FindArticles > Health Services Research > Oct, 1999 > Article > Print friendly
Simulating the effects of employer contributions on adverse selection and health plan choice - Health Plan Choice
M. Susan MarquisArchitects and proponents of managed competition have argued that employers must contribute equally to ali health plan premiums, a practice that would force employees to bear the real cost differences across health plans (Enthoven and Kronick 1989; Ellwood, Enthoven, and Etheridge 1992). They assert that employees will choose more competitive plans that offer better value when they face the real cost differences. Despite these assertions, recent data indicate that 88 percent of large employers that offer multiple health plan options continue to subsidize more expensive health plans (Hunt, Singer, Gabel, et al. 1997).(1) Further, some health policy consultants argue that multiple choice of health plans does not contain costs or insurance premiums (Jones 1990).
One can then ask why employers are so reluctant to implement a strategy that is expected to reduce costs both to themselves and to their employees. Enthoven hypothesizes that this reluctance is the reason that multiple plan offerings have not been effective in containing costs, and that it stems from three sources. First, employers became committed to a policy that paid all of the costs of a fee-for-service plan when costs were much lower. Second, union leaders believe that comprehensive health benefits are a precious bargaining prize not to be eroded. Finally, employers face a collective action problem in which a single employer who adopts such a policy risks losing valued employees to competitors (Enthoven 1990).
Another unacknowledged reason for this reluctance may be drawn from several natural experiments where large employers that offered multiple plan options converted from subsidizing higher cost plans to an equal contribution policy. Data on the conversion experience from the Group Insurance Commission of Massachusetts (which serves state and local employees), the University of California, Harvard University, and large employers in Minnesota all indicate a strong plan-switching response to increased premium costs as intended (Buchmueller and Feldstein 1997; Cutler and Zeckhauser 1997; Feldman and Dowd 1993).
An unintended consequence was that the reform induced risk-based sorting across the plans (Cutler and Reber 1996). The adverse selection within the Harvard system was sufficient to drive the most generous policy out of the market within three years. The Group Insurance Commission of Massachusetts moved to contain the adverse selection by subsidizing premiums on a proportional basis and managing the most generous policy very tightly (Cutler and Zeckhauser 1997).
In the work that follows, we use microsimulation methods to demonstrate and further explore the relationship between employer premium contributions and the stability of insurance markets. We compare a fixed dollar premium contribution with one in which the employer pays a fixed share of the alternative plans. We also explore two contribution policies in which risk differences among plans are incorporated into the employer's calculation in an effort to reduce selection bias, maintain the viability of a multiple-choice offering, and encourage competition based on efficiency and not selection.
THE MICROSIMULATION MODEL
Our simulation includes two behavioral components: a model of family demand for health insurance and an individual model of health services demand. These models were originally developed and tested using data and analyses from the Health Insurance Experiment (HIE), a controlled trial of the effect of health insurance generosity on healthcare use (Buchanan et al. 1991; Keeler, Buchanan, Rolph, et al. 1988; Marquis and Holmer 1996). They have been updated and modified for the current study (Marquis and Buchanan 1992 and 1994). We use the model to simulate health insurance choices over time to investigate the effects of selection on the market equilibrium.
The Insurance Demand Model
The health insurance demand component of the model predicts a family's choice of insurance plan from among a specified set of options. In our methodology, the family has expectations about its health needs and understands how the cost-sharing provisions of different insurance plans will affect its out-of-pocket costs for healthcare given these expectations. In addition, because the structure of an insurance plan affects overall health spending for all enrollees, different plans will have different insurance premiums. The family is assumed to select the insurance plan that results in the highest expected utility. The family's expected utility from any plan depends on the insurance premium, expected out-of-pocket payments for health services given the copayment and deductible provisions of the plan, and how risk averse the family is. Specifically, a family's utility given that health insurance plan j was selected and that health state k occurred is given by:
U(k, j) = - exp(- a * n(k, j)) + w,
where n is the income available to the family for spending after payments are made for health insurance premiums and for out-of-pocket medical care expenses; the parameter a reflects the degree of risk aversion; and w is a stochastic component. The quantity n(k, j) will vary with the chosen plan because premiums and out-of-pocket payments depend on the characteristics of the insurance plan. The expected utility from choosing plan j is:
EU(j) = [summation over k] U(k, j) + E(w).
The family is assumed to choose plan j over plan m if EU(j) [greater than] EU(m).
The parameters of the health insurance demand model derive from Marquis and Holmer's (1996) analysis of responses from families who participated in the HIE about their preferences among hypothetical health insurance policies. The risk parameter a is 0.0006 and the E(w) is a normally distributed variate with an expected value of zero.(2) The risk parameter in our model suggests a price elasticity of demand to purchase health insurance of about -.3; this is consistent with a number of empirical studies of the price response (see, for example. Farley and Wilensky 1985; Holmer 1984; Marquis and Long 1995). Moreover, in other work, we have demonstrated that the insurance model has predictive validity in that it predicted decisions to purchase insurance among a sample of working families that were consistent with their actual behavior (Marquis and Buchanan 1994).
In our analysis, we assume that the employer offers a choice between a health maintenance organization (HMO) plan and another type of managed care plan, such as a preferred provider organization (PPO). Both plans provide medical, mental, and preventive health benefits; they differ in the copayment and deductible provisions. The HMO plan has cost-sharing provisions that amount to about 10 percent of the cost of a visit. The PPO plan has a $100 deductible and a 20 percent cost-sharing provision. The HMO also offers a more restricted choice of physicians than the PPO.
Because most employees who are offered insurance participate in a plan (Long and Marquis 1993), we assume that all employees choose one of these options so that we can focus on the effects of different employer contribution policies on the choice between the plans.(3) The health insurance model predicts which of these employer plans a family would choose based on its income, its expectations about future healthcare needs, the premium the family pays out-of-pocket for each plan, and individual preferences for greater freedom of choice among providers. Most research suggests that people do value choice (see, for example, Tessler and Mechanic 1975: Friedman, LaTour, and Hughes 1984), but little empirical work exists that quantifies these preferences and their effects on decisions. We have therefore adopted a range of assumptions about the value of choice to test the sensitivity of our conclusions. These are described later.
The Health Episode Model
This model is used to generate the family's expectation about the distribution of possible healthcare spending it faces in the next year when making its insurance choice decision. The model is also used to calculate the premiums for each plan and to measure risk selection in each plan.
The health episodes model assumes that healthcare is determined jointly by patients and their doctors. Managed care plans may also influence health spending through a variety of review and management activities, as well as through financial and peer incentives to physicians. Patient demand for healthcare is influenced by the underlying health problem (i.e., the type of signs and symptoms or the clinical condition being experienced) and by the out-of-pocket costs of obtaining healthcare. Once a patient decides to seek care, doctors and healthcare systems largely determine the amount of treatment received. These treatment decisions axe influenced both by the patient's health condition and by financial and peer incentives on the doctor, along with the care monitoring activities of health plans.
The health episodes simulation model was designed as a flexible tool that can incorporate behaviors by patients, physicians, and health plans. The model begins by estimating the underlying "health events or episodes" of each individual in the family. The estimate is characterized by four types of health episodes: (1) those serious enough to involve a hospital stay if treatment is sought, (2) acute outpatient care episodes, (3) chronic outpatient episodes, and (4) well care. A chronic outpatient episode would include all of the routine outpatient care associated with a particular chronic condition, such as arthritis. An acute flare-up of a chronic condition is classified as an acute episode. An acute illness, such as pneumonia, is also classified as an acute episode. Well-care episodes include preventive care, routine gynecology, well-baby check-ups, and annual physicals. The health episodes model includes only those episodes for which people seek care when they do not have to pay out-of-pocket for the care (free care), and it is based on episode occurrence rates observed in the HIE. The episode occurrence rate assumes that doctors have no incentives to limit treatment and that health plans are passive. Thus, total family episodes at time t can be represented as:
[Mathematical Expression Omitted]
where [X.sub.ijk] is the size of the kth episode of type j for family member i.
Within the simulation model, the number of episodes of each type is given by the negative binomial counting processes {[N.sub.ij](t), t [greater than or equal to] 0}, one for each family member and episode type. As a compound Poisson distribution, the negative binomial has a convenient interpretation in this context. If the number of episodes per person of episode type j is assumed to have a Poisson distribution with an intensity parameter, [[Lambda].sub.j], a gamma distributed random variable, then the distribution of the number of episodes of a particular type is given by the negative binomial distribution.
Individual episode propensities, [[Lambda].sub.ij], are assumed to have a measured component, [d.sub.ij], determined by observable characteristics, such as age, sex, ethnicity, and socioeconomic status, that are known to influence the mean level of episodes that a person experiences, and an unmeasured component, [W.sub.j], that is determined by unobservable factors, such as health attitudes and care-seeking propensities. The [d.sub.ij] are calculated using individual characteristics and the negative binomial regression coefficients estimated on the HIE data.(4) The unmeasured component [W.sub.j] introduces the concentration of episodes within some families that is unexplained by measured characteristics and the correlation across episode types. These [W.sub.j] are gamma-distributed random variables with mean equal to one. Thus:
[[Lambda].sub.ij] = [d.sub.ij] [W.sub.j] with [W.sub.j] [approximately] (G: [[Alpha].sub.j], [[Beta].sub.j])
and
{[N.sub.ij](t) [where] [[Lambda].sub.ij], t [greater than or equal to] 0} [approximately] Poisson.
Because the sum of independent Poisson processes is a Poisson process, the model aggregates across family members and episode types generating a single episode stream for the family.
Cost sharing, in both managed care and traditional indemnity plans, reduces spending. Keeler, Buchanan, Rolph, et al. (1988) clearly demonstrated that this spending reduction occurs primarily because patients seek care for fewer episodes. This response to cost sharing increases with the level of cost sharing and differs by type of episode. Let [c.sub.ijt] denote the level of cost sharing that person i faces for episode type j at time t during the year. For traditional insurance plans and PPOs, [c.sub.ijt] is typically 100 percent until a deductible is met, and then it drops to the coinsurance rate and may fall to zero for plans with limits on out-of-pocket spending when the limit is reached. Let [p.sub.j] ([c.sub.ijt]) be the rate of occurrence of episodes of type j given coinsurance [c.sub.ijt] relative to the occurrence rate with free care. The episode loss rate is then given by 1 - [p.sub.j]([c.sub.ijt]). Within the simulation, this is modeled as a Bernoulli censoring process applied to the episode occurrence model.
Managed care plans may also reduce the frequency of some types of episodes, especially high-cost services and hospitalizations. Within the model, this reduction is denoted by [r.sub.j] and acts to increase the Bernoulli censoring rate by a multiple of (1 - [r.sub.j]).
The set of episode costs {[X.sub.ijk]}, come from lognormal distributions that are retransformed. The episode type and characteristics of the individual determine the distribution mean for each random draw. The actual calculation of the mean is done using lognormal regression coefficients estimated on HIE data and observed characteristics of the individual. Data from the National Expenditure Accounts each year are used to update the episode costs from the time of the HIE to the present. Inpatient and outpatient expenditures are tracked separately in the National Expenditure Accounts, so the model uses separate updating factors for inpatient and outpatient episodes. These updating factors [b.sub.j] capture the effects of inflation and technology change through time. Keeler, Buchanan, Rolph, et al. (1988) found that the level of cost sharing had a small effect, denoted by [[Pi].sub.j]([c.sub.ijt]), on episode size and that the effect varied with the level of cost sharing and the type of episode.
Thus, family health expenditures are represented by:
[Mathematical Expression Omitted]
and [Mathematical Expression Omitted] is the original episode occurrence process {[N.sub.ij](t), t [greater than or equal to] 0} that has been subjected to Bernoulli censoring with loss rate [1 - [p.sub.j]([c.sub.ijt])] * (1 - [r.sub.j]) attributable to cost sharing and care management.
The original version of the model was validated extensively. Simulated data on total spending, the components of spending, and utilization measures were compared to actual data for one of the HIE sites. In addition, pseudo claims generated by the model were used to reestimate the set of gamma parameters used in our model. These did not differ statistically from those observed in the actual data. Further, we compared episode-type correlations within and across family members and found that these also replicated the underlying data. Through time, we have compared model-generated premiums with those observed in similar health plans from a number of different survey sources. Our model has consistently performed well. For example, the $2,048 premium for single coverage generated by our model for the PPO option in this article is quite close to the average reported premium of $2,000 in the 1993 Robert Wood Johnson Employer Health Insurance Survey for plans with similar benefits (see also Buchanan et al. 1991; Buchanan and Marquis 1999; Marquis and Buchanan 1992, 1994).
Uses of the Health Episode Model
Premiums for the two plans are determined by using the episode model to simulate spending by individuals and families enrolled in the plan in the previous year; that is, each plan's premium is based on actual experience in the prior year. Predicted spending by families in the HMO is discounted by 20 percent to reflect success in managing care.(5) We initialize the model in period 0 by establishing premiums as if families randomly sort themselves between plans. That is, in the initial period, premium differences are the result only of differences in benefits and efficiency. In the next period (period 1), premiums depend on the actual claims experience of those who selected the plan in the prior period. Over time, then, differences in the premiums of the plans will reflect the health risk of those selecting the plan as well as the differences in cost sharing and the extent of healthcare management. Premiums are set by inflating the expected claims experience by 10 percent to reflect the administration, risk, and sales cost to the insurer.(6) Separate premiums are calculated for single coverage, coverage for two persons, and coverage for families of three or more.
We also use the episode simulation model to produce the distribution of expenditures families face as they evaluate the insurance purchase decision by generating 50 replicates of annual expenditures for each person. We assume that the family does not anticipate that the choice of health plan will affect demand and so evaluates all plans based on a common expenditure distribution. This is analogous to assuming that the family has a reference plan against which it evaluates alternatives. By contrast, the plan premiums are based on the estimated level of expenditures that do account for differences in healthcare demand depending on the generosity of insurance. Thus, our simulations assume that ex ante - in evaluating the decision to purchase insurance - a family does not anticipate behavioral differences depending on the decision,(7) but that ex post - once the plan is purchased - healthcare spending, claims reimbursement, and thus premiums reflect the effect of the insurance choice on health services demand.
Finally, the health episode model is used to estimate risk selection in each plan. We measure risk in a plan as the average level of per person spending by employees and their dependents enrolled in the plan, adjusted for differences in benefits between the plan. The benefit structure we use in measuring risk are those of the PPO plan.
DATA
Database
The database processed in the simulation includes families that were part of the 1993 Current Population Survey (CPS) Employee Benefits Supplement sample, a half-sample of the April 1993 CPS of persons who were working for pay at the time of the survey. Our simulations define families to correspond to insurance units: that is, a family head and spouse and their own children under age 18 or up to age 23 if students. Thus, adult children who live at home and work comprise a second insurance family in our analysis. Our sample includes 17,143 families with at least one employed family head.
The CPS provides information on the demographic and economic characteristics of employees and their dependents who are included in the models of episodes and their costs that are used to generate the family's expected health expenditure distribution and premiums for health insurance plans. Given the expenditure distribution, family income as measured in the CPS, and the characteristics of the plan options, the insurance demand model predicts each family's choice among the options. These choices in turn affect claims expenditures - predicted from the health episode model - and, hence, premiums in the next year.
Measuring Employee Premium Payments
We contrast employee choices among the insurance options under four different employer premium contribution policies. Under the first, the employer contributes a fixed percentage of the total premium, and thus pays a greater share of the cost of more expensive plans. The employer share is set at 84 percent of the premium for single coverage and 71 percent for family coverage.(8) Under the second policy, the employer pays a fixed dollar amount for each plan, regardless of its premium, and so the family pays the full difference in price. The employer contribution for both plans equals the contribution made to the lower-cost plan in the initial period and remains fixed at this same dollar amount over all periods of the simulation.
A fixed dollar contribution policy places all of the increase in premium resulting from adverse selection on the employee, which may worsen selection problems and threaten the viability of some plans. With a constant percentage contribution, the employer subsidizes cost differences between plans that arise from efficiency differences as well as from different risk selections. Bowen and Slavin (1991) describe several other methods that employers and benefits consultants use to adjust for risk in the employer's contribution. We consider two of these methods in this article.
In the first, or ratio method, the ratio of the employer contribution is set equal to the ratio of the plans' risk. To calculate the contribution for each plan, we assume that the average employer contribution is equal to the amount used in the fixed-dollar approach.(9) In the second method, termed "everyone in fee-for-service" by Bowen and Slavin, the employer contribution to the fee-for-service plan is set equal to the average employer contribution plus the difference between the actual fee-for-service plan premium and the premium adjusted for risk.(10) An advantage of the ratio method is that it does not take into account actual premiums, and therefore the employer does not subsidize price differences arising from inefficiencies. A disadvantage is that it adjusts only the employer contribution and so does not fully adjust the employee payment for risk differences. The other method provides a greater adjustment in the employee contribution for risk, but it leads to an employer subsidy of inefficiencies (Bowen and Slavin 1991).
The Value of Provider Choice
Research suggests that restricted choice of providers is viewed as a disadvantage of selecting an HMO plan.(11) However, little evidence exists about the value of free choice of provider and its relative importance in the choice among plans. Therefore, we have carried out our simulation under several alternative assumptions about the perceived cost of choosing an HMO with its greater restrictions on provider choice. First, we assume that this does not factor into decisions; the choice of plans is based only on the expected financial consequences of the decision.
As an alternative, we use results from research on Medicare beneficiaries that estimated the "value of free choice" of provider to be about $75 (in current dollars) per person per year (Marquis and Rogowski 1991). This was an estimate of the dollar transfer that would make beneficiaries indifferent between staying in a PPO with a small network and moving to an HMO. This $75 per person is treated as an additional cost to the family if it chooses an HMO. However, all individuals may not value freedom of choice of physician equally. Higher-risk individuals, in particular, may be reluctant to give up established physician contacts and, hence, view restricted choice as posing a greater cost than do those at less health risk. Therefore, we investigate how choices among the plans differ when all individuals are assumed to place a $75 "cost" on the restricted choice of physicians and under the assumption that only individuals in high-risk families - those with expenditures above the median - do so.
RESULTS
Alternative Contribution Policies
Table 1 compares the fixed dollar and fixed percent contribution policies under the assumption that employees do not attach value to the greater freedom of choice of provider afforded by the PPO and make their choices based only on financial considerations. The table shows total premiums, employee out-of-pocket contributions to premiums, and the number and characteristics of those selecting each plan over four years (period 0 through period 3). By the end of period 3, either only one plan remained in the market or the model had reached a stable solution and enrollment shares remained constant thereafter.
In the initial period (period 0), premiums for the two plans are based on the average health risk in the population. In our example, the HMO plan provides more generous benefits (in lower cost-sharing requirements) than the PPO plan, and its premium is slightly higher than that of the PPO plan: $2,093 and $2,048 per year, respectively, for single coverage. Despite the more generous HMO benefits, the premium difference is small ($45 per year for single coverage) because we have assumed managed care efficiencies in the HMO. When employers contribute a fixed dollar amount to coverage, the difference in employee contribution to the plans is equal to the difference in the total premium of the two plans, as illustrated in the first two columns of Table 1. Given the small difference in employee contribution, the HMO plan with its lower cost sharing is preferred by the majority of employees in the initial period: 84 percent select the HMO plan (see the bottom row for period 0 in Table 1, column 1).
Although those who choose the PPO make up only a small fraction of the employee population, they are at low health risk. Among employees with self-only coverage, those choosing the PPO in the initial period have a health risk (measured as spending adjusted for differences in plan benefits) that is about 25 percent lower than those choosing the HMO in the initial period ($1,512 versus $2,003). Among families of three or more, health spending by those selecting the PPO is 10 percent lower than among those families in the HMO. For the healthiest families, the expected utility gain from the enhanced HMO benefits does not outweigh the additional premium.
Table 1: Choice Between HMO and PPO Under Different Incentives
(assumes that patients evaluate plans only on financial
consequences)
Employer Contribution Policy
Fixed Dollar Amount % of Premium
HMO PPO HMO PPO
INITIAL PERIOD (PERIOD 0)
Premium
Single coverage $2,093 $2,048 $2,093 $2,048
Families of three or more $6,089 $6,053 $6,089 $6,053
Employee Contribution
Single coverage $ 373 $ 328 $ 335 $ 328
Families of three or more $1,792 $1,756 $1,766 $1,755
Health Risk of Purchasers(*)
All families $1,915 $1,612 $1,910 $1,625
Single coverage $2,003 $1,512 $1,984 $1,540
Families of three or more $1,537 $1,412 $1,536 $1,419
Percent Purchasing 84% 16% 84% 16%
PERIOD 1
Premium
Single coverage $2,222 $1,673 $2,204 $1,695
Families of three or more $6,390 $5,825 $6,388 $5,837
Employee Contribution
Single coverage $ 502 $ 0 $ 353 $ 271
Families of three or more $2,093 $1,528 $1,853 $1,693
Health Risk of Purchasers(*)
All families $2,005 $1,629 $1,949 $1,041
Single coverage $2,195 $1,778 $2,191 $1,000
Families of three or more $1,989 $ 809 $1,530 (**)
Percent Purchasing 30% 70% 77% 23%
PERIOD 2
Premium
Single coverage $2,458 $1,956 $2,429 $1,095
Families of three or more $8,335 $3,210 $6,435 NA(***)
Employee Contribution
Single coverage $ 738 $ 236 $ 389 $ 175
Families of three or more $4,038 $ 0 $1,866 NA
Health Risk of Purchasers(*)
All families $2,394 $1,740 $1,770 $1,810
Single coverage $2,491 $1,828 $2,534 $1,406
Families of three or more (**) $1,397 $1,400 (**)
Percent Purchasing 7% 93% 56% 44%
PERIOD 3
Premium
Single coverage $2,735 $2,011 $2,734 $1,546
Families of three or more NA $5,845 $5,967 NA
Employee Contribution
Single coverage $1,015 $ 291 $ 437 $ 247
Families of three or more NA $1,548 $1,730 NA
Health Risk of Purchasers(*)
All families $2,433 $1,706 $1,676 $1,794
Single coverage $2,636 $1,741 $2,237 $1,349
Families of three or more NA $1,412 $1,412 NA
Percent Purchasing 3% 97% 57% 43%
* Average spending per person in year standardized for differences
in benefits, families of 2 members not shown separately.
** Purchased by one percent or fewer of families in category.
*** Not available; purchased by one percent or fewer families in the
category in the previous period.
As a result of the favorable selection in the PPO and adverse selection in the HMO, the premium for the former decreases and the premium for the HMO increases (see period 1 in Table 1). In period 1, the total premium for the HMO is more than $500 higher than the PPO premium, but the difference is less than $40 in period 0. When employees pay this full cost difference out-of-pocket, many fewer employees select the HMO in period 1 (30 percent). Those who stay in the HMO have a greater health risk than those who switch to the PPO; thus, the average health risk of those in the HMO in period I is higher than in period 0. For example, among those enrolled in single coverage, the health risk of HMO participants in period 1 is $2,195 versus $2,003 in the preceding year. As a result, the actual claims payout in period 1 is higher than in the previous period, and so the premium for the HMO again increases relative to the PPO for period 2. These continuing increases in the HMO premium ultimately lead to a collapse of the market for the HMO plan; by period 3, only 3 percent of families remain in the HMO. In our model, the HMO plan is forced out of the market because it is the more expensive plan. In some observed situations, the HMO plan has been the less expensive plan, thereby driving out the more expensive option (e.g., Cutler and Zeckhauser 1997). Thus, our simulation illustrates that adverse selection against high-cost, high-benefit plans can be substantial when employees are given a choice among plans.(12)
A similar premium increase occurs between period 0 and period 1 in our simulation of choice when employers pay a proportional share of the cost of each plan (as shown in the last two columns of Table 1) instead of the same dollar cost of each plan. However, the response to the rising premium is less because employees do not pay the full difference in the cost of the two plans. Although the total premium difference between the plans is more than $500 in period 1, the additional cost to employees who enroll in the HMO plan with self-only coverage is only $82 ($353 versus $271), and the additional cost to families enrolling in the HMO is $160. Because the price difference in the two plans is subsidized, the higher-priced plan remains attractive to many of the individuals at more moderate risk. Overall, 77 percent of employees remain in the HMO in period 1, and virtually all employees who enroll two or more dependents choose the HMO plan. Because some flight from the HMO continues to occur on the part of the more moderate health risks among smaller families, HMO premiums for these families increase again in period 2 and lead to some additional plan switching. However, by period 3, we achieve a stable equilibrium in which both plans continue to survive.
Table 2 contrasts three strategies in which the employer shares in cost differences between plans: the fixed percent of premium policy and the two strategies that adjust directly for risk selection. Under the fixed percent of premium policy, the contributions are proportional to the actual premiums, whereas with the ratio policy, the contributions are proportional to the plan risks. If premium differences between the two plans reflect only risk differences, then the two ratios are the same. However, the ratio policy also incorporates a constraint on the total employer contribution, whereas the fixed percent of premium policy does not. As a result, in our illustration, the total employer contribution increases under the fixed percent of premium policy as consumers select the higher-priced plan, and more individuals and families choose the higher-priced HMO than under the ratio method.(13) The "everyone in fee-for-service" policy also constrains employer contributions, but it makes a greater absolute adjustment to contributions for risk than does the ratio method (Bowen and Slavin 1991). In our example, it leads to the highest enrollment in the HMO than the other risk adjustment mechanisms. [TABULAR DATA FOR TABLE 2 OMITTED] However, enrollment rates in the HMO are greater than a perfectly risk-adjusted mechanism, which are the rates in period 1 (see Table 1) when insurers are assumed to set premiums expecting random selection into plans.
Systematic Variations in Preference for Freedom of Choice
The results are quite similar when we incorporate preferences for freedom of choice of provider into the decision model if we assume that all persons value equally the freedom to choose (see Table 3). Premium increases drive out the HMO within about three years if employers contribute a fixed amount to every plan, whereas both plans attract some enrollees when the added cost of the HMO is subsidized by the employer. However, because the "cost" of the HMO includes the cost of restricted choice,(14) fewer individuals and families select the HMO in any period. The contrast with the results of Table 1 are especially marked for the choices of larger families when employers contribute a proportional share of the premiums. Because we have assumed a per person cost of restricted choice, the "cost" of restricted choice increases with the size of the family. When faced with both the difference in premium costs and the cost of restricted choice, lower-risk families begin to switch to the PPO, driving up the premiums for families of three or more. This eventually induces all families of three or more to switch to the PPO. However, individuals continue to choose both the HMO and PPO options. As a result, the HMO retains about 15 percent of contracts. In contrast, when families paid only the subsidized premium difference and choice was not influenced by preferences for nonmonetary aspects of the plan (as in Table 1), the HMO plan was chosen by virtually all of the larger families.
Table 3: Choice Between HMO and PPO Under Different Incentives
(assumes that all patients value free choice of doctor = $75 per
person)
Employer Contribution Policy
Fixed Dollar Amount % of Premium
HMO PPO HMO PPO
INITIAL PERIOD (PERIOD O)
Premium
Single coverage $2,093 $ 2,048 $ 2,093 $2,048
Families of three or more $6,089 $ 6,053 $ 6,089 $6,053
Employee Contribution
Single coverage $ 373 $ 328 $ 335 $ 328
Families of three or more $1,792 $ 1,756 $ 1,766 $1,755
Health Risk of Purchasers(*)
All families $1,983 $ 1,440 $ 2,093 $1,448
Single coverage $2,083 $ 1,429 $ 2,050 $1,454
Families of three or more $1,585 $ 1,162 $ 1,586 $1,153
Percent Purchasing 78% 22% 79% 21%
PERIOD 1
Premium
Single coverage $2,299 $ 1,572 $ 2,265 $1,600
Families of three or more $6,541 $ 5,009 $ 6,547 $4,973
Employee Contribution
Single coverage $ 579 $ 0 $ 362 $ 256
Families of three or more $2,244 $ 712 $ 1,899 $1,442
Health Risk of Purchasers'
All families $1,982 $ 1,725 $ 2,452 $1,318
Single coverage $2,209 $18,122 $ 2,392 $1,298
Families of three or more (**) $ 1,489 $ 2,452 $ 928
Percent Purchasing 6% 94% 38% 62%
PERIOD 2
Premium
Single coverage $2,492 $ 1,993 $ 2,641 $1,428
Families of three or more NA(***) $ 6,232 $10,013 $3,839
Employee Contribution
Single coverage $ 772 $ 273 $ 423 $ 228
Families of three or more NA $ 1,935 $ 2,904 $1,113
Health Risk of Purchasers'
All families $2,397 $ 1,753 $ 2,556 $1,638
Single coverage $2,537 $ 1,855 $ 2,665 $1,468
Families of three or more NA $ 1,397 (**) $1,397
Percent Purchasing 5% 95% 16% 84%
PERIOD 3
Premium
Single coverage $2,762 $2,040 $2,864 $1,615
Families of three or more NA $5,845 NA $5,845
Employee Contribution
Single coverage $1,042 $ 320 $ 458 $ 258
Families of three or more NA $1,548 NA $1,695
Health Risk of Purchasers(*)
All families $2,501 $1,709 $2,284 $1,626
Single coverage $2,669 $1,755 $2,237 $1,349
Families of three or more NA $1,412 NA $1,412
Percent Purchasing 2% 98% 15% 85%
* Average spending per person in year standardized for differences
in benefits, families of 2 members not shown separately.
** Purchased by one percent or fewer of families in category.
*** Not available; purchased by one percent or fewer families in the category
in the previous period.
Our conclusions are different if not all individuals value choice equally and preferences are systematically related to risk. Table 4 assumes that the higher-risk individuals - those in families whose risk is above the median for their family size - value greater freedom of choice of provider than do lower-risk individuals. This changes the nature of risk selection, especially among families, because our illustration assumes that the cost of restricted choice is a per person cost but that the choice of health plan is a family decision. Under this assumption, it is the high-risk families that select the PPO to retain greater provider choice. This leads to a premium increase for family coverage in the PPO, and the premiums eventually rise to a level that drives all families of three or more to the HMO plan. However, the PPO remains the preferred option among smaller families, and so the two plans continue to coexist.(15)
CONCLUSIONS
Adverse selection is significant and can undermine the managed competition strategy. It can substantially drive up the price of one plan relative to the price of others, so that the plans are not competing on the basis of efficiency, but rather on the basis of selection. In our simulation, for example, the majority of individuals and families preferred the higher-priced HMO plan when premiums were set to reflect only benefits and production efficiency (period 0). However, under some scenarios, selection against the HMO drove the price up to a level where no consumers continued to purchase the HMO. The HMO plan was driven from the market when it was the higher-priced plan, when employer's contributed a fixed dollar amount to the plans, and when variation among individuals in preferences was assumed to be random. Systematic variation in preferences that was correlated with risk led to different outcomes. In addition, had we assumed greater efficiencies in the HMO plan, so that the premium cost was lower but the benefits higher, the PPO plan would have been the one driven from the market. Moreover, our model assumes that insurers and plans price passively - simply adjusting premiums in each period to reflect the prior year experience of the group. However, if insurers and plans take more strategic action in the face of adverse selection, then multiple plans might continue to exist even under the scenarios that led to a collapse of the multiple choice situation in our scenarios. Such insurer or plan action might take the form of a benefit redesign to attract lower risk or an introduction of cost-containment features. Moreover, if a single insurer or plan offers a portfolio of products to the employees in a group, the insurer may require only that the portfolio return a profit. Theoretically, in this case, the insurer or plan will maintain the difference in plan premiums at the difference in the community rates, and no adverse selection will occur (Cave 1985).
Nonetheless, our simulation shows that a fixed employer contribution may be a factor that leads to a collapse of the multiple-choice situation. This result shows the importance of finding good risk adjusters to allow plans to compete on the basis of quality and efficiency. We considered three risk adjustment mechanisms that have been used by employers. Of those we considered, no ideal candidate surfaced. None achieved the result that would be obtained if premiums solely reflected differences in risk. Moreover, some are difficult to implement and require employer information about the true relative risks of the plans. Some - the fixed percent of premium and the "everyone in the fee-for-service" strategies - also would raise contributions for less efficient plans. However, all mitigated the selection problems and maintained a multiple-choice environment. The fixed percent of premium achieved the best balance across plans, but it has the other limitations just discussed.
Table 4: Choice Between HMO and PPO Under Different Incentives
(assumes that high-risk patients value free choice of doctor = $75
per person)
Employer Contribution Policy
Fixed Dollar Amount % of Premium
HMO PPO HMO PPO
INITIAL PERIOD (PERIOD 0)
Premium
Single coverage $2,093 $2,048 $2,093 $2,048
Families of three or more $6,089 $6,053 $6,089 $6,053
Employee Contribution
Single coverage $ 373 $ 328 $ 335 $ 328
Families of three or more $1,792 $1,756 $1,766 $1,755
Health Risk of Purchasers(*)
All families $1,914 $1,643 $1,911 $1,639
Single coverage $2,010 $1,539 $1,992 $1,546
Families of three or more $1,527 $1,521 $1,528 $1,513
Percent Purchasing 82% 18% 83% 17%
PERIOD 1
Premium
Single coverage $2,226 $1,692 $2,208 $1,701
Families of three or more $6,326 $6,544 $6,333 $6,503
Employee Contribution
Single coverage $ 506 $ 0 $ 353 $ 272
Families of three or more $2,029 $2,247 $1,837 $1,886
Health Risk of Purchasers(*)
All families $1,545 $1,894 $1,866 $1,424
Single coverage $2,047 $1,827 $2,156 $1,325
Families of three or more $1,494 (**) $1,493 (**)
Percent Purchasing 44% 56% 72% 28%
PERIOD 2
Premium
Single coverage $2,323 $2,010 $2,392 $1,458
Families of three or more $6,272 NA(***) $6,272 NA
Employee Contribution
Single coverage $ 603 $ 290 $ 383 $ 233
Families of three or more $1,975 NA $1,819 NA
Health Risk of Purchasers(*)
All families $1,676 $1,903 $1,759 $1,827
Single coverage $2,657 $1,670 $2,392 $1,471
Families of three or more $1,397 NA $1,397 NA
Percent Purchasing 48% 52% 58% 42%
PERIOD 3
Premium
Single coverage $2,843 $1,836 $2,595 $1,618
Families of three or more $5,967 NA $5,967 NA
Employee Contribution
Single coverage $1,123 $ 116 $ 415 $ 259
Families of three or more $1,670 NA $1,730 NA
Health Risk of Purchasers(*)
All families $1,447 $1,898 $1,687 $1,783
Single coverage $2,733 $1,773 $2,143 $1,462
Families of three or more $1,412 NA $1,412 NA
Percent Purchasing 38% 62% 58% 42%
Note: Families with per person health risk above the median for the
family size group.
* Average spending per person in year standardized for differences
in benefits, families of two members not shown separately.
** Purchased by one percent or fewer of families in category.
*** Not available, purchased by one percent or fewer families in
the category in the previous period.
Our model produces results that are similar to the findings of the natural experiments that find the employer contribution policy to have a significant effect on choice. When the employer subsidizes the higher-priced plan (as most employers do), the choice mix is quite different than it is when employees face the full difference in price. In our model, and the natural experiments, all employees switched to the low-cost plan within a very few years when employers contributed a fixed dollar amount to all plans. In our scenarios, the subsidy served to offset some of the undesirable effects of selection and permitted the higher-priced HMO plan to continue to compete with the other plan. However, this subsidy would also diminish the incentive to switch from a plan whose higher price reflected inefficiencies. The contribution policy also has an effect on total costs. By encouraging more families to purchase the higher-priced plan, employers who subsidize a part of the price differential face higher total premium expenditures for healthcare. For example, under the scenario depicted in Table 1, total premium expenditures are almost 9 percent higher, given the choice mix resulting from the employer policy that contributes a constant share of premiums, than the policy that pays a fixed dollar amount for each plan.
Our results also show that employee preferences for nonfinancial aspects of the plan can also have an important effect on the types of plans chosen. We find that both high- and low-cost plans can exist if employers subsidize the cost difference. But consumer preferences for the nonfinancial aspects of plans can also lead to this result. We illustrated the case in which a preference for free choice of physicians varies systematically with the individual's health risk. However, there may be other attributes of different kinds of plans that are valued differently by high- and low-risk groups. Some of these would worsen adverse selection and some, as in our illustration, would mitigate it. Not much is known about these preferences and how they vary among different population subgroups. But our results indicate that understanding the value that people place on freedom of choice and other attributes of a plan is important in predicting their responses to policy changes. This knowledge will also help indicate what kind of information consumers need if we are to make managed competition work.
Finally, it is important to think about what these findings might suggest for competition based on quality. Most health services researchers believe that additional quality comes with some cost. At the same time, little evidence exists to indicate that consumers can recognize quality differentials in healthcare. In fact, economists argue that in the absence of other information, consumers sometimes interpret price differentials as indicators of quality differentials because the latter are so difficult to observe. However, when price differentials result from changes in price and employer premium contribution policies, consumers usually have some experience with their health plan and may believe that they know something about the quality of care delivered there, in which case price changes are less likely to be interpreted as signals of change in quality. If the latter is true and plans are as vulnerable to small price differentials as our research suggests, then any quality improvements that increase cost will stand little chance of being introduced. Any competition based on quality will require both better measures of quality and a better understanding of its costs.
ACKNOWLEDGMENTS
The authors gratefully acknowledge programming assistance from Ellen R. Harrison. We also thank Harold Luft, Stephen Shortell, and two anonymous referees for helpful comments.
NOTES
1. This is really an issue for large employers. Here we are referring to firms with 200 or more employees. Sixty percent of these firms offer multiple plans (Hunt, Singer, Gabel, et al. 1997). Small employers are less likely to offer health insurance at all, and when they do, multiple plan offerings are rare (Nichols, Blumberg, Acs, et al. 1997). Data from a study of ten states indicate that among firms with fewer than 50 employees, only 14 percent offer a choice of plans (Cantor, Long, and Marquis 1995).
2. The risk parameter in our model does not vary with family income or other characteristics of the family. In the empirical work from which our parameters derive, Marquis and Holmer (1996) investigated alternative specifications in which the risk parameters were allowed to vary with family attributes. However, they did not find significant differences in risk aversion. The standard deviation of the E(w) is given by 1/[Beta], where [Beta] varies with the family income. See Marquis and Holmer (1996) for the specific functional form; the average value in our sample was .01.
3. Recent work indicates that employee participation rates may be declining (Cooper and Schone 1997). Nonetheless, even this work shows that about 85 percent of employees in large firms participate in their employer's group plan.
4. The individual characteristics in our fitted model are limited to measures that are also available in the Current Population Survey.
5. That is, we predict the level of spending in a plan with comparable benefits and then apply the discount. The RAND Health Insurance Experiment found that spending by persons in an HMO was almost 30 percent lower than spending in a fee-for-service plan with the same benefit structure (Newhouse and the Insurance Experiment Group 1993). The Congressional Budget Office (1993) estimated the price of HMO plans to be 10-15 percent below that of similar fee-for-service plans. We have adopted the average of these estimates.
6. We are investigating choices among employees in a large firm and so have adopted an average loading fee that is observed for groups of 500 and more (U.S. Library of Congress 1988).
7. The assumption that individuals do not perceive that their choice of insurance plan will affect their use of services was also made in the empirical work from which the parameters of the utility function derive, that is, it assumed that the value of insurance comes only from the risk avoided. If the value of the expected additional care is a part of families' evaluation of alternative plans and is incorporated in the responses of the HIE participants about the hypothetical plans, then the estimate of the risk parameter reflects both risk aversion and the expected value of additional care. That is, given the assumed functional form, the model estimation could not disentangle the value of risk avoided and the value of the additional medical consumption.
8. Based on data from the 1989 survey of firms by the Health Insurance Association of America.
9. The contribution for the PPO, C(PPO), is then calculated as: N(PPO)* C(PPO) + N(HMO) * C(PPO) * [R(HMO/R(PPO)] = C(Ave) * N, where N(PPO) and N(HMO) denote the expected enrollments in each plan (from the prior period); N is the total enrollment in both plans; C(Ave) is the average contribution amount the employer budgeted; and R(PPO) and R(HMO) are the risks in each plan. The contribution amount for the HMO, C(HMO), is then C(HMO) = C(PPO) * [R(HMO)/R(PPO)]. Separate calculations are made for each family composition unit for which premiums are established.
10. The contribution for the PPO, C(PPO), is then calculated as: C(PPO) = C(Ave) + (PPO actual premium - PPO adjusted premium) where the PPO adjusted premium = PPO actual premium/[R(PPO)/Group Risk]. The HMO contribution is established to maintain the budgeted, average employer contribution.
11. Tessler and Mechanic (1975); Wolfman (1961); Friedman, LaTour, and Hughes (1984).
12. This finding was also reported in other simulation analyses based on similar models from the RAND Health Insurance Experiment (Marquis 1992).
13. In our model, the result is driven by the choices of families of two or more. Virtually all of them prefer the HMO under the fixed percent of premium policy as the employer contribution increases, whereas they choose the PPO plan under the ratio method.
14. The premiums shown in the table reflect only the monetary cost; they do not include the cost of restricted choice.
15. We investigated several alternative assumptions about the relationship between the value of choice and risk. When those in families with risk above the 25th percentile value choice, we obtain results similar to the case in which all families value choice. When those with risk about the 75th percentile value choice, we obtain solutions as when those above the median value choice. And when those with risk above the 90th percentile value choice, the results are similar to those when choice is not a valued attribute.
REFERENCES
Bowen, B., and E. Slavin. 1991. "Adjusting Contributions to Address Selection Bias." In Advances in Health Economics and Health Services Research, Vol. 12 edited by R. M. Scheffler and L. E Rossiter, pp. 77-96. Greenwich, CT: JAI Press.
Buchanan, J. L., E. B. Keeler, J. Rolph, and M. Holmer. 1991. "Simulating Health Expenditures Under Alternative Insurance Plans." Management Science 37 (9): 1067-90.
Buchanan, J. L., and M. S. Marquis. 1999. "Who Gains and Who Loses with Community Rating for Small Business?" Inquiry 36, no. 1 (spring): 30-43.
Buchmueller, T. C., and P. J. Feldstein. 1997. "The Effect of Price on Switching Among Health Plans." Journal of Health Economics 16 (3): 231-47.
Cantor, J. C., S. H. Long, and M. S. Marquis. 1995. "Private Employment-based Health Insurance in Ten States." Health Affairs 14 (summer): 199-211.
Cave, J. 1985. "Equilibrium in Insurance Markets with Asymmetric Information and Adverse Selection." In Advances in Health Economics and Health Services Research, Volume 6, edited by R. M. Scheffler and L. F. Rossiter. Greenwich, CT: JAI Press.
Congressional Budget Office. 1993. Managed Competition and Its Potential to Reduce Health Care Spending. Washington, DC: U.S. Government Printing Office. May.
Cooper, P. F., and B. S. Schone. 1997. "More Offers, Fewer Takers for Employment-based Health Insurance: 1987 and 1996." Health Affairs 16 (6): 142-49.
Cutler, D. M., and S. Reber. 1996. Paying for Health Insurance: The Trade-off Between Competition and Adverse Selection. Cambridge, MA: National Bureau of Economic Research, Working Paper 5796.
Cutler, D. M., and R. J. Zeckhauser. 1997. Adverse Selection in Health Insurance. Cambridge, MA: National Bureau of Economic Research, Working Paper 6107.
Ellwood, P. M., A. C. Enthoven, and L. Etheridge. 1992. "The Jackson Hole Initiatives for a Twenty-first Century American Health Care System." Health Economics 1 (3): 149-68.
Enthoven, A. C. 1990. "Multiple Choice Health Insurance: The Lessons and Challenge to Employers." Inquiry 27, no. 4 (winter): 368-75.
Enthoven, A. C., and R. A. Kronick. 1989. "A Consumer Choice Health Plan for the 1990s: Universal Health Insurance in a System Designed to Promote Quality and Economy." The New England Journal of Medicine 320 (1): 29-37.
Farley, P., and G. R. Wilensky. 1985. "Household Wealth and Health Insurance as a Protection Against Medical Risks." In Horizontal Equity, Uncertainty, and Economic Well-Being, edited by M. H. David and T. M. Smeeding. Chicago: University of Chicago Press.
Feldman, R., and B. Dowd. 1993. "The Effectiveness of Managed Competition: Results from a Natural Experiment." Washington, DC: American Enterprise Institute, Conference Paper (Health Care Expenditure Controls: Political and Economic Issues).
Friedman, B., S. A. LaTour, E., and E. F. X. Hughes. 1984. "A Medicare Voucher System: What Can It Offer?" In Proceedings on the Conference on the Future of Medicare, pp. 55-78. Subcommittee on Health of the Committee on Ways and Means, U.S. House of Representatives, Washington, DC. (1 February).
Holmer, M. 1984. "Tax Policy and the Demand for Health Insurance." Journal of Health Economics 3 (3): 203-21.
Hunt, K. A., S. J. Singer, J. Gabel, D. Lison, and A. C. Enthoven. 1997. "Paying More Twice: When Employers Subsidize Higher-Cost Health Plans." Health Affairs 16 (6): 150-56.
Jones, S. B. 1990. "Multiple Choice Health Insurance: The Lessons and the Challenge to Private Insurers." Inquiry 27 (1): 161-66.
Keeler, E. B., J. L. Buchanan, J. E. Rolph, J. M. Hanley, and D. M. Reboussin. 1988. The Demand for Episodes of Medical Treatment in the Health Insurance Experiment. R-3454-HHS. Santa Monica, CA: RAND.
Keeler, E. B., and J. E. Rolph. 1982. The Demand for Episodes of Medical Service: Interim Results from the Health Insurance Experiment. R-2829-HHS. Santa Monica, CA: RAND.
Long, S. H., and M. S. Marquis. 1993. "Gaps in Employer Coverage: Lack of Supply or Lack of Demand?" Health Affairs 12 (Supplement): 282-93.
Manning, W. G., J. P. Newhouse, N. Duan, E. B. Keeler, A. Leibowitz, and M. S. Marquis. 1987. "Health Insurance and the Demand for Medical Care: Evidence from a Randomized Experiment." American Economic Review 77 (3): 251-77.
Marquis, M. S. 1992. "Adverse Selection with a Multiple Choice Among Health Insurance Plans: A Simulation Analysis." Journal of Health Economics 11 (1): 129-51.
Marquis, M. S., and J. L. Buchanan. 1994. "How Will Changes in Health Insurance Tax Policy and Employer Health Plan Contributions Affect Access to Health Care and Health Care Costs?" Journal of the American Medical Association 271 (12): 939-44.
-----. 1992. "Subsidies and National Health Care Reform: The Effect on Workers' Demand for Health Insurance Coverage." In Health Benefits and the Workforce. Washington, DC: U.S. Department of Labor, Pension and Welfare Benefits Administration.
Marquis, M. S., and M. R. Holmer. 1996. "Alternative Models of Choice Under Uncertainty and Demand for Health Insurance." The Review of Economics and Statistics 78 (3): 421-27.
Marquis, M. S., and S. H. Long. 1995. "Worker Demand for Health Insurance in the Non-group Market." Journal of Health Economics 14 (1): 47-63.
Marquis, M. S., and J. R. Rogowski. 1991. Participation in Alternative Health Plans: The Role of Financial Incentives in Medicare Beneficiaries' Decisions. R-4105-HCFA. Santa Monica, CA: RAND.
Newhouse, J. P., and the Insurance Experiment Group. 1993. Free For All? Lessons Learned from the RAND Health Insurance Experiment Cambridge, MA: Harvard University Press.
Nichols, L. M., L. J. Blumberg, G. P. Acs, C. E. Uccello, and J. A. Marsteller. 1997. Small Employers: Their Diversity and Health Insurance. Washington, DC: Urban Institute.
Tessler, R., and D. Mechanic. 1975. "Factors Affecting the Choice Between Prepaid Group Practice and Alternative Health Insurance Programs." Milbank Memorial Fund Quarterly/Health and Society 53, no. 2 (spring): 149-72.
U.S. Library of Congress, Congressional Research Service. 1988. Costs and Effects of Extending Health Insurance Coverage. Report prepared for the House Committees on Education and Labor and Energy and Commerce and the Senate Special Committee on Aging (Education and Labor Serial No. 100-EE). Washington, DC.
Wolfman, B. 1961. "Medical Expenses and Choice of Plans: A Case Study." Monthly Labor Review 84 (11): 1186-90.
Address correspondence and requests for reprints to M. Susan Marquis, Ph.D., Senior Economist, RAND, 1333 H Street, N.W., Suite 800, Washington, DC 20005. Joan L. Buchanan, Ph.D. is Senior Researcher, Department of Health Care Policy, Harvard Medical School. This article, submitted to Health Services Research on January 8, 1998, was revised and accepted for publication on September 21, 1998.
COPYRIGHT 1999 American College of Healthcare Executives
COPYRIGHT 2000 Gale Group