Health Inssurance

Paying for Health Insurance: The Trade-Off between Competition and Adverse Selection
Author(s): David M. Culter and Sarah J. Reber

Source: The Quarterly Journal of Economics , May, 1998, Vol. 113, No. 2 (May, 1998),
pp. 433-466
Published by: Oxford University Press
Stable URL: https://www.jstor.org/stable/2586909
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Oxford University Press is collaborating with JSTOR to digitize, preserve and extend access to
The Quarterly Journal of Economics
This content downloaded from

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
All use subject to https://about.jstor.org/terms
compromis
PAYING FOR HEALTH INSURANCE: THE TRADE-OFF
BETWEEN COMPETITION AND ADVERSE SELECTION*
DAVID M. CUTLER AND SARAH J. REBER
We use data on health plan choices by employees of Harvard University to

compare the benefits of insurance competition with the costs of adverse selection.
Moving to a voucher-type system induced significant adverse selection, with a
welfare loss of 2 to 4 percent of baseline spending. But increased competition
reduced Harvard's premiums by 5 to 8 percent. The premium reductions came
from insurer profits, so while Harvard was better off, the net effect for society was
only the adverse selection loss. Adverse selection can be minimized by adjusting
voucher amounts for individual risk. We discuss how such a system would work.
Governments are increasingly turning to market forces as a

way to limit the cost of social insurance. Traditionally, social
insurance programs were operated as nonmarket goods; govern-
ments mandated participation in a central program, collected
revenues to finance the program, and ran the insurance system.
There was no role for competition among suppliers in providing
the basic benefit.
As the costs of social insurance have increased, however, the
centralized model of social insurance is coming under increasing
strain. In the United States, for example, recent proposals have
called for replacing the Medicare program with a health care
voucher for the elderly [Aaron and Reischauer 1995; Cutler 1996].
The voucher would guarantee people a basic insurance plan, but
the plans would be privately run. Competition among plans would
generate plan premiums and enrollments. Similarly, long-
standing proposals for Social Security reform have called for
replacing the current system with a system of individual ac-
counts, where people would make saving, investment, and annu-
itization decisions on their own [Advisory Council on Social
Security 1996; Feldstein 1996].
The trend is not just domestic. In the United Kingdom the
National Health Service has moved to encourage more competi-
tion in recent years, with the establishment of hospital "trusts"
that bid for patients and partially fixed payments to physicians,
* We are grateful to Sally Zeckhauser, Thomas Schmitt, and Lydia Cummings

for providing us data, to them, Jason Furman, Edward Glaeser, Lawrence Katz,
Michael Kremer, Christopher Ruhm, Sujata Sanghvi, Andrei Shleifer, Douglas
Staiger, Richard Zeckhauser, and two anonymous referees for helpful comments,
and to the National Institutes on Aging for research support.
? 1998 by the President and Fellows of Harvard College and the Massachusetts Institute of
Technology.
The Quarterly Journal of Economics, May 1998

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
434 QUARTERLY JOURNAL OF ECONOMICS
from which they contract for medical services. And recent Social
Security reforms have allowed people to opt out of the public
system if they have a suitable private pension.
While encouraging competition seems natural for economists,
competition in insurance markets is somewhat problematic. Con-
sider the case of a voucher for health insurance, the subject of this
paper. Under a voucher system, people will have incentives to
search for efficient plans, and insurers will have incentives to
limit unnecessary care. Both of these are valuable. But adverse
selection is a concern as well. The less healthy naturally prefer
more generous insurance than the healthy. As a result, more
generous insurance plans will cost more than benefit differences
alone would dictate. Healthy people, in turn, will want to avoid
those plans, to keep from subsidizing their less healthy brethren.
If adverse selection is large, it may destroy the market for the
generous insurance entirely. The same could be true for annuity
markets under a privatized social security system.
There is a lengthy literature documenting adverse selection
in annuity markets [Friedman and Warshawsky 1988, 1990;
Mitchell, Poterba, and Warshawsky 1997] and in multiple-choice
health insurance arrangements [Price and Mays 1985; Scheckler
and Schulz 1987; Brown et al. 1993; Feldman and Dowd 1993;
Royalty and Solomon 1995; Riley et al. 1996]. But the literature is
generally silent on the welfare costs of this adverse selection, and
on how these costs compare with the benefits (if any) from
increased competition. In deciding on social insurance reform,
these latter issues are the fundamental ones.
In this paper we analyze empirically the gains and losses
from competitive reforms in health insurance payments. We use
data from a reform carried out at Harvard University. In 1995
Harvard moved from a system of subsidizing generous insurance
to a system of paying a fixed contribution independent of plan
choice (essentially a voucher system). This policy change in-
creased the price to employees of the most generous policy by over
$500 annually.
We use the policy change to estimate the adverse selection
costs and competitive gains of reform. We estimate a substantial
demand response to the pricing reform; a 1 percent increase in the
premiums charged by insurers reduces plan enrollment by 2
percent. Consistent with the theory, the policy change induced
substantial adverse selection. Within three years of the pricing
reform, adverse selection eliminated the market for more gener-

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
PAYING FOR HEALTH INSURANCE 435
ous insurance entirely. We estimate the welfare loss from adverse

selection at about 2 to 4 percent of baseline insurance spending.
On the supply side, increasing plan choice and more stringent
bargaining induced insurers to lower their premiums to Harvard.
These cost savings totaled 5 to 8 percent of baseline health
spending, largely from lower insurer profits. On net, pricing
reform was a benefit for Harvard; the gains are greater than the
losses. For society, however, the efficiency consequences are just
the welfare loss from adverse selection; the reduction in insurance
company profits is only a transfer from health insurance compa-
nies to Harvard.
Because adverse selection is so costly, public and private
insurance programs need to minimize the extent of adverse
selection. A natural way to do so is to vary the voucher amount
with expected spending. In the health insurance context, insurers
would receive more for people who are less healthy than for people
who are healthier, and in the Social Security context, annuity
providers would receive more for those expected to live a long life.
Methods of varying the voucher amount with health status are
termed "risk adjustment." Risk adjustment can be done either on
a prospective basis (using information available at the time of
enrollment), or on a retrospective basis (using ex post spending
data as well). In the latter part of the paper, we discuss how a
risk-adjustment system can be designed.
We begin in Section I with a theoretical discussion of insur-
ance market equilibrium under alternative pricing rules. In
Section II we discuss the experiment we analyze and the data we
employ. Section III considers demand-side responses to the policy
reform. Section IV looks at supply-side changes. Section V dis-
cusses the implications of our findings for the design of social
insurance programs.
I. PRICING RULES AND INSURANCE MARKET EQUILIBRIUM
To demonstrate the issues involved in alternative pricing

rules, consider a situation of individual choice of insurance in a
multiple-option system. For example, the market might be for
health insurance choices in Medicare; the government would
determine how much of a voucher each person is to receive, and
whether the amount varies with the actual insurance policy
chosen. Or, in the example we consider in this paper, the market
could be for health insurance in a large firm. The firm determines

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
the contribution for each worker and how that contribution

depends on the plan chosen.
We focus primarily on the efficiency implications of alterna-
tive pricing rules. There are a number of redistributive issues
involved in pricing-from the young to the old, the sick to the
healthy, the government to the firm and its employees (through
the tax code). We have less to say about redistribution than about
efficiency, however, so we stick to efficiency issues.
Employer contributions for insurance plan k can be expressed
generally as
(1) Ek= Ak + [Pk,
where Ak is a fixed amount (potentially varying by plan), Pk is the

actual plan premium, and [ is the marginal subsidy to insurance.
Traditionally, many large firms paid a percentage, generally 80
percent, of plan premiums (Ak = 0, [ = .8). An increasingly com-
mon alternative policy is for firms to make a fixed, equal contribu-
tion to all plans (Ak= A*; [ = 0).
The rationale for the equal contribution rule is twofold. If
employees pay more of the marginal cost of insurance, their
insurance choices should be more efficient, as those who do not
value generous insurance at its full marginal cost switch to
cheaper plans. In addition, if premium increases by insurers are
translated dollar-for-dollar into increased prices faced by employ-
ees, demand responsiveness will increase, and this may induce
price reductions. These benefits are tempered by adverse selec-
tion, however. If more generous plans disproportionately attract
the sickest employees, prices for more generous policies will be
higher than benefit differences alone would suggest. Premium
differences due to adverse selection are inefficient. Thus, moving
to an equal contribution rule is not guaranteed to increase welfare
and may decrease it.
Demand-Side Changes
We begin by showing the trade-off between adverse selection

and enrollment efficiency on the demand side. Suppose that the
employees in the firm differ in health status, denoted h. For
simplicity we take h to be expected health spending the individual
would incur if he were enrolled in the most generous plan.
The firm offers two plans: a low-cost plan that restricts choice
of providers and provider income (such as a Health Maintenance
Organization [HMO]) and a high-cost plan with fewer restrictions

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
on choice and more generous provider reimbursement (such as a

Fee-for-Service [FFS] plan or a Preferred Provider Organization
[PPO]).' We assume that the more generous plan is a PPO, to
match our later empirical work. For the moment we assume that
everyone will choose insurance; the only question is which policy
they will choose. We return to this issue in our empirical analysis.
Plans must accept anyone who wants to enroll at the same price;
in the firm we analyze (as in most firms), plans must agree to this
condition if they want to be on the menu of policy options.
Demand for the PPO varies with health status. We express
the value of provider choice as g(h). This value is increasing in h if
the less healthy value generosity in insurance more than the
healthier. This is a plausible assumption, and we shall invoke it.
Figure I shows g(h). For any relative out-of-pocket cost of the PPO
(termed the "out-of-pocket premium") Poop, there will be a mar-
ginal person h', where g(h') = Poop. All people who are healthier
than h' (h < h') will enroll in the HMO, while people who are less
healthy than h' (h < h') will enroll in the PPO.2 We show one
particular allocation of people in Figure I.
We denote the mean healthiness of the HMO enrollees (the
conditional mean of h for h < h') as hL(h') and mean healthiness
of PPO enrollees as hu(h'). Premiums will be proportional to
average spending: Pppo = hu(h') and PHMO = axhL(h'), where (X
reflects the cost savings in the HMO net of any higher profits or
administrative expense. As discussed below, ax is a number on the
order of .9. From equation (1) the additional cost to the employee
of enrolling in the PPO is POOP = (1 - A) . (Pppo - PHMO).
The premium difference between the PPO and the HMO,
Pppo - PHMO, depends on h', but the sign of this relat
indeterminate. As the marginal person moves from the PPO to the
HMO, the average cost of the remaining PPO enrollees rises, but
so does the average cost of HMO enrollees. For many distributions
1. An HMO provides medical care for a prepaid fee. The most restrictive form
of HMO is the group/staff model; the physicians in this plan work exclusively for
the HMO and are typically paid a salary or capitated amount (fixed amount per
patient per year). A less restrictive form of HMO is an Independent Practice
Association [IPA]. This plan contracts with specific providers, but the providers do
not work exclusively for the HMO. Providers are generally paid a capitated rate. A
PPO is a plan where providers agree to discount fees in exchange for inclusion in
the PPO "network." The generosity of managed care plans is largely determined by
the size of the provider network, the method of physician payment, and the cost to
the patient for using services outside of the network.
2. The strict delineation by health status is a result of the fact that g depends
only on health status, not on other factors such as risk aversion. Adding such
elements would be straightforward but would not yield any additional insights.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
g(h)
E" '(h)
HMO --------------------------> ho <------------------------------------------------- PPO Health (h)
FIGURE I
Insurance Market Equilibrium
it will be the case that as more people move from the PPO to the
HMO, the relative premium for the PPO will rise, because the
very sick will increasingly be concentrated in the PPO.3 We
assume that the relative PPO premium is increasing as PPO
enrollment falls. Figure I shows the relation between HMO
enrollment and the out-of-pocket cost of the PPO, denoted PP.
The equilibrium in the insurance market is at point E, where
employees have optimally selected plans and premiums are
consistent with those enrollments. This equilibrium may or may
not be stable. As the marginal person moves from the PPO to the
3. This will be true, for example, for most of the distribution if health
spending is distributed lognormally. Empirical studies generally find that medical
spending is distributed lognormally [Newhouse, Beeuwkes, and Warshawsky
1993-1, although we are more interested in the distribution of expected medical
spending than of actual medical spending. There are no empirical estimates of the
distribution of expected medical spending.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
HMO, the PPO premium will increase. If this premium increase

results in further people leaving the PPO for the HMO, there will
be a cycle of increased PPO premiums and reduced PPO enroll-
ment. This is the unstable equilibrium. In order for the equilib-
rium to be stable, it must be the case that the increase in PPO
premiums as the healthiest person leaves the PPO is smaller than
the increase in the PPO reservation value for the new marginal
enrollee, or that the g(h) curve is more steeply sloped than the PP
curve. This is the situation we show in Figure I.
Now suppose that the firm reduces P. For any level of PP
enrollment the cost to the employee of enrolling in the PPO
increases. Thus, the PP line rotates to PP'. The new equilibrium is
at point E', with lower PPO enrollment and a greater price
differential. The dashed line in Figure I shows the factors
responsible for this change. Initially, out-of-pocket premiums
increase because of the pricing reform, and this induces some
people to leave the PPO, so that the equilibrium would be at D.
The movement from E to D is the short-run response to the
reform. Because the remaining PPO pool is less healthy than the
original pool, the PPO premium must rise farther, and more
people will leave the plan. This occurs from D to E'. The
movement between D and E' is the effect of increased adverse
selection.4
Because of adverse selection, even small changes in pricing
rules can have large effects on market equilibrium. Indeed, there
need not be an equilibrium with positive PPO enrollment. If the
value of choice were g(h), for example, the new equilibrium would
have everyone in the HMO (E"). The disappearance of the PPO in
equilibrium E" is sometimes termed an adverse selection "death
spiral."
Adverse selection results in a welfare loss. The allocation is
efficient when people enroll in the HMO only if the resource
savings of their own enrollment decision is greater than the value
they place on the provider choice offered by the PPO. The person
optimally indifferent between the two plans is given by h*, where
(1 - ot)h- = g(h*). This implies an efficient employee price for the
PPO of Ptop = (1 - oL)h*. Under an equal contribution rule the
4. The exact transition from the old to the new equilibrium depends on
insurer and individual behavior. If insurers know demand responses well and
there are no plan-switching costs, the equilibrium will jump from E to E'. If
insurers base premiums on enrollment in the prior year and employees base their
enrollment decisions on current year prices, the equilibrium path will be the
dashed line.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
o g(h)
0)~~~~~~~~
~~~~~AX
ma..
h* Health (h)
FIGURE II
Welfare Implications of Pricing Reform
actual difference in plan premiums will be
(2) PPPo - PHMO = (1 - c)hL(h') + [hu(h') - hL(h')].

The first term in equation (2) is the efficiency savings in the
HMO.5 The second term is adverse selection. Only in particular
circumstances-for example, if people are randomly distributed
across plans-will the equilibrium premium difference be the
efficient premium difference. To the extent that adverse selection
increases the price difference above the efficient level, too few
people will be in the PPO relative to the efficient level.
Figure II shows the welfare consequences explicitly. We
5. Note that this is the resource savings for the average person already in the
HMO, not for the marginal person in-between the PPO and the HMO. Since people
already enrolled in the HMO will be healthier than people at the margin ofjoining
the HMO, this price will be below the optimal price difference that the marginal
employee should face. The efficient price will need to exceed this first term by an
amount that adjusts for the difference between the average and marginal HMO
enrollee.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
assume that initially the employer is subsidizing the PPO more

than is optimal (P < (1 - u)h*), so that there is a deadweight loss
of area A. Moving to an equal contribution rule eliminates area A
but creates deadweight loss of area B, if the new price is greater
than the efficient level (P' > (1 - oL)h*). In addition, the equal
contribution rule involves redistribution of Area C-premium
payments by those who remain in the PPO above the socially
efficient level, which implicitly are redistributed to HMO enroll-
ees (in the form of lower premiums than is optimal).6 Whether
reform on net increases or reduces welfare depends on the relative
size of areas A and B.
Supply-Side Changes
There may also be supply-side effects of changing pricing

rules. Moving to an equal contribution rule will change the
sensitivity of employees to premium increases. If Harvard cannot
credibly threaten to remove an insurer from the set of choices,
then the only constraint on insurer pricing is the elasticity of
demand for the insurance policy. The response of demand to price
increases will be greater under an equal contribution rule than
under a rule where premiums are subsidized at the margin.
Suppose that insurer k increases the premium by lOOx percent, or
XPk dollars. If the employer is subsidizing the premium at rate P a
the margin, the increase in costs to the employee is (1 - r)xPk. As
a share of initial employee payments (EEk), this is (1- -)
xPk/EEk. Generally, employee payments are about 25 percent of
plan premiums, so the increased cost to employees is roughly 400
(1 - A) x percent. The reduction in plan enrollment will therefore
be approximately
(3) Aq/q = 400 (1 - P)x-q

percent, where -i is the elasticity of demand for insurance-the
demand response to an increase in the out-of-pocket cost paid by
the employee. Thus, as employers move toward a fixed payment
for insurance (as ( falls), the demand response to a given
6. We term this a redistribution, but it may also involve efficiency issues. Ex

ante, people may want to purchase insurance against the risk of being sick and
valuing the PPO highly. When the PPO premium is too high, that insurance is
denied them ex post. Thus, all payments above the average cost might also be
termed an efficiency loss. The amount of the redistribution will naturally depend
on whether wages adjust to the policy change; throughout our analysis we assume
that wages are unaffected by the pricing reform.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
premium change increases. Employers will thus capture more of

the surplus from insurance.
In practice, premiums in most firms, including the firm we
analyze, are determined by a negotiation process between firms
and insurers. Insurers submit bids for their insurance policy, but
the proposed rates are then negotiated. Either party to the
negotiations can cancel the plan if the premium is unacceptable.
Moving to an equal contribution rule might have no effect on
premiums in such a circumstance, if bargaining undoes the effect
of greater demand response on profits. But the negotiation may
change the division of surplus in other ways, by changing the
profitability of selling insurance or by convincing insurers of the
seriousness of the firm about reducing costs.
Moving to an equal contribution rule could also change the
quality of insurance plans and thus their price. As employees are
made to pay more for marginal premium increases, the demand
for additional quality improvements will fall. This could lower
quality, and thus premiums. In addition, insurers may change the
quality of their insurance to attract better risks. Certain types of
benefits (oncology services, for example) will attract sicker enroll-
ees; other types of benefits (health club memberships and well-
baby care, for example) are more likely to attract healthier
enrollees. If insurers are concerned about adverse selection under
an equal contribution rule, they may tilt their offerings away from
benefits for the sick and toward benefits that healthy individuals
value.
Pricing reforms could induce quality change for all an insur-
er's enrollees or differentially for the firm moving to an equal
contribution rule. Insurers would like to set quality on a group-by-
group basis, but technology makes this difficult. All enrollees in an
insurance plan generally receive services from the same provid-
ers, and provider compensation is typically not varied by enroll-
ment group. Thus, changes in the quality of providers will almost
certainly be common to all enrollees. But cost sharing can differ
across groups, so cost sharing changes can be targeted to firms
changing their benefits rules.
The net effect of these supply-side factors is likely to lower
premiums, but the magnitude of the premium reduction is an
empirical question.
If premiums are affected by changes in pricing rules, the
efficiency implications of these changes depend on how they are
realized. If premium reductions come from insurer profits, the

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
reduction in premiums is just a transfer of resources from

insurers to employers (and presumably on to employees). There
will be no efficiency consequences of the profit reduction. The
same would be true if premium reductions are financed by lower
insurer payments to health care providers; this would just be a
transfer from health care providers to insurance purchasers.7 If
premium reductions are financed by reduced services provided,
however, there will be social gains or losses equal to the difference
between the value of medical care that is no longer provided and
the cost of that care. The cost of care eliminated may be equal to
its value to consumers, but when insurers change their policies,
there is no presumption that this is the case.
Efficient Pricing
While the equal contribution rule may or may not increase

efficiency relative to particular alternative payment rules, the
equal contribution rule will not be the most efficient rule, so long
as there is adverse selection. If preferences for generous coverage
are correlated with health status, the relative price of generous
insurance will be too high under the equal contribution rule.
There is a pricing scheme that will offset the effects of adverse
selection on premiums, producing a more efficient outcome than
an equal contribution rule. Suppose that the employer contributes
different lump-sum amounts to each plan, taking into account
differences in enrollee health: Appo = hu(h*) and AHMO = hL(h*)
After factoring out this difference, the remaining price variation
between the PPO and the HMO will be the true resource
difference between the plans, Poop = (1 - &)hL(h*). This is the
appropriate price for employees to face.8 In the optimal payment
system, therefore, employer contributions should be fixed, but not
equal across plans. This approach is frequently termed a "risk-
adjusted" pricing rule. We return to how such a system might be
implemented in Section V.
7. Both of these effects assume that the reduction in premiums does not
induce more people to purchase insurance who would not otherwise have done so
or induce providers to leave the medical sector. In the short run this is reasonable.
In the long run the insurance purchase decision and the supply of resources to the
medical sector will be price elastic.
8. As noted in footnote 5, this is subject to the proviso that the person at the
margin between the HMO and the PPO will be sicker than the average person
already enrolled in the HMO and thus should realize a greater cost saving from
enrolling in the HMO. AHMO would thus have to be greater than (1 - OL)hL(h*) to
provide the appropriate incentives for the marginal HMO enrollee. The intuition
for the optimal price is the same, however.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
TABLE I
CHANGES IN EMPLOYEE PAYMENTS RESULTING FROM PRICING REFORM, 1995
Employee payment
Share of
Total Old New enrollment,
Plan premium policy policy Change 1994
Individual
PPO HealthFlex Blue $2773 $ 555 $1152 $597 16%
IPA BayState 2127 489 576 87 5
Pilgrim 2123 382 564 182 2
Tufts 2119 381 564 183 8
G/S HCHP 1945 253 384 131 25
HUGHP 1957 235 396 161 44
HMO average $1980 $ 277 $ 421 $144 84%
Family
PPO HealthFlex Blue $6238 $1248 $2208 $960 22%
IPA BayState 5772 1154 1572 418 9
Pilgrim 5734 1032 1488 456 3
Tufts 5721 1030 1488 458 10
G/S HCHP 5252 683 1056 373 28
HUGHP 5264 632 1068 436 29
HMO average $5395 $ 776 $1191 $415 78%
G/S is a group/staff model HMO. HCHP is Harvard Community Health Plan. HUGHP is Harvard
University Group Health Program, the HMO run by the University. In 1994 there were 3627 individual
policies and 3387 family policies among full-time employees.
Out-of-pocket premiums are for an individual with salary between $45,000 and $70,000.
II. THE HARVARD UNIVERSITY EXPERIENCE
We examine the response to pricing reform empirically using

data on health insurance choices by employees of Harvard Univer-
sity. The Harvard experience is valuable because of a recent
pricing change the University implemented (see Reber [1996] for
more discussion).
Traditionally, Harvard offered its employees a range of insur-
ance choices, shown in Table I. The most generous plan was a Blue
Cross/Blue Shield Preferred Provider Organization (PPO). In
addition, the University offered five HMOs: three Independent
Practice Organizations [IPAs]; and two group/staff [G/S] model
HMOs.9 The group/staff model HMOs are the most restrictive, and
9. There were two other plans offered to employees: a plan available only in
central Massachusetts and an out-of-state Blue Cross/Blue Shield policy. These
were not designed for employees in the Boston area, however, so we omit analysis
of them.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
the IPAs are between the group/staff HMOs and the PPO. One of
the G/S model HMOs is run by Harvard's internal health service,
Harvard University Group Health Program (HUGHP); the remain-
der of the HMOs and the PPO are commercial products provided
by insurers not affiliated with the University.
Harvard historically subsidized the PPO quite generously at
the margin. As Table I shows, while the PPO premium for an
individual was $800 per year more than the premium for an
average HMO, under the old policy the individual paid only $280
more for the PPO than for an HMO. For a family, employees paid
only $500 more for the PPO than for an HMO, even though the
premium was close to $1000 higher. With this set of subsidies,
enrollment in the PPO was stable at about 20 percent of total
enrollees, as shown in Table II.
In the mid-1990s Harvard faced a deficit in the employee
benefits budget. The rising cost of health insurance in particular
was identified as a leading cause.10 The University began looking
for ways to reduce health insurance costs, and beginning in 1995,
the University implemented a health insurance pricing reform.
Under the new policy, Harvard contributes an equal amount to
each plan, regardless of which plan an employee chooses.1"
Harvard's contribution is 85 percent, 80 percent, and 75 percent of
the least expensive commercial policy (excluding HUGHP) for
employees earning below $45,000, between $45,000 and $70,000,
and more than $70,000, respectively.
Because the budget deficit drew substantial attention and the
reforms were significant, the policy change was widely discussed.
Harvard's policy change is thus valuable for learning about the
implications of pricing reform.
The new policy applied to all of the roughly 10,000 full-time
Harvard employees. Two of the unions, representing about 3000
full-time employees, did not agree to the new policy until 1996.
This creates a natural "treatment/control" situation. We divide
employees into those who experienced the policy change in 1995
(termed the "1995 Treatment Group") and those who experienced
the policy change in 1996 (termed the "1996 Treatment Group").
10. Economically, it may not be meaningful to speak of a deficit in one part of a

larger budget, but at Harvard (as at many other companies), accounts for different
services are analyzed separately.
11. The contribution is different for individuals and families. Part-time
workers also receive different contributions. We focus our analysis on insurance
choices for full-time workers.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
TABLE II
TRENDS IN REAL PREMIUMS AND ENROLLMENTS
Year
Measure 1992 1993 1994 1995 1996
Individual
Out-of-pocket cost of PPO $ 290 $ 279 $ 361 - -
1995 treatment group 290 279 361 $ 731 $1414
1996 treatment group 290 279 361 346 1414
Share of enrollees in PPO* 20% 20% 20% - -
1995 treatment group - - 18 14% 9%
1996 treatment group - - 13 12 5
Real premium
PPO $2854 $2794 $2828 $2773 $3228
HMOs 2066 2239 2240 1980 1910
Family
Out-of-pocket cost of PPO $ 439 $ 453 $ 519 - -
1995 treatment group 439 453 519 $1017 $2167
1996 treatment group 439 453 519 522 2167
Share of enrollees in PPO* 20% 20% 20% - -
1995 treatment group - - 25 21% 14%
1996 treatment group - - 11 11 4
Real premium
PPO $6430 $6267 $6395 $6238 $7251
HMOs 5860 6274 6227 5395 5281
Premiums are in 1995 dollars. Out-of-pocket premiums under the new policy are for employees earning
between $45,000 and $70,000.
*Summary data for 1992-1994 include individuals and families together, and are for both part-time and
full-time workers. The divisions into 1995 and 1996 treatment groups include only full-time workers and are
reported separately for individuals and families.
In analyzing the 1995 data, we use the 1996 Treatment Group as a

control. We reverse this situation in 1996.
The policy change had a dramatic effect on the relative price
of different health plans, as shown in Table I. For individuals the
cost of the PPO rose from $555 to $1152, a $597 increase, while the
cost of an average HMO rose by only $144, for a net increase in
PPO costs of $453. For families the net increase in PPO costs was
$545.
III. ENROLLMENT RESPONSES TO POLICY REFORM
The first question we examine is the enrollment response to

this reform. Table II summarizes enrollment and cost information

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
over time. Consistent with a nonzero demand response, PPO

enrollment fell by four percentage points for the 1995 Treatment
Group when the policy was implemented. PPO enrollment among
the 1996 Treatment Group, in contrast, was essentially un-
changed. The reverse situation is true in 1996.
To formalize this relationship, we express latent demand for
the PPO (denoted PPQ*) as a function of the effective out-of-
pocket price of the PPO (P' op and other covariates (X):
(4) PPO t = y In (P o~pit) + XitF

+ 81 1995 Treatment Groupi + 82 Yeart + Eits
where i denotes individuals and t denotes time. In principle, we

could estimate a more detailed model of plan choice, separating
out the type of HMO enrollment as well as PPO enrollment (as in
Feldman et al. [1989]).12 Because the change in out-of-pocket
premiums was essentially the same for the two largest HMOs,
however, our primary source of identification is the decision to
enroll in the PPO or an HMO. We thus focus on this decision.
Both before and after the policy change, employee contribu-
tions to health insurance at Harvard were made on a pretax basis,
since Harvard has a Section 125 ("cafeteria") plan that allows
employee contributions for certain benefits to be excluded from
taxation. Thus, the effective price to the employee of the PPO is
only Pe0op = (1 - T)Poop, where X is the employee's combined
marginal federal, state, and Social Security tax rate and Poop is
the cost of the PPO relative to the cost of an average HMO.13 The
marginal tax rates are imputed based on the worker's salary and
family status.14
12. Feldman et al. [1989] found that the effects of price changes of plans
within the same nest had a greater impact on a plan's market share than price
changes outside the nest. At Harvard about 80 percent of employees were enrolled
in an HMO before the pricing change, so HMOs set their premiums to compete for
current HMO market share as well as switchers from the PPO. This price
competition among HMOs is an important source of savings to Harvard.
13. We use a simple average of HMO premiums to avoid potential endogene-
ity issues from weighting by actual enrollment. In practice, the method used for
averaging HMO premiums does not affect our coefficient estimates.
14. We assume that people with a family policy file joint returns, and people
with an individual policy file single returns. We adjust the salary of people with a
family policy by the average ratio of family income to individual income in the CPS
(separately for men and women) to form income for the family. Since we do not
know the particulars of anyone's income outside of their Harvard salary, we cannot
include direct information on this. We also assume the average number of
exemptions and deductions by income.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
TABLE III
LOGISTIC REGRESSION ESTIMATES OF INSURANCE CHOICE
PPO enrollment Drop coverage

Independent Mean
variable (std devn) 1994-1995 1995-1996 1994-1995 1995-1996
ln(P 0p) 3.64 -.430** -.784** 1.618** .794**

(.44) (.058) (.121) (.359) (.367)
Treatment group .67 .008 -.142 - -
(.089) (.112)
Second year - .058* -.031
(.033) (.086)
Individual policy .52 -.059 -.203** -.487** -.474**
(.067) (.084) (.207) (.245)
Age 30-39 .33 .568*" .775** .443 -.337
(.137) (.151) (.379) (.340)
Age 40-49 .25 1.005** 1.259** .139 -.881**
(.140) (.154) (.409) (.404)
Age 50-59 .16 1.340** 1.565** .141 -.835*
(.150) (.163) (.455) (.485)
Age 60-69 .08 1.470** 1.653** -.484 - 1.652**
(.169) (.182) (.603) (.767)
Salary between $45,000 .18 .491** .531** .278 1.054**
and $70,000 (.094) (.100) (.268) (.287)
Salary > $70,000 .15 .995** 1.176** .156 .918**
(.112) (.124) (.376) (.402)
Female .52 .240** .269** - .039 - .168
(.067) (.072) (.199) (.241)
Distance to HMO (miles) 15.0 .0011 -.0016 -.027 -.003
(61.8) (.0048) (.0052) (.038) (.027)
Distance > = 300 miles .04 .019 1.275 9.796 2.675
(1.427) (1.564) (11.301) (8.048)
Tenure at University 9.5 .023*" .027** -.015 -.004
(9.6) (.004) (.004) (.013) (.017)
Faculty .17 .435** .562** - .060 - .131
(.088) (.093) (.243) (.322)
Hourly .09 -.146 - .090 - .420 - .461
(.127) (.136) (.485) (.547)
Sample size 9,073 16,727 17,741 7,654 8,601
ln(likelihood) -6,694.33 -5,692.25 -608.59 -449.67
Summary statistics are for 1995. Standard errors in the equations for PPO enrollment are adjusted for the
presence of multiple observations per person.
*Statistically significant at the 10 percent level.
**Statistically significant at the 5 percent level.
We include a variety of control variables in the model, the

means of which are reported in the first column of Table III: age,
sex, employee type (faculty, staff, or hourly), job tenure, salary,
single or family plan, and distance from the nearest group/staff

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
model HMO clinic.15 Since we also include year and treatment

group dummy variables, the coefficient y represents the differe
in-differences estimate of the effect of prices on insurance choice.
The most important omitted variable is health status. The theory
predicts that the demand elasticity will be greater for people
whose health status places them at the margin between PPO and
HMO enrollment. Because confidentiality restrictions prevent the
inclusion of health-status data, we estimate an average demand
elasticity over all health states.
The second and third columns of Table III report logit
estimates of equation (4). We estimate the model separately for
1994-1995 and 1995-1996 to allow the elasticity to differ over
time; in each case, the standard errors are adjusted to account for
the fact that we have multiple observations on the same family
unit. Most of the independent variables are consistent with
expectations. Age is a consistent positive and significant predictor
of PPO enrollment, as is being female and job tenure and salary at
the University. Faculty members are also more likely to choose the
PPO. Surprisingly, distance to an HMO does not affect plan choice.
As the first row shows, the relationship between price and
PPO enrollment is negative and significant in both years. The
demand elasticity implied by these estimates (the numerical
elasticity from a 1 percent increase in price) is -0.3 in the first
year of reform and -0.6 in the second year. Recall from Section I
that the response to an increase in the total premium under an
equal contribution rule is roughly four times the demand elastic-
ity. Our estimate of the enrollment response to an increase in total
premiums is therefore about -2. Interestingly, the coefficient on
the Second year variable is about zero or slightly positive. The
price change thus explains all of the change in enrollment over
time.
The estimate of -2 is higher than traditional estimates of
demand responsiveness. Estimates from the 1980s using cross-
firm variation in the generosity of benefits typically suggested
demand elasticities with respect to total premiums of about -.2 to
-.5 [Taylor and Wilensky 1983; Holmer 1984; Farley and Wilen-
sky 1984], although some studies [Phelps 1986; Welch 1986;
Feldman et al. 1989] generated substantially larger elasticities
15. We match employee zip codes to the zip code of the clinics and find the
minimum distance between the employee's home and a clinic. We truncate the
distance at 300 miles and include a dummy variable for people with distances
beyond that amount.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
between -2 and -8. More recently, data from individual choices

within multiple option systems have suggested elasticities with
respect to out-of-pocket premiums of about -.5 or larger [Dowd
and Feldman 1994; Royalty and Solomon 1995; Buchmueller and
Feldstein 1996; Hill and Wolfe 1997].16
To examine the sensitivity of our results to functional form,
we estimated a linear probability model for PPO enrollment. The
coefficient for the 1994-1995 period is -.063 (.008), for an
elasticity of -.4. The coefficient for the 1995-1996 period is
somewhat lower, but still statistically significant (a coefficient of
-.027 (.009), implying an elasticity of -.2). In this framework we
can also instrument for the price change using treatment group/
year interactions, to isolate the component of price changes due
only to the differing implementation dates.17 The price elasticity
from this model is -.2 and significantly different from zero.
Finally, if we estimate a fixed effect linear probability model, the
price elasticity is again -.2, which is also statistically different
from zero.18
Rather than changing their insurance policy, individuals may
decide to stop receiving coverage from Harvard entirely as the
premium increases, if they have insurance through a spouse or
choose to be uninsured, for example. To test this, we estimate logit
models for whether people who had insurance in one year dropped
coverage in the next year. We model this decision as a function of
the price change for the plan in which the individual was enrolled
and the demographic characteristics noted above.19
The last two columns of Table III report these estimates.
Out-of-pocket premiums significantly affect the probability of
dropping coverage. The elasticity of coverage with respect to the
out-of-pocket premium is about - 1. While this elasticity is large,
the baseline rate of dropping coverage is low-only 1 to 2 percent.
16. Dowd and Feldman [1994] estimate demand elasticities nearly identical
to Feldman et al. [1989]. Other studies find very large changes in HMO
enrollment-up to 80 percent-in response to price increases of $5 or $10 per
month.
17. The effective PPO price varies within Treatment groups because of
differing marginal tax rates at different salaries and because there was a small
phase-in for lower-salaried employees in the 1995 Treatment Group.
18. In the 1995-1996 period the elasticities in the IV and fixed effects models
are about -.1. The fixed effects elasticity is significantly different from zero, but
the IV elasticity is not. There is more of a difference across models here, but the
results still support the negative demand response.
19. Since our dependent variable is expressed as a change in coverage, we
omit the controls for the Treatment group and the Second year dummy variable.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
TABLE IV
CHARACTERISTICS OF PLAN ENROLLMENT CHANGES
1994-1995 sample 1995-1996 sample

First year
enrollment HMO PPO HMO PPO
Second year
enrollment HMO PPO HMO PPO H]MO PPO HMO PPO
Share of enrollees 99% 1% 15% 85% 100% 0% 39% 61%

Average age 41** 46** 46** 50** 41 *** 46** 51**
Percent <40 50% 26% 31% 21% 50% * 30% 15%
Percent 40-60 44 68 56 61 45 * 60 66
Percent >60 6 6 13 18 5 * 10 19
Index of spending 0.96 1.09 1.09 1.16 0.97 * 1.09 1.20
Average spending - - - - - - $1893 $2648
Individual and family plans are grouped together. Average spending in the last row is adjusted for
individual/family policies.
**Difference between age of people switching and remaining in plan is statistically significant at the 5
percent level.
***Too few people for reliable estimates.
Thus, even large changes in premiums would have only a small

effect on coverage at Harvard.
Adverse Selection and Enrollment Dynamics
In addition to knowing the average demand elasticity, we

want to know whether the people who disenrolled from the PPO
were disproportionately the healthier enrollees. Table IV reports
evidence on this question. The second row of the table shows the
average age of people who switched plans at the end of each year
relative to the people who stayed in the same plan both years. Age
is a natural indicator of selection since older people use more
medical care than younger people.20
There is clear evidence of age-related differences in plan
changers. In the 1994-1995 enrollment period, the average age of
people who moved from the PPO to an HMO is four years below
the average age of those who remained in the PPO both years and
five years above those who were enrolled in the HMO the entire
time. Both of these differences are statistically significant. The
penultimate row of the table shows a spending index for each
group, weighting people by average age-specific medical spend-
20. Since age is observable, this is not a case of asymmetric information

leading to adverse selection. Because Harvard does not vary its contribution on the
basis of age, however, it is as if the information were asymmetric.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
ing.21 The people who left the PPO were predicted to spend 5
percent less than those who remained in the PPO the following
year and about 15 percent more than the average HMO enrollee.
Thus, the data match well the predictions of the model: younger
(healthier) people are disproportionately in the HMO, older
(sicker) people are in the PPO, and middle-aged people are at the
margin between the two.
Indeed, as a result of this adverse selection, the PPO lost
money in 1995. Recognizing the loss, Blue Cross/Blue Shield
increased the PPO premium in 1996-by about 16 percent in real
terms (see Table II).
With the new, higher premium, the out-of-pocket cost of the
PPO rose markedly. Where families paid $500 more for the PPO in
1994 and $1000 more in 1995, the PPO cost $2000 more than an
HMO in 1996 (see Table II). Not surprisingly, PPO enrollment fell
again in 1996 to only 8 percent, compared with about 20 percent in
1994 (see Table II).
As Table IV shows, disenrollment was again nonrandom.
People leaving the PPO after 1995 were younger than those
remaining in the plan and older than those in an HMO both years.
Indeed, Blue Cross/Blue Shield, the sponsor of the PPO, was
sufficiently concerned about falling enrollment that it compiled
data on average spending in 1995 for those who left the plan at the
end of the year and those who stayed. As the last row of Table IV
shows,22 those who left the PPO at the end of 1995 spent 20
percent below average, while those who remained in the plan in
1996 spent 11 percent above average. Overall, adverse selection is
greater than selection based on age alone, but age explains a large
part of adverse selection.
The large price increase notwithstanding, adverse selection
led to significant financial losses for the PPO again in 1996. By the
21. Data on medical spending are from the 1987 National Medical Expendi-
ture Survey (NMES). We sort the NMES into "health insurance units"-the group
for which health insurance is typically sold and form average spending by age of
household head. These are then matched to the Harvard data. To form the
spending index, we scale spending for individuals and families separately to have a
mean of 1.0.
22. The spending estimates are adjusted for the mix of individuals and
families. The same phenomenon occurred in the BayState HMO (also run by Blue
Cross/Blue Shield). In 1995 BayState lowered its premium, to compete in the new
system. Without cost reductions, however, the plan lost money. In response,
BayState's premium increased by 16 percent in 1996, and two-thirds of the 1995
enrollees left the plan. Data similar to those in Table IV show that the people who
left BayState after its price increased used 23 percent fewer services than the
average BayState enrollee that year.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
beginning of the 1997 premium cycle, it was clear to Harvard and

Blue Cross/Blue Shield that the PPO could not be offered at a
reasonable price, and they agreed to discontinue the plan; the
BayState HMO was ended as well. Blue Cross/Blue Shield
replaced the two plans with a new, more tightly managed HMO
[HMO Blue]. In three years, therefore, the adverse selection
"death spiral" was completed at Harvard.
Harvard was not content to offer only HMOs, however, so the
University encouraged the remaining HMOs to offer plans with
point-of-service (POS) options -which provide some reimburse-
ment for out-of-network use. The price of this option was tied to
the price of the in-network only option. Three of the HMOs agreed
to offer a POS option. In effect, then, Harvard has substituted
three slightly less generous plans for one more generous plan.
The Welfare Loss from Adverse Selection
To estimate the welfare consequences of adverse selection, we

need to know the efficient price for employees to pay if they want
to join the PPO. This information cannot be distilled from our
data, but reviews of the literature on HMO savings (the Congres-
sional Budget Office [1992] and Miller and Luft [1994]) suggest
average savings of about 10 percent (a = .9). Since the average
individual premium in 1995 was $2106 and the average family
premium was $5581, this suggests an optimal PPO out-of-pocket
price of about $200 for individuals and about $550 for families.23
As Table I shows, this is near the relative price for the PPO
under the old payment policy ($277 for individuals and $472 for
families). Thus, the price of the PPO was close to the optimal level
prior to the policy reform. If this is correct, then all of the plan
changes induced by the increase in the PPO price reduced
efficiency.
We can estimate the magnitude of the welfare loss explicitly.
The indirect utility function implied by the logit demand model
[Small and Rosen 1981; Trajtenberg 1989; Feldman 1994] is
(5) log (1 + e(-Yln(P)?M'))

(y/P
where we have incorporated the Treatment Group and Second

Year dummy variables into the X vector.
23. This ignores any difference between the cost of the average enrollee in the
group and the cost of the marginal HMO enrollee under the optimal pricing rule.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
Evaluating equation (5) at the 1994 prices (the optimal price)

as well as an infinite price for the PPO (the effective price when it
is not offered) yields a welfare loss from adverse selection of $165
per insured worker.24 Since average premiums per insured worker
were about $4300, the loss is about 4 percent of baseline spending.
This loss is clearly too high, because it does not account for the
introduction of the out-of-network options. These plans attracted
7 percent of enrollees in 1997. As an alternative estimate, we
simulate the welfare loss if the PPO premium increased just
enough to reduce enrollment to 7 percent. The resulting welfare
loss is about $100 per insured worker, or about 2 percent of
baseline spending.
In addition to uncertainty about the value of the out-of-
network option, there is also uncertainty about the functional
form of the utility function. We have assumed that demand is
logistic, but this need not be the case. For example, employees
always have the option of dropping insurance coverage at Har-
vard and purchasing insurance in the nongroup market. If we
assume that no employee values the PPO more than $6000 for a
family ($2400 for an individual)-roughly the average cost of an
insurance policy-the loss from adverse selection would be $143
per person, or 3 percent of baseline spending. Of course, nongroup
insurance is an imperfect substitute for group insurance; it
frequently contains preexisting condition exclusions, and premi-
ums usually vary with individual health status. As a result of
these constraints, it may be that the sickest people at Harvard
value the generous policy even more than we have assumed. We
have no way of knowing the true distribution of demand for the
PPO (particularly at the tails of the distribution), so we stick with
the logit formulation. As a range of potential welfare losses, we
thus assume that adverse selection resulted in a loss of 2 to 4
percent of baseline spending.
IV. SUPPLY SIDE RESPONSES TO PRICING REFORM
Pricing reforms clearly affect the demand side of the market;

this is one rationale for this type of reform. A second rationale is to
24. As a benchmark for this estimate, note that the welfare loss can be
approximated as AW = l2 AP A Aq. Note from Table II that a price increase o
$1000 for individuals and about $2000 for families reduced PPO enrollment by
roughly three-quarters. If we assume that a price increase for the PPO of one-third
more would reduce PPO enrollment to 0, the implied welfare reduction is $200 per
person (.5 x $2,000 x .2). This is close to our 4 percent estimate.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
TABLE V
CHANGES IN REAL FAMILY PREMIUMS, 1990-1996
Effect of
increased
cost
Harvard Composite rate sharing Net change
1990- 1994- 1995- 1990- 1994- 1995- 1994- 1990- 1994- 1995-
Plan 1994 1995 1996 1994 1995 1996 1995 1994 1995 1996
PPO HealthFlex - -2.4% 15.5% - 3.2% -7.1% -1.5% - -4.1% 22.5%

Blue*
IPA BayState 5.8 -12.5 19.7 6.2 -5.3 1.8 -1.5 -0.4 -5.8 17.9
Pilgrim 3.8 -14.1 2.8 5.2 -3.5 -3.1 -1.5 -1.4 -9.1 5.9
Rifts 5.6 -12.8 -3.6 4.5 -7.6 -3.1 -1.5 1.1 -3.7 -0.5
G/S HCHP 3.9 -8.6 -3.3 4.9 -2.4 -2.1 -1.5 -1.0 -4.7 -1.1
HUGHP 5.2 -20.7 -8.7 - - - -3.0 - - -
Difference column is the change for Harvard net of the change for the composite rate and the effect of
increased cost sharing. Composite premiums are for large employers (generally 26+ or 51+ employees). The
premiums are a family policy for the insurer's "standard" plan with outpatient prescription drug coverage.
*Plan begins in 1992. The Boston area premium is for HMO Blue, an HMO that is similar to the
HealthFlex Blue PPO.
lower health insurance premiums. In this section we evaluate the

response of premiums to the pricing reform.
The first three columns of Table V show the change in
premiums at Harvard from 1990-1994, 1994-1995, and 1995-
1996 (we do not report premium changes in 1996-1997 because
the set of plans offered is so different). From 1990 to 1994 real
premiums rose about 4 to 5 percent per year for all plans. With the
implementation of the equal contribution rule in 1995, real
premiums fell by 2 to 21 percent. In 1996, most of the HMO
premiums continued to decline but by a smaller amount, and two
of the plans (PPO and the BayState HMO) had large premium
increases. Thus, the evidence suggests a substantial premium
reduction in 1995, with continued, less substantial declines in
subsequent years.
We want to decompose these premium changes into the three
components noted in Section I: increased demand responsiveness
from increased marginal prices to enrollees; effects from negotia-
tions between Harvard and its insurers; and changes in the
quality of insurance. Changes in premiums resulting from in-
creased demand responsiveness are difficult to separate from
changes in premiums resulting from more stringent negotiations,
since the two occurred at the same time. We thus consider these
two factors together and focus on the difference between quality
and nonquality factors.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
We adjust for quality in two parts. To account for quality

changes that are common throughout the insurers' policies,25 we
compare the growth of premiums at Harvard with the growth of
"composite" premiums charged by the insurers serving Harvard. A
composite premium is not a premium charged to any particular
group; rather, it is the estimate of expected medical costs for a
typical company buying insurance. When a company first pur-
chases insurance, it is generally charged the composite rate
adjusted for its industry and demographic mix. Over time, the
actual premium charged to the company is a blend of the
composite rate and the company's realized experience. Thus, the
composite premium is like an "average" premium, although it is
not the average for any specific group. We gathered composite
premiums for the specific insurers in our sample for policies that
are as close as possible to the policy offered Harvard employees: a
"standard plan" for a large employer in Eastern Massachusetts,
including coverage for outpatient pharmaceuticals.26
The middle columns of Table V show changes in the composite
premiums. Premium growth for the composite plans was about
the same as the premium growth at Harvard in the 1990-1994
and 1995-1996 periods. But composite rate premium reductions
were much smaller in 1994-1995 than they were at Harvard.
Once we have adjusted for changes in composite premiums,
the only residual quality change is changes in cost sharing for
Harvard employees. Harvard did alter its cost sharing when it
moved to the equal contribution rule. Prior to the change, most
policies had a $5 copay per visit, and HUGHP had no copay. After
the change, all policies charged $10 per visit. The consensus
among a number of insurance actuaries we spoke to is that each
$5 increase in copayment reduces premiums by about 1.5 percent.
The next column of Table V shows the adjustment we use for
changes in copayments.
The final columns of Table V show the net change in premi-
ums at Harvard-the Harvard change less the change in the
composite rate and less the change due to increased cost sharing.
25. These may be true changes in quality (for example, contracting with fewer
well-trained physicians) or changes in factor prices in the Boston area that
translate into lower premiums.
26. Massachusetts requires HMOs to file composite premiums with the State
Insurance Department, from whom we acquired the data. The State does not
require PPO rates to be filed; we thus cannot compare HealthFlex Blue with the
comparable product in the Boston area. Instead, we compare HealthFlex Blue to
HMO Blue, Blue Cross/Blue Shield's HMO.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
In the 1990-1994 period, the net change in premiums was small.

Premium growth at Harvard was within 1.5 percent of the area
average for all plans. In the 1994-1995 period, premiums fell at
Harvard by 5 to 10 percent, depending on the plan. In the
1995-1996 period, the situation is mixed, with some plans increas-
ing premiums and some reducing them.
To analyze the aggregate impact of these changes, we com-
pare the growth of actual spending per insured worker at Harvard
with the growth of simulated spending if premiums at Harvard
had increased at the same rate as composite premiums in the
Boston area and enrollment had not changed over time. In
forming the growth of actual spending, we need to make one
adjustment. Cost growth at Harvard fell both because premiums
fell and because more people moved to cheaper plans. The second
of these effects is not a resource savings to Harvard; it is just the
purchase of fewer goods. We thus subtract from actual spending
growth the part that results solely from employees moving into
cheaper plans; the residual is the pure savings to Harvard.27
Figure III shows the growth of actual spending per insured
worker and simulated spending per insured worker. Between
1990 and 1994 the two grew at roughly the same rate; the
difference is on the order of one percentage point or less. In
1994-1995 costs fell by 8 percent more at Harvard than in the
simulation. In the subsequent two years actual cost growth was
again very close to simulated cost growth. Thus, Figure III
suggests a one-time 8 percent reduction in the level of spending
relative to the baseline as a result of the policy change.
Since there is no composite premium to compare with Har-
vard's internal HMO, one might be worried that the premium
growth for that plan includes some quality change that we could
not account for. If we assume that the quality-adjusted premium
change at HUGHP was the same as at HCHP (the other group/
staff model HMO), the overall savings to Harvard from reduced
premiums is 5 percent in 1994-1995, again with no reversal of
this trend in the next two years. We therefore take as our range of
savings estimates a one-time reduction in premiums of about 5 to
8 percent, with premiums continuing at this lower level after-
wards.
27. To make this calculation, we assume that HMOs save 10 percent of

average PPO costs, and that HMOs with a point-of-service option save 5 percent.
Using the average premium to proxy for the cost savings overstates the resource
savings, because it assumes no adverse selection. Thus, we will understate the
pure price savings to Harvard.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
15
10
20
-5
-10
I~~~~~~~~~~~~~~ I
-15 l l l
1991 1992 1993 1994 1995 1996 1997
Year
| Actual 0 Simulated
FIGURE III
Actual and Simulated Cost Growth at Harvard, 1990-1997
Note: Actual cost growth is net of changes in cost sharing and savings from
more employees enrolling in cheaper plans.
This finding is consistent with conversations we have had

with executives from Harvard and the insurance companies who
were involved in the negotiations. Harvard told the insurers about
the policy change before it occurred, and bargained strenuously
for lower rates in the first year. Several insurers stressed that
being the lowest cost plan, or near the lowest cost plan, became
quite important. Some of the insurers, for example, indicated that
in response to the change they decided to "buy" market share at
Harvard by submitting lower bids. The reduced premiums, they
indicated, came at the expense of profits. The negotiations with
Harvard were also quite contentious. In one case, Harvard
bargained down the "experience factor" used to adjust the insur-
er's standard premium for Harvard-specific utilization. In an-
other case, an insurer submitted a proposed premium which
Harvard accepted, but was then told that other plans had cut
premiums even more and was encouraged to reconsider its bid.
The plan resubmitted a lower premium.
Thus, while results based on only six plans and three years of

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
postreform data cannot be definitive, the data on premiums and

our anecdotal evidence suggest that the pricing reform, coupled
with tighter negotiations, resulted in lower premiums to Harvard.
The net savings was about 5 to 8 percent of baseline spending,
largely coming from lower insurer profits. The savings did not
diminish after the first year, but they did not increase either.
Private and Social Gains
The fact that the savings to Harvard (5 to 8 percent of

baseline spending) were greater than the welfare loss from
adverse selection (2 to 4 percent of baseline spending) means that
for Harvard as a whole, the reform was a potential Pareto
improvement. The losers from adverse selection could have been
compensated with enough money to make everyone in the Univer-
sity better off.
For social efficiency, however, we care about the source of the
cost savings. As noted above, the bulk of the savings, at least
initially, came from lower profits. Socially, this reduction in profits
is just a transfer from insurance company shareholders to Har-
vard (and its employees). There are no efficiency implications of
this transfer. Thus, the "first round" impact is that society as a
whole is worse off by the increase in adverse selection.
There is an important dynamic issue, however. As more
employers engage in this type of policy change, it may be that
insurers will reduce utilization rather than just profits. Since
providers ultimately control the provision of medical services,
reducing the amount of care that is provided must involve limiting
provider behavior. But this is costly; it involves monitoring
provider behavior more carefully, installing new information
systems, denying payments, and renegotiating contracts. Indeed,
if only one insurer implemented these measures, providers might
just avoid that plan entirely. Thus, it is possible that if only some
firms push for cost reductions these cost savings measures would
not be implemented, but if many firms implement competitive
reforms, these changes might make sense. Indeed, some reports
have suggested that the overall reduction in premium growth we
noted was a result of this type of competitive pressure [Center for
the Study of Health Systems Change 1997].
Some evidence on this point comes from the experience of
Harvard's internal HMO [HUGHP]. Initially, HUGHP financed
its premium reductions in 1995 by drawing down a "wasting fund"
the plan maintains to keep premiums below the other group/staff

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
model HMO. This is analogous to the profit reduction of the other

insurers. But more recently, HUGHP has reduced costs enough so
that the wasting fund is no longer being depleted. Lower revenues
for HUGHP are now matched by lower costs, rather than addi-
tional losses.28
The social consequences of these cost savings depend on how
they are realized. If the savings come entirely from reduced
payments to providers, with no changes in the allocation of
resources to the medical sector, they would again be transfers, in
this case from medical providers to the rest of society. If the
savings come from reductions in resource utilization, in contrast,
there will be efficiency implications depending on the value of the
services saved relative to their cost. We do not know about the
source of the recent cost savings at HUGHP, nor about long-term
cost savings for the other insurers serving Harvard, so we cannot
provide further evidence on this issue. This is clearly an impor-
tant topic for future research.
V. IMPLICATIONS FOR SOCIAL INSURANCE REFORM
Our results on insurance market dynamics show that adverse

selection is a serious concern. Because of adverse selection, the
most generous policy could not be sustained under an equal
contribution rule. We estimate the welfare loss from adverse
selection at 2 to 4 percent of baseline spending. On the positive
side, we estimate that the policy reform reduced the premiums
Harvard faced by about 5 to 8 percent. These savings are
sufficiently large that the adverse effects of selection could be
offset and everyone at the University could be better off. Socially,
the first round effects of savings to Harvard appear to be a
redistribution from insurer profits more than a net saving in
resource utilization. Thus, society is worse off by the adverse
selection loss.
One might wonder whether our results about adverse selec-
tion are unique to Harvard, or whether they apply to multiple-
choice insurance arrangements more generally. While we do not
have definitive evidence on this question, we suspect that our
findings are quite general. Harvard is not alone in finding
difficulty maintaining more generous plans. Stanford University
28. Because some of HUGHP's savings come from cost reductions, the savings
are not just a transfer from one part of the University to another, as they would be
if they were deficit financed.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
and the State of Minnesota, for example, also moved to equal

contribution rules and were forced, a few years later to discon-
tinue their most generous plans. There is a suggestion in each
case that adverse selection was to blame [Royalty and Solomon
1995; Feldman and Dowd 1993]. Similarly, the Massachusetts
Institute of Technology also moved to an equal contribution rule in
the 1980s and was forced in 1997 to discontinue its indemnity
policy. Our conversations with officials running the MIT plan
suggested that the growing presence of adverse selection was an
important factor in this decision. And even in cases where
generous insurance options are still available, adverse selection is
a concern. Cutler and Zeckhauser [1997] document the difficulties
that the Group Insurance Commission (GIC) of Massachusetts -
the Agency that manages insurance for state and local employees
in Massachusetts-has had in maintaining an indemnity plan.
The GIC subsidizes 85 percent of the additional premium of the
indemnity policy, but even with that subsidy, additional steps
have had to be taken to limit the costs of the indemnity policy and
prevent adverse selection, including switching to a mail-order
drug benefit and negotiating more strenuously with hospitals and
clinical laboratories. While we have not examined all of these case
studies in enough detail to make definite conclusions, all suggest
that adverse selection is an important concern, and that it
impinges greatly on the feasibility of alternative payment policies.
The question raised by our results is how to realize the gains
from competition without the losses from adverse selection. One
potential strategy was discussed in Section I: employer payments
could be "risk-adjusted" to account for the differential selection of
employees across plans [Newhouse 1996; Newhouse, Beeuwkes,
and Chapman 1996; Cutler and Zeckhauser 1997].
The simplest form of risk adjustment is prospective adjust-
ment. Suppose that the employer measures expected spending in
each plan at the beginning of the year, using information on the
demographic mix of enrollees and their past medical conditions.
Then, the employer could vary the payments to plans based on
these health status differences. Plans with less healthy enrollees
would get more than the average, and plans with healthier
enrollees would get less than the average.
For example, Table IV shows that the average enrollee in an
HMO is about 20 percent healthier (based on age) than the
average enrollee in the PPO. If Harvard implemented an age-
based risk-adjustment system, it would reduce its payments to

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
the HMOs by 20 percent of average costs and increase its

payments to the PPO by 20 percent of average costs. Note that
Harvard's contribution would not vary with actual premiums
charged, so that insurers would still face the higher demand
responsiveness.
If adverse selection were entirely on an age basis, this
risk-adjustment system would exactly compensate for nonrandom
selection. Plans would perceive their enrollee mix, net of the
additional payments, as exactly equal to the average in the group
as a whole. Prices would differ across plans only because of
efficiency differences and profits. In such a circumstance, indi-
vidual choices over their insurance options would be efficient.
The difficulty, however, is if there is selection along dimen-
sions not accounted for by the risk-adjustment system. In our
example above, suppose that even within a given age group, sicker
employees opted into the PPO and healthier employees opted into
the HMOs. Then, there would still be adverse selection even after
the risk-adjustment system were implemented, and the problems
that we noted above might still occur. Of course, an actual
risk-adjustment system would be based on medical conditions as
well as demographic status, but the problem still remains. As an
example, we might have a risk-adjustment system that compen-
sates for the share of diabetics in each plan, but if diabetics who
know they are at greatest risk for renal failure opt into the PPO
while those who know they are at lower risk opt into the HMOs,
the risk-adjustment system will again be imperfect.
An alternative to prospective risk adjustment is retrospective
risk adjustment-measuring differences in utilization at the end
of the enrollment period and making payment adjustments at
that point. For example, Harvard could measure how much
insurers spent on care during the year and give additional
amounts to plans with above average spending, taking the money
from plans with below average spending. If the adjustment were
complete, there would be no difference in ex post costs across
plans. Each plan could therefore charge an average rate, again
with variation only because of efficiency differences and profits.
A related concept to retrospective risk adjustment is reinsur-
ance. Suppose that Harvard purchased reinsurance for very high
cost medical cases-for example, the amount of spending an
individual incurs above $25,000 in a given year. The reinsurance
would be financed by reducing payments to all insurers. Then,
very high cost cases would not show up differentially in one plan,

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
and that part of adverse selection would be minimized. Reinsur-

ance is likely to be particularly important in the medical care
context because medical spending is so skewed. The top 10
percent of medical care users account for nearly 70 percent of total
spending [Berk and Monheit 1992]. If high cost users differen-
tially select into more generous plans, as seems likely, reinsur-
ance might offset a great deal of adverse selection.
The weakness of retrospective risk adjustment is relatively
obvious, however: it reduces the incentives for plans to monitor
utilization carefully. This is true whether risk adjustment is based
on all spending or is limited to very high cost cases. Indeed, this is
exactly the same problem as when employers subsidize premiums
at the margin. In both cases, the incentives for insurers to control
costs are blunted.
The choice between prospective and retrospective risk adjust-
ment is analogous to the problem of optimal insurance with moral
hazard [Arrow 1963]. In the optimal insurance context, the goal is
to balance risk-sharing and moral hazard. In the risk-adjustment
context, the goal is to balance incentives for efficiency and adverse
selection. And just as an interior insurance policy (neither full
insurance nor self-insurance) is optimal in the health insurance
design problem, so too a mix of prospective and retrospective risk
adjustment is likely to be optimal in the plan payment situation.
For example, Harvard might implement a risk-adjustment sys-
tem that makes transfers ex ante on the basis of demographics
and health status, and then makes additional payments ex post to
partially compensate plans with very high cost users.
A number of unanswered questions about the most appropri-
ate risk-adjustment methodology remain, and current methods
can account for only a fraction of the predictable variation in
medical costs. Still, even an imperfect risk-adjustment system
could ameliorate the losses from adverse selection, particularly if
it could prevent the elimination of the market for generous
insurance.
Harvard has not implemented a risk-adjustment system,
although the requirement that premiums for the point-of-service
plans be tied to the premiums for the underlying HMO is a limited
form of risk adjustment. Indeed, most employers that have moved
to competitive pricing systems have not implemented risk adjust-
ment. One might wonder why this is the case if risk adjustment is
so valuable. We do not have a complete answer to this puzzle, but
there are several possibilities. One potential reason may be

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
adverse selection. Employers that fear attracting or differentially

retaining high cost employees may want to raise the price of more
expensive insurance above its actuarial level.29 This equilibrium
would not be efficient, since high risk people would not get full
insurance even though they might value additional coverage more
than it costs. A second explanation is the magnitude of the change
involved. Moving to an equal contribution rule and implementing
risk adjustment both involve substantial internal costs; firms may
prefer to do only one at a time. At Harvard, for example, there was
some discussion of implementing a risk-adjustment system after
the equal contribution rule was in place. Risk adjustment may
also change the distribution of employer and employee costs for
health insurance. Implementing risk adjustment generally raises
the cost of less expensive plans while lowering the cost of more
expensive plans. If employers cannot vary their contribution rate
at will (as in many unionized settings), implementing risk adjust-
ment could lead to undesirable changes in employer or employee
payments.
Since risk adjustment has rarely been tried, we do not know
the optimal combination of prospective and retrospective risk
adjustment in designing a choice-based insurance system. Design-
ing appropriate risk-adjustment systems is essential, however, if
social insurance programs are to become more competitive. Imple-
menting a system of Medicare vouchers, we suspect, would raise
the same issues as came up at Harvard. So too would a privatized
Social Security system where individuals are left to purchase
annuities in a private market. Some risk-adjustment mechanism
must be found to minimize adverse selection, or we will forgo a
large part of the potential gains from market reform of social
insurance programs.
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH

COUNCIL OF ECONOMIC ADVISERS
REFERENCES
Aaron, Henry, and Robert Reischauer, "The Medicare Reform Debate: What Is the
Next Step?" Health Affairs, XIV (Winter 1995), 8-30.
Advisory Council on Social Security, Report of the 1994-96 Advisory Council on
Social Security, Volume I: Findings and Recommendations (Washington, DC:
- GPO, 1996).
29. We have no evidence that Harvard considered this factor, but other
employers might.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
Arrow, Kenneth, "Uncertainty and the Welfare Economics of Medical Care,"

American Economic Review, LIII (1963), 941-969.
Berk, Mark L., and Alan C. Monheit, "The Concentration of Health Expenditures:
An Update,"Health Affairs, XI (Winter 1992), 145-149.
Brown, R. S., et al., The Medicare Risk Program for HMOs-Final Summary
Report on Findings from the Evaluation, Final Report under HCFA Contract
Number 500-88-0006 (Princeton, NJ: Mathematica Policy Research, Inc.,
1993).
Buchmueller, Thomas C., and Paul J. Feldstein, "Consumers' Sensitivity to Health
Plan Premiums: Evidence from a Natural Experiment in California," Health
Affairs, XV (Spring 1996), 143-151.
Center for the Study of Health System Change, Health System Change in Boston,
Mass. (Washington, DC: Center for the Study of Health System Change,
1997).
Congressional Budget Office, "The Effects of Managed Care on Use and Costs of
Health Services," mimeo, June 1992.
Cutler, David M., "Restructuring Medicare for the Future," in Robert Reischauer,
ed., Setting National Priorities (Washington, DC: The Brookings Institution,
1996).
Cutler, David M., and Richard J. Zeckhauser, "Adverse Selection in Health
Insurance," NBER Working Paper No. 6107, July 1997.
Dowd, Bryan, and Roger Feldman, "Premium Elasticities of Health Plan Choice,"
Inquiry, XXXI (Winter 1994/5),438-444.
Farley, Pamela, and Gail Wilensky, "Household Wealth and Health Insurance as
Protection against Medical Risks," in Martin David and Timothy Smeeding,
eds., Horizontal Equity, Uncertainty, and Economic Well-Being (Chicago:
University of Chicago Press, 1984), pp. 323-354.
Feldman, Roger, "The Welfare Economics of a Health Plan Merger," Journal of
Regulatory Economics, VI (1994), 67-86.
Feldman, Roger, and Bryan Dowd, "The Effectiveness of Managed Competition:
Results from a Natural Experiment," Institute for Health Services Research,
University of Minnesota, April 6, 1993.
Feldman, Roger, Michael Finch, Bryan Dowd, and Steven Cassou, "The Demand
for Employment-Based Health Insurance Plans," Journal of Human Re-
sources, XXIV (Winter 1989), 115-142.
Feldstein, Martin, "The Missing Piece in Policy Analysis: Social Security Reform,"
American Economic Review, LXXXVI (May 1996),1-14.
Friedman, Benjamin, and Mark Warshawsky, "Annuity Prices and Saving Behav-
ior in the United States," in Zvi Bodie, John Shoven, and David Wise, eds.,
Pensions in the U. S. Economy (Chicago: University of Chicago Press, 1988),
pp. 53-77.
Friedman, Benjamin, and Mark Warshawsky, "The Cost ofAnnuities: Implications
for Saving Behavior and Bequests," Quarterly Journal of Economics, CV
(February 1990), 135-154.
Hill, Stephen C., and Barbara Wolfe, "Testing the HMO Competitive Strategy: An
Analysis of Success and Failure in Wisconsin," Journal of Health Economics,
XVI (1997), 261-286.
Holmer, Martin, "Tax Policy and the Demand for Health Insurance," Journal of
Health Economics, III (December 1984), 203-221.
Miller, Robert, and Harold Luft, "Managed Care Plan Performance Since 1980,"
Journal of the American Medical Association, CCLXXI (19), (May 18, 1994),
1512-1519.
Mitchell, Olivia S., James M. Poterba, and Mark J. Warshawsky, "New Evidence
on the Money's Worth of Individual Annuities," MIT mimeo, 1997.
Newhouse, Joseph P., "Reimbursing Health Plans and Health Providers: Efficiency
in Production versus Selection," Journal of Economic Literature, XXXIV
(September 1996), 1236-1263.
Newhouse, Joseph P., Melinda Beeuwkes, and John D. Chapman, "Risk Adjust-
ment and Medicare," Harvard University mimeo, 1996.
Newhouse, Joseph P., and the Insurance Experiment Group, Free For All? Lessons
From the Rand Health Insurance Experiment (Cambridge, MA: Harvard
University Press, 1993).

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC
Phelps, Charles E., "Large-Scale Tax Reform: The Example of Employee-Paid

Health Insurance Premiums," University of Rochester Center for Economic
Research Working Paper, No. 37, 1986.
Price, James R., and James W. Mays, "Biased Selection in the Federal Employees
Health Benefits Program,"Inquiry, XXII (Spring 1985), 67-77.
Reber, Sarah J., "Does Price Matter? An Analysis of Harvard Employees' Health
Insurance Decisions Under Managed Competition," senior thesis Harvard
University, 1996.
Riley, Gerald, Cynthia Tudor, Yen-pin Chiang, and Melvin Ingber, "Health Status
of Medicare Enrollees in HMOs and Fee-for-Service in 1994," Health Care
Financing Review, XVII (Summer 1996), 65-76.
Royalty, Anne Beeson, and Neil Soloman, "Health Plan Choice: Price Elasticities in
a Managed Competition Setting," Stanford University, October 1995.
Scheckler, William E., and Rockwell Schulz, "Rapid Change to HMO Systems:
Profile of Dane County, Wisconsin, Experience," Journal of Family Practice,
XXIV (1987),417-424.
Small, Kenneth A., and Harvey S. Rosen, "Applied Welfare Economics with
Discrete Choice Models,"Econometrica, XLIX (1981), 105-130.
Taylor, Amy, and Gail Wilensky, "The Effect of Tax Policies on Expenditures for
Private Health Insurance," in Jack A. Meyer, ed., Market Reforms in Health
Care: Current Issues, New Directions, Strategic Decisions (Washington, DC:
American Enterprise Institute, 1983).
Trajtenberg, Manuel, "The Welfare Analysis of Product Innovations, with an
Application to Computed Tomography Scanners," Journal of Political Economy,
XCVII (1989), 444-479.
Welch, W. Pete, "The Elasticity of Demand for Health Maintenance Organiza-
tions," Journal of Human Resources, XXI (Spring 1986), 252-266.

78.192.218.89 on Sat, 20 Nov 2021 12:54:34 UTC

Health Inssurance

Uploaded by

Copyright:

Available Formats

Health Inssurance

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Health Inssurance

Uploaded by

Copyright:

Available Formats

Paying for Health Insurance: The Trade-Off between Competition and Adverse Selection

Author(s): David M. Culter and Sarah J. Reber

Stable URL: https://www.jstor.org/stable/2586909

This content downloaded from

DAVID M. CUTLER AND SARAH J. REBER

We use data on health plan choices by employees of Harvard University to

Governments are increasingly turning to market forces as a

* We are grateful to Sally Zeckhauser, Thomas Schmitt, and Lydia Cummings

This content downloaded from

This content downloaded from

ous insurance entirely. We estimate the welfare loss from adverse

I. PRICING RULES AND INSURANCE MARKET EQUILIBRIUM

To demonstrate the issues involved in alternative pricing

This content downloaded from

the contribution for each worker and how that contribution

(1) Ek= Ak + [Pk,

where Ak is a fixed amount (potentially varying by plan), Pk is the

We begin by showing the trade-off between adverse selection

This content downloaded from

on choice and more generous provider reimbursement (such as a

This content downloaded from

HMO --------------------------> ho <------------------------------------------------- PPO Health (h)

This content downloaded from

HMO, the PPO premium will increase. If this premium increase

This content downloaded from

actual difference in plan premiums will be

(2) PPPo - PHMO = (1 - c)hL(h') + [hu(h') - hL(h')].

This content downloaded from

assume that initially the employer is subsidizing the PPO more

There may also be supply-side effects of changing pricing

(3) Aq/q = 400 (1 - P)x-q

6. We term this a redistribution, but it may also involve efficiency issues. Ex

This content downloaded from

premium change increases. Employers will thus capture more of

This content downloaded from

reduction in premiums is just a transfer of resources from

While the equal contribution rule may or may not increase

This content downloaded from

II. THE HARVARD UNIVERSITY EXPERIENCE

We examine the response to pricing reform empirically using

This content downloaded from

10. Economically, it may not be meaningful to speak of a deficit in one part of a

This content downloaded from

Measure 1992 1993 1994 1995 1996

In analyzing the 1995 data, we use the 1996 Treatment Group as a

III. ENROLLMENT RESPONSES TO POLICY REFORM

The first question we examine is the enrollment response to

This content downloaded from

over time. Consistent with a nonzero demand response, PPO

(4) PPO t = y In (P o~pit) + XitF

where i denotes individuals and t denotes time. In principle, we

This content downloaded from

PPO enrollment Drop coverage

ln(P 0p) 3.64 -.430** -.784** 1.618** .794**

We include a variety of control variables in the model, the

This content downloaded from

model HMO clinic.15 Since we also include year and treatment

This content downloaded from

between -2 and -8. More recently, data from individual choices

This content downloaded from

ln(P 0p) 3.64 -.430 -.784 1.618 .794