is thus
affected not only by the out-of-pocket prices
as determined by coinsurance and other ele-
ments of benefit design but also by ‘‘manage-
ment’’ of care.
Two basic approaches have been proposed
in the literature for capturing these manage-
ment mechanisms in an economic model of
managed care. The first approach views man-
aged care as setting quantities for individuals
who may be heterogeneous with respect to
illustrates therelationship
severity of illness and demand. For example,
Pauly and Ramsey (1999) assumed that in
order to control costs, the HMO was able to
imposeuniform-across-severity
restrictions on the amount of treatment that
was offered. The other approach views man-
aged care as rationing services by a ‘‘shadow
price.’’ For example, Frank, Glazer, and
McGuire (2000) considered the MC as giving
consumers all services that are valued above
the shadow price and denying care for all uses
for which the value was below the shadow
price. Both approaches imply that, under man-
aged care, observed demand response to out-
of-pocket cost will be less elastic than that
under indemnity arrangements. We use the
Pauly and Ramsey approach to illustrate
why managed care should decrease demand
response.
Figure 1 illustrates the idea of a quantity
limit, qr, of the kind envisioned by Pauly
and Ramsey. At level of cost sharing c1, the
consumer uses q1as before. Now, however,
when the coinsurance falls to c2, the consum-
er’s demand increases to q2, but constraints in
the MC limit use to qr. Thus, measured
demand response under a managed care plan
is thus the difference between qrand q1rather
than the larger difference between q2and q1.
Since quantity limits in managed care will
sometimes bind, we expect the demand
responsetodemand-sidecostsharingtobeless
under a MC than in a conventional plan.
Following Manning and Mullahy (2001),
our point of departure for estimation of
quantity
FIGURE 1
The Effect of Changes in Demand-Side Cost Sharing on Use in Traditional and MCs
q1
qr
q2
Quantity
c2
c1
Demand
Coinsurance
level of rationing
in managed care
LU, FRANK & MCGUIRE: DEMAND RESPONSE UNDER MANAGED CARE3

Page 4

demand response is the two-part generalized
linear model (GLM). The first part of the
model estimates a logit equation for the prob-
ability of any ambulatory medical care use.
Probðy1.0Þ 5 expðX
0bÞ=ð1 þ expðX
0bÞÞ;
where y1takes 0 or 1 indicating whether or not
an individual uses medical services and X rep-
resents a group of independent variables.
The second part consists of a GLM eq-
uation on the level of expenditure given the
positive use of medical services. The GLM
directly models both the mean and the
variance functions on the original scale of
expenditure.
varðy2=XÞ 5 jðlðX
0bÞÞk;
where y2is the level of spending, j . 0, l(X#b)
represents mean function E(y2/X), l is the
inverse link between the expectation of the
observed raw-scale y2and the linear predictor
X#b, and k is nonnegative and finite.
To choose the link function between
the observed expenditures and the right-hand
side variables, we examine the distribution
of expenditures. The histograms of the log-
transformed nonzero data and residuals ap-
pear symmetric. The kurtosis of the log-scale
residuals from the regression is 3.02, whereas
the normal distribution is 3.0. This suggests
that with an appropriate variance function,
the GLM is efficient. We choose to use the
log as a link function, a common choice in
health expenditure studies according to Bun-
tin and Zaslavsky (2004). Conducting a Park
test to estimate the relationship between the
mean and the variance as suggested by Man-
ning and Mullahy (2001), we find that the best
model falls between variance proportional to
mean model (Poisson like) and the standard
deviationproportional
(Gamma like). We find little difference
betweenresultsfromthesetwomodelsbyplot-
ting predictions of the alternative models
against true means by deciles. We report the
results from Gamma-like model below.
Wecompare the
response to price between MCs and non-
MCs by examining how demand prices affect
expenditures on medical services under two
types of insurance plans. We do not find a sig-
nificant difference of the average price paid
to providers per visit between these two types
tomeanmodel
estimated demand
of plans (Appendix Table A8). The price ef-
fects on expenditure can be interpreted as
changes in quantity. We identify the differen-
tial response between two types of plans by
specifying an interaction term between man-
aged care dummy variable and prices varia-
bles. If the estimated interaction terms are
positive and significant, this implies that the
change of quantity demanded under MCs is
smaller than that under non-MCs.
It has long been recognized that selection
interferes with the inferences about demand-
side cost-sharing effects on the utilization of
health-care services based on observational
data. Individuals with different characteristics
may enroll in different plans if they have
options (Cutler and Zeckhauser, 1998; Nichol-
son et al. 2004). Clearly, the availability of mul-
tiple health plans creates the opportunity for
biased selection (Call et al. 1999; Nicholson
et al. 2004). By focusing on an insured popu-
lation where a single health plan is offered, we
expect that the adverse selection problems
stemming from having multiple choices of
insurance plans can be minimized among indi-
viduals with a single health plan. We also
exclude those two-worker couples with more
than one insurance plan offered from the sam-
ple to further reduce the problem of selection.
A fewpreviousstudiesusethismethodtomea-
sure the degree of risk selection between HMO
and non-HMO plans (Polsky and Nicholson,
2004; Reschovsky, 1999/2000). They find no
evidence that HMO plans attract a dispropor-
tionate share of low-risk enrollees when a sub-
sampleonlyincludesemployeeswithnochoice
of health plan type.
To investigate the effectiveness of our strat-
egy in minimizing selection bias, we examine
whether the job with no-option plan is more
likely to attract individuals with certain health
status and personal characteristics associated
with health status. To conduct this test, we
compare sociodemographic and health sta-
tus variables between privately insured em-
ployees with no option of insurance plans to
privately insured employees. We also investi-
gate whether our study population is different
from the privately insured population with
multiple options by comparing the average
values of sociodemographic and health sta-
tus variables between these two populations.
Finally, we examine the mean differences in
sociodemographics andhealthstatusvariables
between privately insured employees enrolled
4CONTEMPORARY ECONOMIC POLICY

Page 5

in MCs with no-option and privately insured
employees enrolled in conventional plans with
no-options. If selection through employment
exists, we would expect to observe some sys-
tematic differences on health status variables
betweenthese different groups.Wefindnosig-
nificant differences in health status and some
sociodemographic variables between these
three different groups (Appendix Tables 1–3).
Results from multivariate regression models
show that (1) there is no significant relation-
ship between the health status variables and
a dummy variable indicating a single plan
option and (2) among those who are offered
single plans,there isno significantrelationship
between health status and plan types (Appen-
dix Tables 4 and 5).
The population with a single plan option
tends tooverrepresentemployeesofsmallfirms
(Long and Marquis 1999). In our sample,
about 60% of individuals with no-option insur-
ance plans are in small firms (less than 50
employees).Wetestthedifferenceofhealthsta-
tus and sociodemographic variables between
employees in small firms (less than 50 employ-
ees) and the employees in larger firms (more
than 500 employees). We do not find a signifi-
cantdifferenceonhealthstatusandsomesocio-
demographic variables (such as age, gender,
and education) between these two types of
firms (Appendix Table A6). The multivariate
regression results also show that there is no sig-
nificant relationship between the health status
variables and a dummy variable indicating
a small firm (Appendix Table A7). No signifi-
cant differences in health status variables imply
that our strategy effectively reduces a concern
about selection bias among individuals with
asinglehealthplan.The limitationofthisstrat-
egy is that employees may choose to work in
firms that offer different types of insurance
plans based on their unobserved preference
for health care. Without controlling for these
unobserved factors, the current method may
not entirely eliminate selection bias.
III.EMPIRICAL IMPLEMENTATION
A. Data
The data used in this study come from 1996
MEPS, which was conducted by the Agency
for Healthcare Research and Quality (http://
www.meps.ahrq.gov). MEPS is designed to
provide nationally representative estimates of
health-care use, expenditures, sources of pay-
ment, and insurance coverage for the civilian
noninstitutionalized population of the United
States. The household component (HC) col-
lects data on a sample of families and individ-
uals drawn from a nationally representative
subsample of households that participated in
theprioryear’sNationalHealthInterviewSur-
vey. There are about 21,571 persons in the HC
files of 1996. Among them, 18.2% have public
insurance, 66.7% of them have private insur-
ance, and 15.1% of them are not insured.
We use the following parts of the MEPS
survey: self-reported information about so-
ciodemographic variables, health status va-
riables, and chronic diseases; self-reported
information or provider billing records on
expenditures on outpatient visits, office-based
visits, and emergency care; and information
collected from the booklets on insurance type
or policies (e.g., deductibles and the coinsur-
ance rate) obtained from surveyed household.
Among the privately insured individuals, only
about 54% of individuals provided booklet
information on their insurance policy. MEPS
documentation warns users to be cautious in
generalizing their results due to significant
nonresponse on this item. The weights in sur-
veys such as MEPS are designed to be applied
to the entire sample to make it nationally rep-
resentative. In this article, we focus on a delib-
erately selected sample. Our aim was not to
draw inferences for people who have a choice
of health plan. The estimates apply to people
who do not have plan choices. Therefore, the
weights would not serve to augment the infer-
ences that could be drawn. We include cova-
riates as controls that also account for the
variables that MEPS uses for stratification
(demographics).
Among individuals who offer booklet
information, we only include those persons
(and their dependents) who reported that their
employers in 1996 offered them a single health
insurance plan. Two-worker couples with
more than one insurance plan offered are also
excluded. There are 1,774 individuals in the
final sample.
B. Model Specification
Dependent Variables. We focus on two depen-
dent variables: a dummy variable indicating
some use of ambulatory medical services
andthetotalexpenditureon general
LU, FRANK & MCGUIRE: DEMAND RESPONSE UNDER MANAGED CARE5

Page 6

ambulatory visits conditional on any use. The
expenditure data in MEPS were derived from
both the MEPS HC and medical provider
components (MPCs). The MPC contacted
medical providers identified by household
respondents. The charge and payment data
from medical providers were used if complete;
otherwise HC data were used if complete.
Ambulatory visits include outpatient visits,
office-based visits, and emergency care visits.
Expenditure in our analysis, therefore, is
defined as the sum of payments made by indi-
viduals and private insurance for care received
from emergency room visits, office visits, and
outpatient visits. The data show that about
81% of individuals used ambulatory visits in
1996 (Table 1).
Our key interest is the responsiveness of
quantity demanded to demand-side cost shar-
ing under MCs compared to that under
indemnity plans. We treat ‘‘expenditure’’ on
ambulatory visits as a quantity index (in dol-
lars). Some research has found that managed
care may also affect price paid, not only quan-
tity (Cutler, McClellan, and Newhouse, 2000).
Therefore, we determine whether we can use
expenditure as an indicator of quantity
demanded by checking the difference of aver-
age unit price between managed care and
non-MCs. The average unit prices are not
TABLE 1
Definitions and Summary Statistics
Variable Name
N
Mean Standard Deviation
Dependent variables
Use: using general health-care services
Expenditure: expenditure on ambulatory visits ($)
Sociodemographic variables
Age
Female
Nonwhites
Living in suburban area
Living in the Northeast
Living in the West
Living in the Midwest
Living in the South
Family size
High education
Middle education
Low education
High income
Middle income
Low income
Health status variables
Perceived good health
Activity limitation
Having chronic disease before 1996
Having chronic disease in 1996
Insurance and plan variables
Deductible ($)
Coinsurance rate (%)
MC: plans with exclusive or preferred providers
HMO: plans with exclusive provider
PPO: plans with preferred provider
Fee-for-Services plans: plans with any provider
1,774
1,444
0.810.39
8602,145
1,774
1,774
1,774
1,774
1,774
1,774
1,774
1,774
1,774
1,774
1,774
1,774
1,774
1,774
1,774
30.11
0.51
0.89
0.71
0.18
0.18
0.28
0.35
3.42
0.06
0.26
0.68
0.36
0.44
0.20
17.23
0.50
0.31
0.45
0.39
0.39
0.45
0.48
1.39
0.23
0.44
0.47
0.48
0.50
0.41
1,774
1,774
1,774
1,774
0.93
0.13
0.19
0.10
0.26
0.34
0.39
0.30
1,774
1,774
1,774
1,774
1,774
1,774
134
14.64
0.67
0.20
0.47
0.33
232
9.88
0.47
0.40
0.50
0.47
Notes: The sample includes individuals who were offered a single private health insurance plan by employers in 1996
MEPS data. Spouses and children are also included if they are the dependents of insurance policy holders. Ambulatory
visits include outpatient visits, office-based visits, and emergency care visits.
6CONTEMPORARY ECONOMIC POLICY

Page 7

statistically different between managed care
and non-MCs. We also run a multivariate
regression model to examine the effects of
MC on the unit price of visit. We find that
the managed care dummy variable has no sig-
nificant effects on the unit price of visit
(Appendix Table A8). We conclude that dif-
ferences in spending between managed care
and non-MCs can be treated as differences
in quantity. Furthermore, any differences in
average price levels between managed care
and non-MCs will be accounted for by plan
main effects.
The total expenditure on ambulatory visits
for each individual is the sum of facility and
physician expenses of all visits during the year.
The mean expenditure was $860 (Table 1).
The distribution of expenditures is skewed
to the right: less than 1% of users have costs
more than $10,000 and about 50% of users
have costs less than $260.
Independent Variables. The independent vari-
ables fall into three categories: plan types, pri-
ces, interactions between plans and prices;
sociodemographic characteristics; and health
status. The summary statistics for these varia-
bles are provided in Table 1.
In 1996 Health Insurance Plan Abstraction
Linked Data, information on provider ar-
rangements (exclusive provider, unrestricted
choice of provider, and mix of preferred and
unrestricted choice of provider) is available.
Following Miller and Luft (1994), we take
the selection of network physicians as the sin-
gle most important feature that distinguishes
a MC from a conventional plan. The term
MC refers to any plan that uses a network
of providers, such as HMO and PPO plans.
In the sample, about 20% of individuals
enrolledinHMOplansand47%ofindividuals
enrolled in PPO plans. We create a dummy
variable MC to represent MCs and conven-
tional plans based on the restriction of pro-
viders. If a plan has an exclusive provider or
preferred provider, we define variable MC 5
1; if provider choice is unrestricted, we define
MC 5 0. About 67% of individuals report
enrolling in MC plans and 33% of individuals
reportenrollinginnonmanagedplans(Table 1).
Among enrollees in MC plans, 76% use some
general medical services. Among enrollees in
non-MCs, about 73% use some general med-
ical services. Users in MC plans have lower
average spending ($799) than users in non-
MCs ($963) (Table 2). Though HMO and
PPO plans may function differently in many
aspects, our focus is on comparing demand
response between traditional indemnity plans
and any plans with management mechanism
other than prices. To test the sensitivity of
the results to the definition of managed care,
we also run regressions by separating out
HMOs and PPOs.
Two variables measure the price paid by
the consumers: the deductible and the coinsur-
ance rate. The deductible, the amount the con-
sumer must pay before coverage implies, is
measured in dollars. In MEPS 1996, the coin-
surance rate is defined as the percent of expen-
ditures that are to be paid by consumers after
any deductible has been met and before any
maximums take effect. About 44% of individ-
uals have plans which pay a specified percent-
age of the charges, about 55% of individuals
have plans which pay the remainder after
a copayment by the covered person, and the
remaining 1% of individuals has plans which
pay a specified percentage after a copayment
by the covered person (Medical Expenditure
Panel Survey, 1996). In MEPS 1996, copay-
ments are converted to coinsurance by assum-
ing that the covered person faces the average
cost of a physician office visit for MEPS
households in 1996, which was $91. The coin-
surance rate variable has values expressed in
TABLE 2
Summary Statistics of Utilization and Price Variables of Different Types of Plans
Use (%) Expenditure ($)Deductible ($) Coinsurance Rate (%)
MC (51) [N 5 1,340]
MC (50) [N 5 683]
0.76 (0.43)
0.73 (0.45)
799.10 (2068.90)
962.75 (2410.18)
79.90 (166.46)
241.04 (300.82)
12.29 (8.03)
20.08 (12.15)
Notes: The sample includes individuals who were offered a single private health insurance plan by employers in 1996
MEPS data. Spouse and children are included if they are the dependents of insurance policy holders. Values in parenthesis
are standard deviation.
LU, FRANK & MCGUIRE: DEMAND RESPONSE UNDER MANAGED CARE7

Page 8

terms of percentages that range from 0 to 100.
In our sample, the mean values of the de-
ductible and coinsurance rate are $134 and
14.6%, respectively (Table 1). The correlation
between these two price variables is only .20,
suggesting that we should be able to estimate
their effects in the same regression. Table 2
shows that the mean values of price variables
vary greatly between MCs and non-MCs. The
t test suggests that conventional plans have
a significant higher mean value of deductible
($241 vs. $80) and coinsurance rate (20.1%
vs. 12.3%) than MCs.
To test the possible endogeneity of the
deductible and coinsurance rate, we follow
the suggestion of Davidson and Mackinnon
(1993) and use an augmented regression test.
We take deductibles as dependent variable
and run an ordinary least square regression
on all other exogenous variables and an instru-
ment variable indicating whether or not the
firm has employees less than 50. The instru-
ment variable is significant at the .05 level.
Weobtaintheresidualsfromtheordinaryleast
square regression and include them in the orig-
inal model. A significant nonzero coefficient
for the residual term is an indication that the
suspected variable is endogenous. We find that
the coefficient of residuals is not significant
with a p value of .93 in logistic regression
and .53 in the regression of GLM. This test
supports our assumption that deductible vari-
able is exogenous. The same test is applied to
coinsurance rate. The instrument variable has
a p value of .08. The coefficient of residuals is
not significant in the logistic model with p val-
uesof.33and.48intheGLM.Hence,wereject
the endogeneity of coinsurance rate.
Sociodemographic
female, nonwhite, living in suburban area,
and four regional indicators. A set of dummy
variables for education level are high educa-
tion (more than 16 yr), middle education
(between 12 and 16 yr), and low education
(less than or equal to 12 yr). Income is indi-
cated by dummy variables for high (more than
400% of the federal poverty line), middle
(200%–400% of the federal poverty line),
and low income (below 200% of the federal
poverty line). The variables ‘‘living in the Mid-
west,’’ ‘‘low income,’’ and ‘‘low education’’
are the reference groups in the models.
Health status variables include an indicator
for an ‘‘activity limitation,’’ a dummy variable
for whether a person had activity limitation,
variables areage,
and a dummy variable for individual’s per-
ceived physical health; it is 1 if perceived phys-
ical health taking on values ‘‘excellent,’’ ‘‘very
good,’’or‘‘good’’and0otherwise.Inaddition,
we also construct indicators for whether a per-
son had a chronic disease before and in 1996.
IV. RESULTS
Estimates from the logistic regression
model of the probability of using any medical
care and the GLM regression model of the
level of spending conditional on use of care
are reported in Table 3.
As Table 3 shows, the coinsurance rate var-
iablehasanegativeeffectonthelevelofspend-
ing (?.015) and the estimate is statistically
significant at the .001 level, suggesting enroll-
ees with high coinsurance rate spend less than
enrollees with low coinsurance rate under con-
ventional plans. The effects of the interaction
between the managed care indicator ‘‘MC’’
and coinsurance rate on the level of spending
is positive (.029) and significant at the .001
level, indicating that, with a high coinsurance
rate, the expenditures fall less under MC plans
than under conventional plans. The effect of
coinsurance rate on the probability of using
some medical services is negative, but the esti-
mate is not significant. The effects of deduct-
ible on probability and the level of spending
are close to zero and not significantly different
from zero. The plan indicator ‘‘MC’’ has neg-
ative effects on the level of average spending
(?.64) and significant at the .001 level.
We calculate the price elasticity under con-
ventional plans using coefficient of coinsur-
ance rate in the second part of the model
since the effect of coinsurance rate in the first
part is not significant. The marginal effect of
coinsurance rate on the level of spending is
?10.6, suggesting that as coinsurance rate
increases by one unit, the level of average
spending drops by $10.60. The implied price
elasticity from the second part of the two-part
model is ?0.196.
The estimates of health status and sociode-
mographic variables are presented at Appen-
dix Table A9. Health status variables are the
most powerful factors explaining the likeli-
hood of using medical care and total spend-
ing. For example, individuals with chronic
diseases or with activity limitations are more
likely to seek medical care and to spend more.
Among sociodemographic variables, female
8 CONTEMPORARY ECONOMIC POLICY

Page 9

individuals are more likely to seek health care.
Younger individuals and individuals living in
West area (comparing to individuals living in
Midwest area) are more likely to use medical
services. Comparing to individuals with low
income, individuals with high income are
more likely to seek health care and spend
more. Family size has significant negative
effect on total spending. Older people and
individuals living in Northeast area (relative
to living in Midwest area) are more likely to
spend on medical services.
Finally, we test the robustness of the above
results by separating out HMO and PPO plans.
Estimates of price and plan variables are
reported in Table 4. We find thatthe sign, mag-
nitude,andsignificanceofeffectsofcoinsurance
rate and plan variables are consistent with
results from the combination of these two
groups:theeffectofcoinsurancerateundercon-
ventional plan on the level of spending is nega-
tive (?0.015) and significant at the .001 level.
The interaction variables between ‘‘HMO’’/
‘‘PPO’’ and coinsurance rate are positive
TABLE 3
Estimated Coefficients of Plan Type and Price Variables in a Two-Part Model with MC Dummy
Logistic Regression (N = 1,774)GLM Regression (N = 1,444)
Coefficient EstimatesCoefficient Estimates
Plan type effect
MC
Price effect
Deductible
Coinsurance
Interaction effect
MC ? deductible
MC ? coinsurance
Marginal effect of coinsurance rate
under conventional plans
0.206 (0.270)
?0.644*** (0.172)
?0.0002 (0.0004)
?0.004 (0.008)
?0.0003 (0.0002)
?0.015*** (0.004)
?0.0001 (0.001)
0.0006 (0.013)
?0.0003 (0.0003)
0.029*** (0.008)
?10.60*** (3.02)
Notes: The sample includes individuals who were offered a single private health insurance plan by employers in 1996
MEPS data. Spouse and children are included if they are the dependents of insurance policy holders. The estimates of
sociodemographic and health status variables are presented separately in Appendix Table A9. Values in parentheses are
standard errors.
***Significant at the .001 level.
TABLE 4
Estimated Coefficients of Plan Type and Price Variables in a Two-Part Model with HMO
and PPO Dummies
Logistic Regression (N = 1,774) GLM Regression (N = 1,444)
Plan type effects
HMO
PPO
Price effects
Deductible
Coinsurance
Interaction effects
HMO*deductible
PPO*deductible
HMO*coinsurance
PPO*coinsurance
0.070 (0.345)
0.330 (0.285)
?0.695 (0.223)***
?0.605 (0.189)***
?0.0002 (0.0004)
?0.004 (0.0007)
?0.0003 (0.0002)
?0.015 (0.004)***
0.0009 (0.0009)
?0.0006 (0.0007)
?0.002 (0.025)
?0.002 (0.014)
0.0001(0.0005)
?0.0005 (0.0004)
0.029 (0.015)*
0.028 (0.008)***
Notes: The sample includes individuals who were a offered single private health insurance plan by employers in 1996
MEPSdata.Spouseand childrenare includedif they are the dependentsof insurancepolicyholders.Values in parentheses
are standard errors.
*Significant at the .1 level; ***significant at the .001 level.
LU, FRANK & MCGUIRE: DEMAND RESPONSE UNDER MANAGED CARE9

Page 10

(0.029 and 0.028) and significant at the .1 level
or .001 level, respectively. Both ‘‘HMO’’ and
‘‘PPO’’ have negative and significant effects
on the level of spending (?.695 and ?.605).
V. DISCUSSION AND CONCLUSIONS
Using 1996 MEPS, we find that the demand
forambulatorymedicalserviceunderaconven-
tional plan is responsive to price changes and
the implied price elasticity from the analysis
is?0.196.Thisisconsistentwithpreviousstud-
ies. In the Health Insurance Experiment study,
the pure price response for overall medical care
is on the order of ?0.2 (Keeler and Rolph,
1988). The price elasticity for outpatient med-
icalcareis?0.17atlowcoinsurancerate(0–25)
and ?0.31 at higher values (25–95) (Newhouse
and The Insurance Experiment Group, 1993).
The response is smaller under MCs. A rise
in coinsurance rate has more effect on average
predicted expenditure of conventional plans
than MC. We find no effect of a deductible
on spending for any type of plan. MCs have
a negative and significant effect on total
spending. This finding is in agreement with
the idea that rationing mechanisms of MCs,
as an alternative to controlling moral hazard,
can reduce the costs as the demand-side cost
sharing does in conventional plans.
Since most privately insured Americans are
enrolled in some type of managed care ar-
rangement, our estimates of demand response
to out-of-pocket prices of health care under-
score the importance of taking account of
the modern institutional context within which
health care is delivered. Our estimates suggest
that the welfare loss from moral hazard and
the cost savings from increase cost sharing will
be less than it was in the 1970s and 1980s.
Thus, policy makers interested in implement-
ingmeasuresthatreducecoverageandincrease
demand-side prices will realize fewer efficiency
gains than might be anticipated from demand
response estimates from an earlier era. In par-
ticular, policies that encourage the supply
‘‘high-deductible’’ health plans will reduce
health insurance costs mainly by shifting costs
to consumers. It is notable that a move to
higher consumer cost sharing in managed care
goes against the tenets of optimal insurance. A
low demand response should be tied to lower,
not higher, cost sharing.
It is important to note that the above find-
ings need to be interpreted with caution due to
the following limitations. First, the selection
problem cannot be eliminated entirely with
our strategy and we may not fully rule out
the influence of unobserved factors on both
utilization and insurance coverage. Second,
the low response rate of insurance coverage
may raise questions of generalization of the
study to individuals without choice of insur-
ance plans. To further understand the influ-
ence of demand-side
consumers’ behavior under managed care
requires a better understanding of how plans
ration medical care. We have identified, but
not triedtodistinguish
approaches: a shadow price and rationing
quantity. Both approaches imply that the
price elasticity is smaller under MCs com-
pared with that under conventional plans.
However, they have different implications
about the normative interpretation of demand
response. The shadow price approach allows
managed care mechanism to affect individuals
with different health status, while the ration-
ing quantity approach may only affect the
heavy users. Other rationing mechanisms
may also exist. Sorting out how managed care
affects utilization is an important step to
determining the features of optimal demand-
side cost sharing in these plans.
costsharingon
between,two
APPENDIX
In the Appendix, we provide econometric tests on
selection issues (Appendix Tables A1–A7). Appendix
TablesA1–A3presentthettestonthesignificanceofmean
differences of sociodemographics and health status varia-
bles between privately insured employees and privately
insured employees with single-choice insurance plans,
between privatelyinsured employees withmultiplechoices
and privately insured employees with single choice, and
between privately insured employees with single choice
in conventional plans and privately insured employees
with single choice in MCs. The results show that there
is no significant difference in mean values of health status
variables between those groups. Some sociodemographic
variables such as locations, education, and income levels
are significantly different in mean values between those
groups. Appendix Table A4 presents estimates from
a logistic regression of having a job with single-choice
insuranceplansonsociodemographicvariablesandhealth
status variables. Appendix Table A5 presents estimates
from a logistic regression of having a job with single-
choice MC on sociodemographic variables and health sta-
tus variables. We want to examine whether an individual
with certain health status is more like to obtain a job with
single-choice insurance plan or single-choice MC. We find
that there is no significant relationship between health
status variables and having jobs with single-choice insur-
ance plans. However, some sociodemographic variables
10CONTEMPORARY ECONOMIC POLICY

Page 11

are significantly associated with having jobs with certain
types of insurance plans. For example, individuals with
higher income or higher education are less likely to have
jobs with single-choice insurance plans (Appendix Table
A4). For these with single-choice insurance plans, individ-
uals with higher income are less likely to have MCs
(Appendix Table A5). The testing results suggest that
health status variables may not lead to selection concerns.
We have controlled for sociodemographic variables such
as income, education, and living areas in analysis. How-
ever, we cannot fully eliminate the selection problem since
some unobserved factors may be correlated with health or
sociodemographic variables and utilizations.
According to Long and Marquis (1999), the popula-
tion with a single plan option tends to overrepresent
employees of small firms. We want to examine whether
individuals with certain health status are more likely to
have jobs in small firms. Appendix Table A6 shows the
t test results on the significance of mean differences of
health status and sociodemographic variables between
employees in small firms and the employees in larger firms
in the sample. We do not find significant mean differences
on health status variables. The multivariate regression
results in Appendix Table A7 show that there is no signif-
icant relationship between the health status variables and
a dummy variable indicating a small firm.
In Appendix Table A8, we investigate whether the
average unit price of ambulatory under MCs is signifi-
cantly different from that under conventional plans.
Appendix Table A8 shows that the mean difference of
average unit price is not statistically different between
the managed care and the non-MCs. We also run a regres-
sion model which defines the unit price of each visit as
dependent variable. Variables of sociodemographics,
health status, insurance coverage variables, and a dummy
variable indicating play types are independent variables.
We find that the managed care dummy variable has no
significant effects on the unit price of visit. We conclude
that the differences in expenditure between managed care
and non-MCs are not due to the difference of unit price
between these two types of plans.
Appendix Table A9 presents the estimates of sociode-
mographics and health status variables for the two-part
model presented in the Table 3. Variables of health status,
age, gender, income, and living areas are significant in
either the first part or the second part of the model.
TABLE A1
Mean Difference of Sociodemographics and Health
Status Variables between Privately Insured Employees
(N 5 6 778) and Privately Insured Employees with
Single-Choice Insurance Plans (N 5 2,439)
Mean
Difference
Standard
Error
Sociodemographics variables
Age
Female
Nonwhites
Living in suburban area
Living in the Northeast
Living in the West
Living in the Midwest
Living in the South
Family size
High education
Middle education
Low education
High income
Middle income
Low income
Health status
Perceived good health
Activity limitation
Chronic disease before 1996
Chronic disease in 1996
0.200
0.028*
0.008
0.064**
0.019*
0.028*
?0.024*
?0.023*
?0.024
0.030**
0.054**
?0.084**
0.062**
?0.033**
?0.028**
0.260
0.012
0.008
0.008
0.010
0.010
0.010
0.011
0.035
0.007
0.012
0.012
0.012
0.012
0.008
0.007
?0.001
0.0003
0.002
0.006
0.008
0.011
0.008
Notes: The general population includes individuals
younger than 65 yr who were privately insured and were
employed in 1996 MEPS data. The individuals with single
choice are those who were younger than 65 yr and offered
single private health insurance plan by employers in 1996
MEPS data. Spouses and children are not included in
comparison.
*Significantat the .1level; **significant atthe .05 level.
TABLE A2
Mean Difference of Sociodemographics and Health
Status Variables between Privately Insured Employees
with Multiple Choices (N 5 4,339) and Privately
Insured Employees with Single Choice (N 5 2,439)
Mean
Difference
Standard
Error
Sociodemographics
Age
Female
Nonwhites
Living in suburban area
Living in the Northeast
Living in the West
Living in the Midwest
Living in the South
Family size
High education
Middle education
Low education
High income
Middle income
Low income
Health status
Perceived good health
Activity limitation
Chronic disease before 1996
Chronic disease in 1996
0.313
0.044**
0.013
0.099**
0.030**
?0.044**
?0.037**
?0.036**
?0.038
0.047**
0.083**
?0.131**
0.097**
?0.052*
?0.044*
0.280
0.013
0.009
0.010
0.010
0.010
0.011
0.012
0.031
0.006
0.013
0.012
0.013
0.012
0.009
0.0117
?0.002
0.001
0.003
0.006
0.009
0.012
0.009
Notes: The employees with multiple choices are those
who were younger than 65 yr and offered multiple pri-
vate insurance plans in 1996 MEPS data. The employees
with single choice are those who were younger than 65 yr
and offered single private health insurance plan in 1996
MEPS data. Spouses and children are not included in
comparison.
*Significant at the .1 level; **significant at the .05
level.
LU, FRANK & MCGUIRE: DEMAND RESPONSE UNDER MANAGED CARE11

Page 12

TABLE A3
Mean Difference of Sociodemographics and Health Status Variables between Privately Insured Employees with Single
Choice in Conventional Plans (N 5 598) and Privately Insured Employees with Single Choice in MCs (N 5 1,179)
Mean Difference Standard Error
Sociodemographics
Age
Female
Nonwhites
Living in suburban area
Living in the Northeast
Living in the West
Living in the Midwest
Living in the South
Family size
High education
Middle education
Low education
High income
Middle income
Low income
Health status
Perceived good health
Activity limitation
Chronic disease before 1996
Chronic disease in 1996
0.714
?0.024
0.047**
?0.190**
?0.012
?0.109**
0.075**
0.045*
0.008
?0.003
?0.039
0.040
?0.108**
0.073**
0.034
0.808
0.020
0.014
0.021
0.018
0.018
0.021
0.023
0.066
0.011
0.021
0.022
0.020
0.023
0.019
?0.005
0.017
0.004
0.008
0.012
0.016
0.019
0.015
Notes: The employees with asingle-choiceconventionalhealth insurance plan or MCare thosewhowereyoungerthan
65 yr and offered single private insurance plan by their employers in 1996 MEPS data. Spouses and children are not
included in comparison.
*Significant at the .1 level; **significant at the .05 level.
TABLE A4
The Multivariate Regression Results of the Relationship between the Health Status Variables and a Dummy Variable
Indicating Single-Choice Plan while Controlling for Sociodemographic Variables (N 5 6,778)
Coefficient Estimates
p . jzj Value
Sociodemographics
Age
Female
Nonwhites
Living in suburban area
Living in the South
Living in the Northeast
Living in the Midwest
Family size
High education
Middle education
High income
Middle income
Health status
Perceived good health
Activity limitation
Chronic disease before 1996
Chronic disease in 1996
?0.001 (0.003)
?0.098 (0.083)
?0.0.093 (0.079)
?0.459 (0.068)***
0.243 (0.078)**
0.096 (0.090)
0.314 (0.084)***
?0.010 (0.018)
?0.540 (0.093)***
?0.377 (0.059)***
?0.335 (0.086)***
?0.127 (0.085)
.594
.239
.241
.000
.002
.284
.000
.586
.000
.000
.000
.131
?0.155 (0.119)
?0.088 (0.080)
?0.042 (0.068)
?0.045 (0.086)
.195
.273
.539
.599
Notes: The multivariate regression model defines a dummy variable indicating a single-choice plan as dependent vari-
able and health status and sociodemographics as independent variables. Individuals in model were offered private insur-
ance plans by their employers in 1996 MEPS data and were younger than 65 yr. Values in parentheses are standard errors.
**Significant at the 0.05 level; ***significant at the 0.01 level.
12 CONTEMPORARY ECONOMIC POLICY

Page 13

TABLE A5
The Multivariate Regression Results of the Relationship between the Health Status Variables and a Dummy Variable
Indicating MC Single-Choice Plan (vs. Non-MC Single-Choice Plan) while Controlling for Sociodemographic Variables
(N 5 1,774)
Coefficient Estimate of Small Firm Variable
p . jzj
Sociodemographics
Age
Female
Nonwhites
Living in suburban area
Living in the South
Living in the Northeast
Living in the West
Family size
High education
Middle education
High income
Middle income
Health status
Perceived good health
Activity Limitation
Chronic disease before 1996
Chronic disease in 1996
?0.006 (0.004)
0.108 (0.104)
?0.457 (0.188)**
0.652 (0.113)***
0.160 (0.128)
0.194 (0.155)
0.877 (0.171)***
0.001 (0.040)
0.030 (0.240)
0.198 (0.128)
0.587 (0.156)***
0.083 (0.139)
.084
.300
.015
.000
.210
.211
.000
.981
.901
.121
.000
.549
?0.005 (0.220)
?0.107 (0.169)
0.065 (0.147)
0.033 (0.173)
.982
.527
.656
.849
Notes: The multivariate regression model defines a dummy variable indicating a single-choice MC plan as dependent
variableandhealthstatusandsociodemographicsvariablesasindependentvariables.Individualsinmodelwereofferedprivate
insuranceplansbytheiremployersin1996MEPSdataandwereyoungerthan65yr.Valuesinparenthesesarestandarderrors.
**Significant at the 0.05 level; ***significant at the 0.01 level.
TABLE A6
MeanDifferenceofSociodemographicsandHealthStatusVariablesbetweenPrivatelyInsuredEmployees in LargeFirms
(N 5 1,059) and Privately Insured Employees in Small Firms (N 5 222)
Mean DifferenceStandard Error
Sociodemographics
Age
Female
Nonwhites
Living in suburban area
Living in the Northeast
Living in the West
Living in the Midwest
Living in the South
Family size
High education
Middle education
Low education
High income
Middle income
Low income
Health status
Perceived good health
Activity limitation
Chronic disease before 1996
Chronic disease in 1996
0.887
?0.033
?0.105**
0.045
0.041
?0.089**
?0.030
0.078*
?0.151
0.003
0.038
?0.038
0.093**
?0.059
?0.034
0.834
0.037
0.023
0.033
0.029
0.028
0.033
0.035
0.109
0.019
0.036
0.037
0.036
0.036
0.029
?0.033
0.032
0.045
0.002
0.018
0.026
0.033
0.025
Notes: The small firm is defined as firms with less than 50 employees. Individuals were offered private insurance plans
by their employers in 1996 MEPS data and are younger than 65 yr. Spouses and children are included in comparison.
*Significant at the .1 level; **significant at the .05 level.
LU, FRANK & MCGUIRE: DEMAND RESPONSE UNDER MANAGED CARE 13

Page 14

TABLE A7
The Multivariate Regression Results of the Relationship
between the Health Status Variables and a Dummy
Variable Indicating Small Firms while Controlling for
Sociodemographic Variables (N 5 1,281)
Coefficient
Estimate of Small
Firm Variable
p . jzj
Sociodemographics
Age
Female
Nonwhites
Living in suburban areas
Living in the South
Living in the Northeast
Living in the West
Family size
High education
Middle education
High income
Middle income
Health status
Perceived good health
Activity limitation
Chronic disease before 1996 ?0.055 (0.205)
Chronic disease in 1996
?0.002 (0.007)
0.134 (0.168)
0.993 (0.238)*** .000
?0.037 (0.191)
?0.064 (0.204)
?0.035 (0.238)
0.977 (0.309)*** .002
0.030 (0.059)
0.294 (0.3340)
?0.060 (0.173)
?0.439 (0.254)
?0.088 (0.253)
.768
.423
.848
.754
.882
.608
.378
.730
.084
.727
0.488 (0.312)
?0.198 (0.242)
.118
.412
.789
.927 0.024 (0.264)
Notes: The multivariate regression model defines
a dummy variable indicating small firm as dependent vari-
able and health status and sociodemographics variables as
independent variables. The small firm is defined as firms
with less than 50 employees. Individuals in model were
offered private insurance plans by their employers in
1996 MEPS data and were younger than 65 yr. Values
in parentheses are standard errors.
***significant at the 0.01 level.
TABLE A8
The Results of Testing Unit Price Difference between
Non-MC and MC Plans
(1) Tests on difference of average unit price between
non-MC and MC
Mean ($) Standard ErrorPr . jtj
Non-MC
MC
Difference
175.85
157.2
18.65
12.67
7.79
14.13 .21
(2) Multiple regression result of the relationship between
unit price of visits and MC dummy
EstimateStandard Error Pr . x2
MC
?0.040.13 .77
Notes: The average unit price is calculated by: (sum of
the total expenditure of each visit/sum of each visit). The
regression model defines the unit price of each visit as
dependent variable. Variables of sociodemographics,
health status, insurance coverage variables, and a dummy
variable indicating play type are independent variables.
Individuals in model were offered a single-choice private
insurance plans by their employers in 1996 MEPS data
and were younger than 65 yr.
TABLE A9
Estimated Coefficients of Sociodemographic and Health
Status Variables in Two-Parts Model
Logistic
Regression
(N = 1,774)
GLM
Regression
(N = 1,444)
Coefficient
Estimates
Coefficient
Estimates
Age
Female
Nonwhites
Living in the suburban
Living in the Northeast
Living in the West
Living in the South
Family size
High education
Middle education
High income
Middle income
Perceived good health
Activity limitation
Chronic disease before 1996
Chronic disease in 1996
?0.011*
0.672**
0.064
?0.084
0.102
?0.584**
?0.186
?0.025
0.208
?0.082
0.337*
0.143
?0.654*
0.596**
1.085**
1.259**
0.007*
0.109
0.067
0.126
0.565**
0.010
0.140
?0.093**
0.037
0.137
0.231*
0.274**
?0.237
0.740***
0.210*
0.611**
Notes: The sample includes individuals who were
offered a single private health insurance plan by employers
in 1996 MEPS data. Spouse and children are included if
they are the dependents of insurance policy holders. Values
inparenthesesarestandarderrors.Thereferencegroupsare
‘‘living in Midwest,’’ ‘‘low education,’’ and ‘‘low income.’’
*Significant at the .1 level; **significant at the .05 level;
***significant at the 0.01 level.
14CONTEMPORARY ECONOMIC POLICY

Page 15

REFERENCES
Buntin, M. B., and A. M. Zaslavsky. ‘‘Too Much Ado
about Two-Part Models and Transformation?’’
Journal of Health Economics, 23, 2004, 525–542.
Call, K. T., B. Dowd, R. Feldman, and M. Maciejewski.
‘‘Selection Experiences in Medicare HMOs: Pre-
enrollment Expenditures.’’ Health Care Financing
Review, 20(4), 1999, 197–209.
Cutler, D. M., M. McClellan, and J. P. Newhouse. ‘‘How
Does Managed Care Do It?’’ Rand Journal of Eco-
nomics, 31, 2000, 526–548.
Cutler, D. M., and R. J. Zeckhauser. ‘‘Adverse Selection
in Health Insurance,’’ in Frontiers in Health Policy
Research, edited by A. Garber. Cambridge, MA:
MIT Press, 1998, 1–32.
Davidson,R., and J. G. Mackinnon. Estimation and Infer-
ence in Econometrics. New York: Oxford University
Press, 1993.
Frank, R. G., J. Glazer, and T. G. McGuire. ‘‘Measuring
Adverse Selection in Managed Health Care.’’ Jour-
nal of Health Economics,19, 2000, 829–854.
Glied, S. ‘‘Managed Care,’’ in Handbook of Health Eco-
nomics, edited by A. Culyer, and J. P. Newhouse.
Amsterdam: North Holland, 2000, 707–745.
TheKaiserFamilyFoundationandHealthResearchandEdu-
cationalTrust.‘‘EmployerHealthBenefitsAnnualSur-
vey.’’ 1998, 2006. Accessed July 10, 2007. http://
www.kff.org/insurance/ehbs-archives.cfm.
Keeler, E. B., and J. E. Rolph. ‘‘The Demand for Epi-
sodes of Treatment in the Health Insurance Exper-
iment.’’ Journal of Health Economics, 7, 1988,
337–367.
Long, S. H., and S. M. Marquis. ‘‘Comparing Employee
Health Benefits in the Public and Private Sectors.’’
Health Affairs, November/December, 1999, 183–193.
Ma, C. A., and T. G. McGuire. ‘‘Optimal Health Insur-
ance and Provider Payment.’’ American Economic
Review, 87, 1997, 685–704.
Manning, W. G, and J. Mullahy. ‘‘Estimating Log Mod-
els: To Transform Or Not To Transform?’’ Journal
of Health Economics, 20, 2001, 461–494.
Medical Expenditure Panel Survey. HC-017: 1996
Health Insurance Plan Abstraction Linked Data,
1996, D33. Accessed on February 20, 2006. http://
www.meps.ahrq.gov/mepsweb/data_stats/download_
data_files_detail.jsp?cboPufNumber5HC-017.
Miller, R. H., and H. S. Luft. ‘‘Managed Care Plan Per-
formancesince1980:ALiteratureAnalysis.’’Journal
of the American Medical Association, 271, 1994,
1512–19.
National Center for Health Statistics. ‘‘Health United
States.’’ 2003. Accessed May 28, 2006. ftp://ftp.cdc.
gov/pub/Health_Statistics/NCHS/Publications/Health_
US/hus03/.
Newhouse, J. P., and The Insurance Experiment Group.
Free for All? Lessons from the RAND Health Insur-
ance Experiment. Cambridge, MA: Harvard Univer-
sity Press, 1993.
Nicholson,S.,K.M.Bundorf,R.M.Stein,andD.Polsky.
‘‘The Magnitude and Nature of Risk Selection in
Employer-Sponsored Health Plans.’’ Health Services
Research, December, 2004, 1817–38.
Pauly, M. V., and S. D. Ramsey. ‘‘Would You Like Sus-
penders to Go with That Belt? An Analysis of Opti-
mal Combinations of Cost Sharing and Managed
Care.’’ Journal of Health Economics, 18, 1999,
443–458.
Polsky, D., and S. Nicholson. ‘‘Why Are Managed Care
Plans Less Expensive: Risk Selection, Utilization, or
Reimbursement?’’ Journal of Risk and Insurance, 71,
2004, 21–40.
Reschovsky, J. D. ‘‘Do HMOs Make A Difference?
Data and Methods.’’ Inquiry, 36, 1999/2000, 378–389.
Zeckhauser, R. J. ‘‘Medical Insurance: A Case Study of
the Trade-Off between Risk Spreading and Appro-
priate Incentives.’’ Journal of Economic Theory, 2,
1970, 10–26.
LU, FRANK & MCGUIRE: DEMAND RESPONSE UNDER MANAGED CARE 15