The Effect of Benefits, Premiums, and
Health Risk on Health Plan Choice
in the Medicare Program
Adam Atherly, Bryan E. Dowd, and Roger Feldman
Objective. To estimate the effect of Medicare1Choice (M1C) plan premiums and
benefits and individual beneficiary characteristics on the probability of enrollment in a
Data Source. Individual data from the Medicare Current Beneficiary Survey were
combined with plan-level data from Medicare Compare.
Study Design. Health plan choices, including the Medicare1Choice/Fee-for-Service
decision and the choice of plan within the M1C sector, were modeled using limited
information maximum likelihood nested logit.
Principal Findings. Premiums have a significant effect on plan selection, with an
estimated out-of-pocket premium elasticity of ?0.134 and an insurer-perspective
drug benefits having the largest marginal effect. Sicker beneficiaries were more likely to
choose plans with drug benefits and diabetics were more likely to pick plans with vision
on individual characteristics.
Key Words. Medicare, health plan choice, Medicare1Choice, managed care,
Under the current structure of the Medicare program, payments to
Medicare1Choice (M1C) managed care plans are determined by a complex
administrative formula that takes data on historical Medicare fee-for-service
(FFS) costs and inflates it at an annual rate determined by Congress. When
M1C plans enter low-payment counties, they typically offer limited benefit
packages and charge premiums. Conversely, in high-payment counties, plans
offer generous benefit packages——including prescription drug benefits——and
often charge no out-of-pocket premiums. As a result, the Medicare program,
which is financed by nationally uniform tax rates, provides a decidedly
uneven benefit package that varies by health plan and county of residence
Geographic disparities in benefits and concerns about rising costs have
generated interest in alternative approaches to paying health plans in the
Medicare program. The difficulty in improving the current administrative
pricing mechanism is that the government does not directly observe M1C
plan costs, making the determination of an appropriate payment level by the
Centers for Medicare and Medicaid Services (CMS) difficult. M1C plans are
required to report an estimate of their profits to Medicare. If profits are above
the plan’s profit rate on commercial business, the plan must offer additional
benefitsto dissipate the extra earnings. Because these benefitsare offered only
can be greater than the marginal value of the benefits to beneficiaries.
Therefore, some of the extra benefits offered may be economically inefficient.
One alternative that has been proposed is to replace the current M1C
payment mechanism with a system whereby plan payments would be
a financial incentive to pick the low-bid plans (Dowd, Feldman, and
Christianson 1996). This ‘‘competitive pricing’’ alternative relies on competi-
tion to determine government payments, out-of-pocket premiums, and
optional benefits. But this approach has two potential problems. First, if
Medicare beneficiaries are unresponsive to differences in out-of-pocket
premiums, plans will lack a strong incentive to submit low bids. Second,
(Feldman and Dowd 2000; Cutler and Zeckhauser 1997).
This study examines the factors that determine beneficiaries’ choices of
Medicare health plans. We estimate the effect of out-of-pocket premiums and
benefits on the probability of choosing a M1C plan versus FFS Medicare, as
well as the choice of a specific plan within the M1C ‘‘sector.’’ We also
examine the relationship between benefits and adverse selection within the
benefit elasticities using Medicare health plans that uses individual level data.
This study was supported in part by the Center for Medicare and Medicaid Services (contract no.
500-92-0014). The authors would like to thank our project officer, Ron Deacon, for advice and
assistance. Opinions expressed herein are solely those of the authors.
Address correspondence to Adam Atherly, Ph.D., Assistant Professor, Department of Health
Policy and Management, Emory University, 1518 Clifton Rd. NE, Atlanta, GA 30322. Bryan E.
Dowd, Ph.D., and Roger Feldman, Ph.D., are with the Division of Health Services Research,
University of Minnesota, Minneapolis.
848HSR: Health Services Research 39:4, Part I (August 2004)
benefit elasticities among M1C plans using aggregate market share data.
Previous research has demonstrated that employees’ choices of health plans
Lave 1997; Cutler and Reber 1998; Dowd and Feldman 1994/1995; Feldman
et al. 1989). However, less is known regarding the impact of premiums on
choice in the Medicare population. Buchmueller (2000) used data from the
University of California retiree health benefits program to examine the effect
of premiums on the probability of switching health plans and to estimate the
price elasticity of demand for FFS versus managed care. Buchmueller
estimated that a $10 premium increase would lead to a 0.8 percent increase in
the probability of switching among managed care plans. The estimated
elasticity for FFS versus managed care was ?0.16.
There are only two national studies of beneficiary sensitivity to out-of-
pocket premiums in the Medicare population (Feldman et al. 1993; Dowd,
Feldman, and Coulam 2003); both used aggregate market share data. The price
elasticity of health plan choice estimated by Feldman et al. (1993) was ap-
proximately ?2. Dowd, Feldman, and Coulam (2003) examined the factors
affecting choice within the M1C sector and found that a $10 increase in the out-
of-pocket premium would result in a loss of four percentage points of M1C
market share (e.g., from 25 to 21 percent). Dowd, Feldman, and Coulam (2003)
included a large number of benefits in their plan choice equation and found that
outpatient drug benefits greater than $800 per year and coverage of dental, eye
using a measure of selection based on expected Medicare reimbursements.
Shefound littleevidencethat coverageofparticularservices(e.g.,prescription
drugs) or higher premiums induced biased selection by beneficiaries.
However, healthier beneficiaries were attracted to plans with lower primary
care copayments and larger networks of primary care physicians, while sicker
beneficiaries were attracted by lower specialty care copayments, larger
specialty care panels,and higher perceivedplanquality. Feldman, Dowd, and
Wrobel (2003) found that offering a prescription drug benefit was associated
with adverse selection, but some benefits such as dental coverage were
associated with favorable selection.
Benefits, Premiums, Health Risk and Health Plan Choice849
Our study is based on the expected utility model of health plan choice. We
assume that each beneficiary chooses the health plan that yields the highest
expected utility. An individual’s utility from a given plan is determined by a
combination of plan characteristics and interactions between individual and
plan characteristics, as given by the following equation:
Uij¼ xjb1þ ðx?
where Uijrepresents the utility associated with plan j for individual i, xjis a
vector of plan characteristics, siis a vector of individual characteristics, b1and
enter this model only as interactions with plan characteristics because the
main effect of individual characteristics is differenced away.
The standard method to estimate models of this type is McFadden’s
single choice among j alternatives. The number of alternatives in this study is
equal to the number of M1C plans available in the individual’s county plus
the FFS option. The number of alternatives is thus the same for all individuals
in a given county, but varies across counties.
McFadden’s method assumes that the error terms in the model are
independent and homoskedastic. Because a Hausman test (Greene 2000, p.
865) rejected this assumption in our data, a nested logit model was employed
instead. Nested logit splits the M1C sector into a separate ‘‘branch’’ from the
FFS sector and estimates an interbranch choice model for the choice of M1C
versus FFS sectors plus a within M1C choice of a particular M1C plan.
Formally, the probability of joining a particular plan j in sector k is:
Pð j;kÞ ¼ Pð jjkÞ?PðkÞ ¼eb0
where xi again represents plan attributes and xi*si the plan–individual
interactions that influence choice within sector, and yirepresents individual
and market characteristics that influence choice of sector. There are K nests
and J choices within the nest. We assume that individuals firstchoose between
the two sectors and then, if they choose the M1C sector, choose one of the
M1C plans available in their county. The ‘‘inclusive value’’ is Ikand is
850 HSR: Health Services Research 39:4, Part I (August 2004)
An estimated coefficient is tkon the inclusive value, Ik. The inclusive
value represents the expected utility of the M1C nest for a particular
individual and is typically calculated for each of the K nests. This analysis
presents two problems that deviate from the standard nested logit framework.
First, the FFS sector is a degenerative branch (i.e., there is no within-nest
choice). Although individuals in the FFS sector have the option to purchase
on the individual’s choice set of supplemental insurance options. The
degenerative FFS branch prevents estimation of the within-nest b for FFS
and requires tFFSbe fixed equal to one (Hunt 2000). Second, because there is
no variation in plan attributes in the FFS sector (FFS Medicare benefits are
uniform nationwide), the model is estimated sequentially from the bottom-up
using Limited Information MaximumLikelihood (LIML). LIMLestimation is
consistent, although not asymptotically efficient.
Empirically, it is necessary to predict both the within-nest choice of M1C
plans and the choice of nest. The variables for the within-nest choice equation
are the beneficiary’s out-of-pocket premium, benefits offered by M1C plans
beyond the statutory requirement, and interactions of benefits with individual
characteristics (precise variable definitions are described below). The model
a given plan, interacted with plan characteristics. Individual characteristics do
not enter the model directly. Instead, it is hypothesized that certain plan
characteristics may provide greater utility for some individuals than others.
angina pectoris [CHD], and hypertension) interacted with a prescription drug
indicator variable. This interaction indicates whether individuals with these
chronic illnesses systematically select plans with prescription drug benefits.
The four chronic illnesses were selected by using the 1997 Medicare Current
Beneficiary Survey (MCBS) Cost and Use file (which contains data on out-of-
pocket spending)and regressingout-of-pocketspending on prescription drugs
in the FFS sector on indicators for 18 chronic illnesses. These four chronic
Benefits, Premiums, Health Risk and Health Plan Choice851
illnesses had the largest marginal effects on out-of-pocket drug spending.
Other variables might have a large impact on choice, but we were concerned
about the subset of those variables that affect expenditures, as well.
A variable indicating that the individual completed a college education is
more education select plans with prescription drug benefits, controlling for the
premium. Given that drug coverage in zero premium M1C plans may be
underpriced relative to its actuarial value, drug coverage should be economic-
likely to understand this. A diabetes indicator variable is interacted with a
measure oftheplan’s vision benefits, becausewe expectthatdiabeticsaremore
interested in vision services (diabetes is the leading cause of blindness in the
United States). Finally, premium is interacted with income to measure whether
higher income individuals have less-elastic demand for health plans.
The variables for the choice of M1C or FFS sector are individual
characteristics (age, health status, income, and marital status), the cost of
Medigap coverage, and the county M1C payment rate. The previously cited
health (15excellent, 55poor) and the number of chronic illnesses.1Income
is represented by an indicator variable equal to one if the individual’s income
is more than $20,000.2Marital status might affect plan choice through
coordination of choice by spouses or through the availability of income
substitutes for professional care. Medigap premiums are measured at the
county level, and individuals who live in counties with higher Medigap
premiums are expected to be more likely to join an M1C plan. (Medigap
plans are an economic substitute for the additional benefits offered by M1C
plans.) The government payment rate to M1C plans in the county is in-
cluded as a proxy for additional benefits that are not captured in our model.
Higher payment rates are expected to be positively associated with joining a
The main data source is the 1998 Medicare Current Beneficiary Survey
(MCBS). The MCBS is a rolling cohort survey of Medicare beneficiaries. The
852 HSR: Health Services Research 39:4, Part I (August 2004)
50 states, the District of Columbia, and Puerto Rico. Details of the survey are
available in Adler and Phil (1994). This analysis utilized data on health care
expenditures and plan choices from the Access to Care portion of the survey,
as well as data on health status from the personal interview portion of the
The 1998 MCBS Cost and Use dataset contained 21,020 observations.
Several groups were excluded in order to restrict the analysis to those
beneficiaries for whom we could characterize the plan choice set and who
were likely to consider a M1C plan. Individuals were excluded if someone
else paid for their additional coverage beyond the basic Medicare benefits.
These included currently employed individuals and those who had a
supplemental policy from their former employer (n55,038) or were eligible
for Medicaid (n55,566).3The disabled population under age 65 was
excluded (n53,202).4The included sample is quite similar to the excluded
aged population, with a mean age of 75.6 (versus 75.5 for the excluded
sample), 58.7 percent female (versus 61.1 percent), and 21.1 percent with fair
or poor self-rated health (versus 25.9 percent).
Individual data from the MCBS were matched to the Medicare
Compare dataset, which provided information on benefit packages offered
information on emergency/urgent care, cost sharing for outpatient and
inpatient care, and prescription drug benefits, as well as terms of coverage for
mental health, preventative services, hearing services, dental care, and vision
In the MCBS, individuals were asked about M1C enrollment for each
monthduring1998.Individualswhoreported beinginaMedicareM1C plan
during January 1998 were characterized as having chosen an M1C plan;
individuals who reported not being in an M1C plan on that date were
an M1C, the ‘‘plan number’’ reported by the MCBS was matched to plans
that reported operating in the individual’s county. A successful match was
achieved forall but 269 of the3,824 M1C enrollees(a 93 percentmatch rate).
Nineteen of the failed matches were enrolled in Medicare cost-reimbursed
plans; the other 249 were enrolled in risk plans that were not listed as
operating in the individual’s county of residence. Because the choice set for
analysis. One observation had to be excluded due to missing individual data.
This left a sample of 7,062, including 3,555 M1C enrollees and 3,507 FFS
Benefits, Premiums, Health Risk and Health Plan Choice853
One limitation in theMedicare Comparedata is that 39 percent of plans
offer multiple options. For example, a plan might offer an option with a zero
premium and limited drug benefits and an option that includes a better drug
benefit but also charges a higher premium. The MCBS plan codes link
to the plan, but not to the particular option within the plan. When a plan
included multiple options, the benefits and premium associated with the
lowest premium option were used. To control for possible omitted variable
bias, an indicator for plans that offer multiple options is included in the
Finally, Medigap premiums were drawn from an Abt Associates survey
of the largest Medigap providers in the nation. The survey collected premium
data for five Medigap insurers nationally; together, these five insurers issue
morethan 50 percent of the combined group and individualMedigap policies
in 1996. We assume that Medigap premiums in 1998 were generally
proportionate to the premiums in 1996.
Frequencies and means for plan benefits are presented in Table 1 (along
with the benefits’ expected relationships to the probability of choice). These
descriptive statistics represent sample means and frequencies of plans in the
analysis sample, which are not equivalent to unweighted means and
frequencies among all plans. Because large plans operate in many counties,
plans, so their benefit options will receive a greater weight. The means and
frequencies therefore show the proportion of plans presented to the sample
Two different models are presented in this section. First, results from a nested
logit model that includes only plan characteristics are presented. Second,
results are presented that include both plan characteristics and interactions
between individual and plan characteristics.
Premium elasticities are estimated both from the beneficiary’s perspec-
tive (representing a change in out-of-pocket premium) and the insurer
perspective (representing a change in out-of-pocket premium plus the plan
payment). The premium elasticity is given by:
ejk¼ bpremium? premiumj ? ½ð1 ? pjjkÞ þ tð1 ? pkÞpjjk?
854 HSR: Health Services Research 39:4, Part I (August 2004)
The first half of the bracketed term ðbpremium? premiumj?½1 ? pjjk?Þ represents
the within-sector price elasticity (the effect of a premium change on the
ðbpremium? premiumj ? t½1 ? pjjk?Þrepresentstheintersectorpriceelasticity(the
effect of a premium change on the probability of joining the M1C sector).
Table 2 presents the results of the nested logit model with plan
characteristics only. Premium has a negative sign and is statistically significant
(po.001). The total out-of-pocket premium elasticity of demand is equal to
?0.13. This indicates that a 10 percent increase in a M1C plan’s premium is
associated with a 1.3 percent decrease in the plan’s enrollment. The estimated
within sector out-of-pocket premium elasticity is ?0.12, while the cross sector
out-of-pocket premium elasticity is equal to ?0.01. This indicates that most of
the 1.3 percent who disenroll from the M1C plan as a result of the premium
increase will choose another M1C plan, if available, other than FFS Medicare.
The insurer-perspective within-sector premium elasticity is equal to
?3.87, while the cross-sector elasticity is ?0.69. These larger elasticities
Table1: Medicare M1C Plan Characteristics
Out-of-pocket premium (dollar)
Any drug benefit with a limit over $300?
Dollar copayment for generic
Is the M1C plan a Staff Model HMO?
Is the M1C plan a Group Model HMO?
Dollar copayment for primary care
HMO for profit?
Does the plan offer any dental coverage?
Does the plan offer any vision coverage?
Dollar copayment for outpatient
mental health care
Does the plan offer any
Dollar copayment for ambulance services
Dollar copayment for emergency room$33.81—— Negative
Does the plan offer multiple options? 39.3%—— Positive
Benefits, Premiums, Health Risk and Health Plan Choice855
reflect the larger value of the basepremium usedto calculate the elasticity. If a
plan increased its out-of-pocket premium by 10 percent of the total premium,
its enrollment would decline by 45.6 percent.
expected change in market share for a particular plan j for a given dollar
change in premium:
ð1Þ Change in Market Share Planjk¼bpremium? MSk? MSjjk½ð1 ? MSjjkÞ
þ t ? ð1 ? MSkÞ MSjjk?
While this market-share equation is analogous to the elasticity equation given
above, it has a decided advantage over the elasticity because it is not
dependent on a particular premium value. The first half of (1) represents the
Table2: Nested Logit Results
Variable Coefficient Standard Error T-statisticPjZj > 0
Within Nest Choice
M1C / FFS Choice
Out of pocket premium
Primary care copayment
Mental health copayment
Emergency room copayment
Multiple plan options
Income over $20,000
Number of chronic illnesses
County Medigap premium
Inclusive value parameter
856 HSR: Health Services Research 39:4, Part I (August 2004)
within sector change in market share ðbpremium? MSjjk? ½1 ? MSjjk?Þ, multi-
plied by the proportion of the sample in the M1C sector (MSk). The second
half ðbpremium? MSk? MSjjk? t?½1 ? MSk? ? MSjjkÞ represents the intersector
change in market share. The probability that an individual picks the M1C
sector (MSk) is 0.308. The conditional probability of picking a particular plan
within the M1C sector ðMSjjkÞ is equal to 0.252.6
The change in the jthplan’s unconditional market share (including the
FFS sector) for a $1 change in the jthplan’s premium is ?0.00062. This
in the plan’s market share. The typical M1C plan in this model has an
unconditional market share (i.e., including FFS) of 7.76 percent, which
suggests that if the typical plan raised its premium $10, its market share would
beneficiaries switching to FFS as a result of the premium increase.
The presence a drug benefit is associated with an increased probability
of joining a particular M1C plan (b50.589, po.001). Similarly, the
probability of joining is negatively correlated with higher copayments for
generic prescription drugs (b5 ?.080, po.001).
Individuals are more likely to join a staff model plan, relative to the
reference group (IPA model), while individuals were not more likely to join
group model plans. For-profit status of the M1C plan is strongly (t514.75)
and negatively (b5 ?.917) associated with enrollment. Overall, beneficiaries
prefer nonprofit staff model plans, although the large coefficient on the
nonprofit variable suggests that it may be correlated with omitted variables,
such as the brand name advantage enjoyed by plans such as Blue Cross/Blue
Shield or Kaiser.
Among other plan characteristics, vision coverage, multiple benefit
options, and having copayments for primary care, emergency room services,
coverage are not. Although the coefficients for vision coverage, emergency
room copayments, ambulance services, and mental health coverage have the
be due to omitted plan characteristics associated with higher primary care
copayments, or plans with superior provider networks that may have higher
primary care copayments. The positive coefficient for multiple benefit
packages indicates that more popular plans offer multiple benefit packages
Benefits, Premiums, Health Risk and Health Plan Choice857
In the model predicting sector choice, M1C enrollees are younger
(b5 ?.039, po.001), and healthier as measured by self-rated health.
Interestingly, M1C joiners were more likely to have a chronic illness.
Demographically, M1C members were more likely to be married (b5.279,
po.001) and less likely to have income over $20,000 (b5 ?.562, po.001).
County average Medigap premiums are positively correlated with the
probability of joining a M1C plan, reflecting the substitution between
M1C plans and Medigap, with a Medigap premium elasticity of 0.209.
To better understand the magnitude of the effects reported in Table 2,
the coefficients for key significant variables were transformed to marginal
probabilities (Table 3). The most important plan characteristic is the drug
benefit. Offering a drug benefit increases the probability of selecting a
particular M1C plan by 43.7 percent and the probability of selecting the
percent) and mental health copayment (2.2 percent) are much less important.
characteristics and interactions of individual and plan characteristics. The
interaction of high income and premium was significant and positive,
indicating that the premium elasticity for M1C plans decreases as income
increases. College education, interacted with the plan offering a drug benefit,
was significant (p5.025) and positive, indicating that the attractiveness of
drug coverage increases with education.
Table3: Marginal Probabilities and Elasticities
Elasticity / Probability
Within M1CM1C / FFS Total
Within Nest Choice
M1C / FFS Choice
Any mental health copayment
County Medigap premium
Income over $20,000
Number of chronic illnesses
858 HSR: Health Services Research 39:4, Part I (August 2004)
Our two interactive markers of selection indicated that drug and vision
benefits were attractive to high-cost beneficiaries. First, the coefficient on the
interaction between chronic illness and drug benefits is positive, indicating
that beneficiaries with these chronic illnesses are more likely to choose
plans with drug benefits (b50.102, p50.03). Second, we found that
diabetics were more likely to join plans that offer vision benefits (b50.021,
Table 4 also includes plan characteristics, for which the results are
generally similar to Table 2, and individual characteristics. The coefficient of
Table4: Nested Logit Coefficients Including Individual Interactions
Variable Coefficient Standard ErrorT-statisticPjZj > 0
Within Nest Choice
Any drug benefit
Primary care copayment
Mental health copayment
Emergency room copayment
Multiple plan options
Income over $20,000*premium
College education*drug benefit
Diabetes, hypertension, arthritis and
coronary heart disease*drug benefit
0.02190.0128 1.7190 0.0856
M1C / FFS Choice
Income over $20,000
Number of chronic illnesses
County Medigap premium
Inclusive value parameter
Benefits, Premiums, Health Risk and Health Plan Choice859
self-rated health is negative, indicating favorable M1C sector selection.
However, the coefficient of chronic illness is not a significant predictor of
M1C sector choice in the interactive model.
Proponents of replacing the currentM1C payment mechanism witha system
based on competitive pricing assume that market forces will discipline M1C
premiums are significantly associated with the probability of enrollment. The
estimated out-of-pocket premium elasticity of demand from the consumer’s
perspective is low (?0.134) but similar to those reported in other studies. The
same elasticity from the insurer’s perspective (?4.57) is much larger.
Alternatively, looking at the effect of premiums on changes in market share,
we found that a typical plan would lose .62 percentage points of their market
share with a $10 increase in premium.
Is the loss of .62 percentage points of a plan’s market share enough to
the scope of this analysis, but the insurer perspective elasticity shows that total
revenue will decline if a plan raises its premium. Although the impact of an
increase in out-of-pocket premiums on profits cannot be estimated directly, it
is not unreasonable to speculate that they may decline as well, providing an
incentive for plans to bid competitively. However, the elasticity is not likely
large enough to drive a slightly higher-priced plan from the market, as some
health plans claimed during the attempts to demonstrate competitive pricing
(Dowd, Coulam, and Feldman 2000).
The price elasticities are estimated over a fairly small range of premiums
(the mean premium is $10.17, with nearly 80 percent of plans charging a
premium less than $20, and nearly two-thirds of the selected plans not charging
plan charged a premium. Therefore, the results of this study accurately reflect
the responsiveness of M1C enrollees to the current incentive structure,
although they may not accurately predict responses to larger premium
variation. Similarly, the supplemental plan benefits observed in the data are
be generalized to benefit packages substantively different from those observed.
Controlling for Medigap premiums, we found that lower-income
beneficiaries are more likely than those with higher incomes to join the
860 HSR: Health Services Research 39:4, Part I (August 2004)
M1C sector. As Medigap premiums have increased, the M1C sector has
become a refuge for Medicare beneficiaries whose income is too high for
Medicaid but too low to afford a Medigap plan. Plan withdrawals from the
M1C sector are therefore likely to have a disproportionate impact on lower
income Medicare beneficiaries.
Beneficiaries prefer plans with better benefits, controlling for
premiums. The benefit with the largest marginal effect on enrollment is
prescription drug coverage. A plan dropping a typical prescription drug
benefit would lose market share equivalent to a $62 increase in monthly
premium. This large effect suggests that the addition of a prescription drug
benefit to the FFS sector may significantly decrease M1C enrollment.
rated health, but unfavorable selection based on the number of chronic
illnesses (in the model without interactions). Unfavorable selection into the
M1C sector may be motivated in part by a desire for prescription drug
coverage. Adding a prescription drug benefit to the FFS sector may therefore
increase FFS costs more than expected by diverting high cost beneficiaries
with chronic illnesses away from the M1C sector.
The second question investigated by this study was whether particular
M1C plan benefits encourage enrollment by higher-cost beneficiaries.
Several interactions between individual and plan characteristics were
significant, suggesting that individuals sort themselves systematically into
plans based on individual characteristics. The significant interaction between
income and premium suggests that the price elasticity of demand for health
insignificantly different from zero for higher income individuals. This
finding——that poorer individuals are more concerned about $10 monthly
premium differences than higher income individuals——is not surprising, but it
shows that competition may lead to segmentation of the M1C population
based on income.
The finding that individuals with a college education are more likely to
select a plan with a drug benefit suggests that they are better able to determine
the value of this benefit in a heavily subsidized environment. Alternatively,
more-educated Medicare beneficiaries may have greater demand for
prescription drug coverage. Similar to the premium results, this finding
suggests that competition over drug benefits may lead to segmentation based
Finally, the significant interactions between chronic illness and a
prescription drug benefit and between diabetes and vision care indicate that
Benefits, Premiums, Health Risk and Health Plan Choice861
M1C plans that offer these benefits will enroll less-healthy beneficiaries as a
result. In 1998, M1C plans in high-payment areas were sufficiently
overcompensated to the point that they were willing to accept some adverse
planpayments more into line withcosts, plans may begin to dropbenefits that
attract high-cost beneficiaries.
One potential solution to this problem would be to establish a uniform
benefit structure for M1C plans. However, a uniform benefit structure could
stifle innovation by plans. Also, requiring a single benefit structure for all
beneficiaries would be optimal only if Medicare beneficiaries had uniform
preferences, which is unlikely. A second option would be to create multiple
‘‘packages’’ of benefits that could be offered together, as was done with
Medigap plans (Atherly 2001). However, although the model plan structure
has improved consumers’ ability to select a Medigap plan (Rice, Graham, and
Fox 1997), it appears that adverse selection has effectively eliminated options
with a prescription drug benefit. This suggests that the ‘‘package’’ approach
being driven from the market through adverse selection.
A third option is to risk adjust plans’ bids. The CMS has explored and
implemented a number of different risk adjustment systems over the years.
Under a competitive-pricing system with satisfactory risk adjustment, plans
considering offering a benefit would compare only the marginal cost of the
benefit to the marginal beneficiary’s willingness to pay. The impact of benefits
on risk selection could be ignored because the costs associated with selection
Since these data were collected and analyzed, the M1C program has
been in decline as plans have reduced service areas, increased premiums, and
reduced benefits. Not surprisingly, enrollment in the program has fallen from
6.2 million in 1999 to less than 5 million in 2002. A key benefit reduction by
from 23 percent to 69 percent, drug copayments increased, and coverage for
brand name drugs was reduced (Achman and Gold 2002). As government
payments to health plans have declined, it is likely plans are becoming more
concerned about the type of adverse selection we found. Although reductions
in prescription drug coverage decrease M1C plan expenditures through the
creation of a healthier risk pool, this is not necessarily a desirable goal for the
program. Unless government payments are tied to the benefits offered by
plans or a more effective risk-adjustment mechanism is used to compensate
862 HSR: Health Services Research 39:4, Part I (August 2004)
plans that attract higher-risk enrollees with richer benefit packages, M1C
plans are likely to continue to reduce the benefits favored by high-cost
1. The chronic illness include diabetes, heart disease, hypertension, myocardial
infarction, stroke, skin cancer, other cancer, and arthritis.
3. Medicaid eligibility was not directly observed, but it was proxied by eliminating
those with incomes under $10,000.
4. One-hundred-eighteen individuals were in more than one of these categories.
5. CMS recently has completed the first analysis of enrollment in the individual
products marketed by M1C plans and provided us with the following results.
Among all M1C enrollees who have outpatient prescription drug coverage in
January 2002, 87 percent obtained coverage through their plan’s basic benefit
package (the one we use for analysis) not through an optional supplementary
package soldby the same plan. Among M1C enrollees who do not have outpatient
prescription drug coverage, 73 percent were not offered any prescription drug
coverage by their plan, whereas only 27 percent were offered a drug option and
turned it down. Thus, for drug coverage, which is probably the most important
a very important role in determining who has drug coverage and who does not.
6. Because we have excluded some types of Medicare beneficiaries (e.g., those with
employer-sponsored supplemental insurance), these probabilities are not general-
izable to the Medicare population as a whole.
Achman, L., and M. Gold. 2002. Trends in Medicare1Choice Benefits and Premiums,
1999–2002. Report no. 580. New York: Commonwealth Fund.
Adler, G., and M. Phil. 1994. ‘‘A Profile of the Medicare Current Beneficary Survey.’’
Health Care Financing Review 15 (4): 153–63.
Atherly, A. 2001. ‘‘Medicare Supplemental Insurance: Medicare’s Accidental Step-
child.’’ Medical Care Research and Review 58 (2): 131–61.
Buchmueller, T. 2000. ‘‘The Health Plan Choices of Retirees under Managed
Competition.’’ Health Services Research 35 (5): 949–76.
Buntin, M. 2000. ‘‘Risk Selection and Product Differentiation in Medicare HMOs.’’
Unpublished manuscript. Santa Monica, CA: RAND.
Cutler, D., and S. Reber. 1998. ‘‘Paying for Health Insurance: The Tradeoff between
Competition and Adverse Selection.’’ Quarterly Journal of Economics 113 (2):
Benefits, Premiums, Health Risk and Health Plan Choice863
Cutler, D. M., and R. J. Zeckhauser. 1997. Adverse Selection in Health Insurance. Download full-text
Cambridge, MA: Harvard University and the National Bureau of Economic
and Competitive Pricing.’’ Health Affairs 19 (5): 9–29.
Dowd, B., and R. Feldman. 1994/1995. ‘‘Premium Elasticities of Health Plan Choice.’’
Inquiry 31 (4): 438–44.
Dowd, B., R. Feldman, and J. Christianson. 1996. Competitive Pricing for Medicare.
Washington, DC: AEI Press.
Dowd, B., R. Feldman, and R. Coulam. 2003. ‘‘The Effect of Health Plan
Characteristics on Medicare1Choice Enrollment.’’ Health Services Research 38
(1, part 1): 113–35.
Feldman, R., and B. Dowd. 2000. ‘‘Risk Segmentation: Goal or Problem?’’ Journal of
Health Economics 19 (4): 499–512.
Feldman, R., B. Dowd, and M. Wrobel. 2003. ‘‘Risk Selection and Benefits in the
Medicare1Choice Program.’’ Health Care Financing Review 25 (1): 23–36.
Feldman, R., M. Finch, B. Dowd, and S. Cassou. 1989. ‘‘Demand for Employment-
Based Health Insurance Plans.’’ Journal of Human Resources 24 (1): 115–42.
Feldman, R., C. Wisner, B. Dowd, and J. Christianson. 1993. ‘‘An Empirical Test of
Competition in the Medicare HMO Market.’’ In Competitive Approaches to Health
Care Reform, edited by R. Arnold, R. Rich, and W. White. Washington, DC:
Urban Institute Press.
Greene, W. 2000. Econometric Analysis, 4th ed. Upper Saddle River, NJ: Prentice Hall.
Hunt, G. 2000. ‘‘Alternative Nested Logit Model Structures and the Special Case of
Partial Degeneracy.’’ Journal of Regional Science 40 (1): 89–113.
McBride, T. 1998. ‘‘Disparities in Access to Medicare Managed Care Plans and Their
Benefits.’’ Health Affairs 17 (6): 170–80.
McFadden, D. 1974. ‘‘Conditional Logit Analysis of Qualitative Choice Behavior.’’ In
Frontiers in Econometrics, edited by P. Zarembka, pp. 105–42. New York:
Rice, T., M. Graham, and P. Fox. 1997. ‘‘The Impact of Policy Standardization on the
Medigap Market.’’ Inquiry 34 (2): 106–16.
Royalty, A., and N. Solomon. ‘‘Price Elasticities in a Managed Competition Setting.’’
Journal of Human Resources 34 (1): 1–41.
Scanlon, D., M. Chernew, andJ. Lave. 1997. ‘‘Consumer Health Plan Choice: Current
Knowledge and Future Directions.’’ Annual Review of Public Health 18: 507–28.
864HSR: Health Services Research 39:4, Part I (August 2004)