Objective: To determine the sensitivity of employees’ health insurance decisions - including the decision to not choose health maintenance organization or fee-for-service coverage - during periods of rapidly escalating healthcare costs.
Study Design: A retrospective cohort study of employee plan choices at a single large firm with a "cafeteria-style" benefits plan wherein employees paid all the additional cost of purchasing more generous insurance.
Methods: We modeled the probability that an employee would drop coverage or switch plans in response to employee premium increases using data from a single large US company with employees across 47 states during the 3-year period of 1989 through 1991, a time of large premium increases within and across plans.
Results: Premium increases induced substantial plan switching. Single employees were more likely to respond to premium increases by dropping coverage, whereas families tended to switch to another plan. Premium increases of 10% induced 7% of single employees to drop or severely cut back on coverage; 13% to switch to another plan; and 80% to remain in their existing plan. Similar figures for those with family coverage were 11%, 12%, and 77%, respectively. Simulation results that control for known covariates show similar increases. When faced with a dramatic increase in premiums - on the order of 20% - nearly one fifth of the single employees dropped coverage compared with 10% of those with family coverage.
Conclusions: Employee coverage decisions are sensitive to rapidly increasing premiums, and single employees may be likely to drop coverage. This finding suggests that sustained premium increases could induce substantial increases in the number of uninsured individuals.
(Am J Manag Care. 2004;10:41-47)
The rapid rise in health insurance premiums over the last 2 years raises questions about what will happen to the employer-provided insurance market. Because most Americans obtain health insurance through the workplace, a concern is that employees and their dependents will end up without coverage or with much less adequate coverage. Whereas some small firms will almost certainly respond to rapid premium increases by dropping coverage for all employees, large firms are unlikely to do so because virtually all offer coverage and have been doing so for years. Rather, large firms are likely to require employees to pay a larger portion of their health insurance premiums. The percentage of covered workers whose employers pay the full cost of single coverage declined from 30% in 2001 to just 23% 1 year later.1 This fact raises the question of how employees respond to these price changes - will they switch to less generous coverage or forego coverage completely?
Previous studies have examined the demand response to premium-sharing arrangements.2-5 For example, in a 1984 study of the health plan choices made by employees in 20 Minneapolis firms, Feldman and colleagues2 estimated nested logit models of insurance coverage in firms offering health maintenance organization (HMO) and fee-for-service (FFS) insurance. They found that employee choices were sensitive to out-of-pocket premiums and that employees choosing HMOs were more sensitive to price. This result may reflect the status quo in 1984, when HMOs were generally the lower cost option. Today, HMOs are not necessarily the least expensive option; more recent work by Feldman and associates6 indicated that firms that offer HMOs do not have lower healthcare costs than companies that do not.
Our goal was to examine health plan choices within a flexible benefit plan. Firms offering these plans give employees a fixed benefit allocation. Employees then decide how to allocate these "credits" between health benefits, pensions, life insurance, or other benefits. Funds not spent on health insurance can be used to purchase other benefits or increase take-home earnings; thus, employees pay the full marginal cost of electing a more expensive health plan. Such cafeteria-style plans cover 13% of workers in medium and large firms, and the proportion is growing; thus these plans are interesting to study in their own right.7
Our analysis was similar in spirit to that of Buchmueller and Feldstein,8 who had examined a policy change by the University of California that capped employer contributions at the cost of the least expensive health plan offered. They found that premium increases induced high rates of plan switching. Buchmueller and Feldstein speculated that the well-publicized policy switch may have influenced their results. Cutler and Reber9 also examined demand response to a substantial benefits policy change at Harvard University and found similar large effects. Our work complemented findings from both studies by confirming this price sensitivity using multiyear data, although the firm in question did not experience any dramatic change in its compensation policy.
Our data came from a single large company in the United States that offered a flexible benefits plan. Employees paid out-of-pocket for the difference in premiums between the chosen plan and a low-cost catastrophic health insurance policy. Employees paid the additional premium cost on a pretax basis. We acquired data on employees' health plan decisions from 1989 through 1991, allowing us to examine how changes in relative prices affected plan choices over 3 years. The advantage of this period was that it was a time when firms experienced premium increases well above the rate of inflation—similar to the rapid premium growth from 2000 through 2002 (for which data were not available).
Modeling Plan Changes
In modeling health insurance choices, we focused on the probability that an individual would choose an alternative in year t + 1 that differed from the one chosen in year t. The responsiveness of demand could then be measured by looking at how this probability changed with price changes. Unfortunately, it is not entirely clear in a reduced-form analysis exactly what the appropriate measure of price should be. Economists working with firm-level data have used variation in employer contributions, tax rates, loading fees, or standardized policies to proxy for price changes.10-13 Because our data was from a single firm, we could not exploit this type of variation. Rather, we hypothesized that the choice of a health plan was a function of the relative premiums within the set of feasible plan alternatives, similar to the assumption of Long and colleagues.13
As with many flexible benefits plans, the employer provided a partial subsidy to the purchase of health insurance. In this case, the employer paid the full cost of a catastrophic FFS plan. If employees elected a more generous FFS health plan with a lower deductible or any of the 50 HMOs offered, they were required to pay the additional premium cost either through credits or payroll deductions. We used data from the 3-year period of 1989 through 1991. While these data are old, this period has the advantage that it was a time of large premium increases, both within and across plans. It also is comparable to periods covered in other studies.
Multinomial Logit Model of Choice. The multinomial logit (MNL) model of choice requires the well-known condition of the independence of irrelevant alternatives. The independence of irrelevant alternatives is tantamount to assuming that the stochastic portions of the conditional utility functions are uncorrelated across alternatives, and imposes the restriction that the crossprice elasticities are the same across all alternatives.14 One alternative to the MNL is the nested MNL. This model allows for correlation across subgroups of alternatives (closer substitutes) so that price elasticities are more elastic within groups than across groups. Unfortunately, the structure of our dataset did not allow for the estimation of such a model, since we did not have any right-hand side variables except premium that varied across choices. The right-hand side variable was actually the "increase or decrease in cost associated with the status quo plan" - ie, the plan chosen in the previous period. Thus our ability to draw certain inferences was limited; for example, we could not infer what would happen if another plan were added to the system.
We first estimated an MNL model in which each employee at time t faced 3 choices: (1) keep the plan that he or she chose in time t-1; (2) switch plans; or (3) choose the free option, which means dropping health insurance or switching to the catastrophic FFS plan. (Although the catastrophic plan with individual coverage was always a free alternative, approximately 11% of the sample chose no insurance. This choice may have been related to coverage through a family member or partner outside the firm. Unfortunately, we did not have any information on alternate sources of insurance. Therefore, we treated the decisions of dropping health insurance or switching to the catastrophic option as equivalent in this report.)
We assumed that the probabilities of undertaking each of these actions was a function of the change in the relative premium, controlling for a limited set of covariates, including total compensation, age, sex, tenure in the job, marital status, plan and state dummies, and possibly some interactions. We included compensation on the job because higher-income employees may be less responsive to price changes and because the deductible in the catastrophic plan was equal to 5% of salary. We hypothesized that workers with greater job tenure may have had more inertia about changing plans, and that workers with greater expected use of healthcare (older workers, women, or workers with families) may have been less likely to change plans if they had to change providers, as a change between HMO and FFS plans might require. Plan dummies were included to capture the relative attractiveness of individual plans and state dummies were included to account for systematic, unmeasured geographic differences.
Our measure of price was based on relative premiums, defined as the premium of an individual's plan divided by the average for all competing plans in that market. It is useful to note that the actual change in an individual's expenses could deviate from this amount if he or she switched plans; hence, we used the term "incipient premium increase" to describe this variable.
The data consisted of 3 years (1989-1991) of earnings and benefit information for 14,221 employees at a single US company. Not all employees worked all years, so the entire dataset consisted of 31,907 employee-years of data. After eliminating employees aged 65 years or older, and a few observations with inconsistent or incomplete data, the employees were geographically dispersed across 47 states, with most living in California, New Jersey, or Texas. Table 1 presents descriptive statistics across all employee years. Employees were on average 37 years old with 6 years of experience with the company; two thirds were female.