Printer Friendly

The effect of HMOs on premiums in employment-based health plans.

Jensen, G. A., "Employer Choice of Health Insurance Benefits." Unpublished Ph.D. diss., University of Minnesota, Department of Economics, March 1986.

Jensen, G., R. Feldman, and B. Dowd. "Corporate Benefit Policies and Health Insurance Costs." Journal of Health Economics 3, no. 3 (Winter 1984): 275-96.

Jensen, G. A., M. A. Morrisey, and J. W. Marcus. "Cost Sharing and the Changing Pattern of Employer-Sponsored Health Benefits." Milbank Quarterly 65, no. 4 (1987): 521-50.

Health maintenance organizations (HMOs) are an integral part of the competitive health care strategy that relies on competition to improve the performance of the health care system. Reliance on HMOs rests on three propositions: first, that HMOs reduce the use and cost of health care for their own enrollees; second, that the premium for HMO insurance coverage accurately reflects the lower utilization and cost of HMO enrollees; and third, that the competitive effect of HMOs will force other health plans to make similar improvements in health care use and cost.

Most researchers accept the first proposition. Early studies reviewed by Luft (1981) suggest that HMO hospitalization rates are about 30 percent lower than those of conventionally insured populations. A controlled trial (Manning, Leibowitz, Goldberg, et al. 1985) showed that persons randomly assigned to a prepaid group practice (PGP) had about 40 percent fewer hospital admissions than persons who received free fee-for-service (FFS) medical care. Estimated expenditures for all services were about 25 percent lower in the PGP than in the FFS alternative. In a study that used econometric techniques to correct for voluntary enrollment in HMOs, we found that employees who chose HMOs had about 30 percent fewer hospital days than a comparable voluntarily enrolled FFS population (Dowd et al. 1991). Use of physicians' services was about the same in the two populations.

Evidence from market level research is much less supportive of the third proposition of the competitive hypothesis. McLaughlin (1987) estimated that the growth of PGP market shares in 25 large metropolitan areas did not lead to a statistically significant reduction in overall hospital expenses per capita. Zellner and Wolfe (1989) have argued that this finding may be biased due to specification error.

Despite uncertainty surrounding the market level effects of HMOs, it would still appear that firms can reduce their health insurance costs by offering HMOs. That is, the 30-40 percent utilization savings from HMOs should result in lower total health insurance premiums for the employer and employees. However, even if HMOs represent a more efficient medical care delivery system than FFS medicine, it does not follow that HMOs will satisfy the second proposition of the competitive hypothesis. This point has been virtually overlooked in the debate on competitive health care strategy. An exception is an article by Dowd and Feldman (1985), who pointed out that employers and employees will not benefit from superior HMO efficiency unless the HMO's premium is driven down to equal its marginal cost. Competition among health plans must be present in order for this to occur.

Dowd and Feldman drew attention to the fact that competition among health plans usually occurs within a specific context -- the market for employment-based health insurance. HMO enrollment almost always occurs through an employment-based health plan. Therefore, if HMOs reduce health care costs, evidence of the savings should appear in lower health care premiums paid by employers and employees, holding constant other factors, such as generosity of coverage. Despite the importance of this form of competition, there is almost no evidence that HMOs reduce health insurance premiums in firms where they are offered. Lack of research in this area is surprising, given that employers are concerned with the costs of health insurance in general and that many employers have reported poor experience with HMOs.

Honeywell, Inc., for example, concluded that its health care costs increased by an extra $4-5 million per year after HMOs gained a 70 percent enrollment share in the firm (Lee 1985). Many companies accuse HMOs of "shadow pricing," that is, setting their premiums just below those of commercial carriers. HMOs can profit from shadow pricing if they tend to enroll a disproportionate share of young, healthy workers in the firm. One New York company that was currently paying $2,000 per year for each employee's medical bills accepted an HMO's offer to provide care for $1,400 per person. But overall costs went up as young workers flocked to the HMO. The average employee joining the HMO was 25 years old and had been submitting only $400 in claims per year (Wall Street Journal August 11, 1988).

Some favorable selection can be anticipated in setting premiums. For example, J. C. Penney reduced its HMO premiums after discerning that its HMO enrollees were younger than those who elected the company's indemnity plan (Lee 1985). However, the HMO premium can be set lower than the FFS premium without eliminating HMO profits if the HMO enjoys a large selection advantage. As Luft (1985) has noted, the difference between cost and price is at the heart of employers' dissatisfaction with HMOs.

These examples from employers typically are collected by private health care consultants in the course of advising their business clients. The data often are confidential and the examples usually are limited to case studies of particular companies. This article attempts, by statistically analyzing a large data set, to document the assertion that HMOs reduce premiums in employment-based health plans. Sections I and II present our theoretical model of HMO premium pricing and the methods needed to estimate the model. Section III describes the data. This is followed in Section IV by our empirical specification of the model. Estimates of the equations are presented in Section V. Section VI calculates the effect of HMOs on premiums and summarizes the results. Section VII is the concluding discussion.

I. THEORETICAL MODEL

HMOs reduce average premiums in employment-based health plans if the FFS premium in a randomly selected firm that offers only a FFS plan is greater than the weighted average premium in a similar firm that also offers an HMO. This condition can be written as:

E(|P.sub.F~/H = 0) |is greater than~ E(|P.sub.H~ * H + |P.sub.F~ * (1 - H)/H |is greater than~ 0) (1)

where |P.sub.F~ = FFS premium, |P.sub.H~ = HMO premium, and H = HMO enrollment share.(1) We require a theory about the way premiums and market shares for HMOs and FFS health plans are determined. To develop that theory fully, and to correct for the nonexperimental nature of our data, we will need to explain why some firms offer HMOs and others do not.

Our hypothesis is that HMOs maximize expected profit (the excess of expected revenue over expected cost) in any firm. This is equivalent to maximizing expected profit per employee since firm size is assumed to be fixed. Expected profit per worker is the product of the probability of offering the HMO and profit per worker once the HMO is offered:

E(|Pi~) = PROB * H * (|P.sub.H~ - |C.sub.H~) (2)

where PROB is the probability of offering the HMO and |C.sub.H~ is the HMO's average cost.

Suppose that the probability of being offered was 1.0, provided that the HMO wanted to be offered. Then the HMO would concentrate only on the difference between its premium and average cost, that is, the expression in parentheses in equation 2. If |P.sub.H~ was greater than |C.sub.H~ for at least one positive value of enrollment share, then the HMO would mandate that it be offered. Furthermore, once the HMO was offered, economic theory predicts that it would set its premium to equalize marginal revenue and marginal cost per employee.

In contrast to this simple model, the HMO's expected profit function is affected by rules that spell out conditions under which a firm must offer an HMO. These rules are contained in the federal Health Maintenance Organization Act and subsequent regulations that require a firm to offer an HMO if it employs 25 persons or more and the HMO requests to be offered.(2) The regulations do not require the firm to offer all HMOs, however, and if more than one HMO of a particular type is available in the market area, the firm has the option of selecting which of the eligible HMOs will be offered. Therefore, the probability that a firm with more than 25 employees will offer an HMO is identical to the probability that at least one HMO in the market area wishes to be offered. The probability that a firm with fewer than 25 employees will offer an HMO is the probability that at least one HMO wants to be offered and the employer is willing to offer that HMO.

Thus, HMOs face two distinct demand functions. The first demand function is the employer's determination of whether or not the HMO will be offered to the firm's employees. A higher premium reduces the probability of being offered, other things equal. The only exception is a market area with just one HMO. The second demand function is the employees' demand for a specific health plan, chosen from the menu of plans offered by the employer. A higher premium will reduce the HMO's enrollment share in the firm but may increase profits. In setting its premium, the HMO must take both demand functions into account. It must balance the potential gain in profit, if it is offered, against the increased risk that an employer will select another plan.(3)

The risk that the employer will offer another health plan is influenced by a variable that we call "employer shopping." This denotes the conditions that the employer lays down for health plans to gain entry into the firm. It also refers to the firm's aggressiveness in negotiating premiums and other conditions with the plans that are offered. Firms that shop aggressively will be willing to replace one HMO with another HMO subject to the federal requirement that they must offer at least one HMO. These firms also will replace a traditional insurer with a self-insured arrangement if they believe the FFS premium is too high. Other firms, however, may find that their employees have a strong preference for a particular health plan. This could occur because the employees are loyal to the health care providers affiliated with that plan. These employers cannot drop the preferred health plan without offending their employees. Labor unions may also have strong preferences for certain types of health plans. The firm must consider recruitment and retention of employees in selecting the efficient level of pay (Dowd and Feldman 1987).

If the employer lays down stricter terms for gaining entry, health plans will find that a small premium increase will cause a large drop in the probability that they will be offered to the employees. Aggressive employer shopping, therefore, can result in competition among HMOs to enter the firm. Such competition may drive the probability of entry low enough to cause all HMOs to lose interest in the firm, particularly if an HMO's total enrollment capacity is limited and there are more lucrative opportunities available elsewhere in the form of less aggressive employers.

Finally, we hypothesize that firms will be more likely to offer HMOs if their work force has diverse preferences for different health plans. A firm with a heterogeneous work force may be able to improve the welfare of some employees if it offers an HMO. Considering all factors involved, we hypothesize that the probability of offering an HMO depends on HMO premiums; employer shopping (SHOPPING); characteristics of the employer and employees that measure loyalty to particular health plans (LOYALTY); and diversity of employee preferences (DIVERSITY):

PROB = PROB(|P.sub.H~, SHOPPING, LOYALTY, DIVERSITY) (3)

Once offered to employees, the condition for HMO profitability is that |P.sub.H~ exceeds average cost for at least one positive value of enrollment share. The average cost expression in our model is straightforward. We hypothesize that the HMO's average cost in a firm of fixed size is related to exogenous |X.sub.H~ variables that affect employees' demand for covered HMO services. Average cost also rises as the HMO enrolls a larger share of the firm's employees. Numerous studies have shown that HMOs tend to attract young, healthy workers away from the FFS plan.(4) As the HMO's enrollment share grows, however, its enrollees will tend to become more representative of the average worker and this will raise average costs. Thus, average cost can be written as:

|C.sub.H~ = |C.sub.H~(H, |X.sub.H~) (4)

Next, we consider the employees' demand for HMO enrollment. This is a function of differences in characteristics of health plans that are offered to employees, including the difference in the employee's out-of-pocket premium expense (Feldman et al. 1989; Short and Taylor 1989). Since many employers pay a fixed premium contribution, the difference in the employee's out-of-pocket premium is often equal to the difference in total premiums (Feldman and Dowd 1982). Thus, we hypothesize that the demand for HMO enrollment depends on the difference in total premiums between the HMO and the FFS plan. Using the inverse demand notation in which premium is the dependent variable, we can write:

|P.sub.H~ = |P.sub.H~(|P.sub.F~, H, SHOPPING, LOYALTY, DIVERSITY, HMO CHARACTERISTICS) (5)

To the extent that HMOs and FFS plans are substitutes for each other, an increase in the FFS premium will increase the demand for HMO coverage. Therefore, we expect that the coefficient of |P.sub.F~ in the inverse HMO demand function will be positive. The coefficient of HMO enrollment share will be negative if the demand curve slopes downward, as predicted by economic theory.

The other variables in the HMO demand equation capture different types of demand-shift effects. Employer shopping will reduce the premium the HMO can charge for any given level of enrollment. On the other hand, greater employee loyalty, more diversity of preferences, and favorable HMO policy characteristics should increase the premium. These demand-shift factors will be explained in greater detail in the succeeding section.

Next, we consider the premium for the FFS plan. Most FFS plans are "experience rated," that is, their premiums are based on the cost of covered services used by employees in each firm plus administrative expenses.(5) Many experience-rated firms have "administrative services only" (ASO) contracts with firms that do claims processing, utilization review, and other services. We assume that employers can reduce the charges of ASO contractors, and thus premiums for the FFS plan, by shopping aggressively. Employers dissatisfied with available ASO contracts can fall back on their own administrative staff to manage the FFS plan.

Despite the assumption that FFS premiums are driven by costs of covered services plus a markup, the FFS premium is nevertheless endogenous in our model. This occurs because the costs of covered services depend on which employees enroll in the FFS plan. As the HMO market share rises, the average health status of remaining FFS employees is assumed to worsen, thus making the FFS premium depend on HMO enrollment share. Considering all factors, if we let |X.sub.F~ stand for characteristics of the firm, employees, and the FFS insurance policy that determine the cost of covered FFS services and the administrative charge of the FFS insurance policy, we can write the FFS premium equation as:

|P.sub.F~ = |P.sub.F~(H, SHOPPING, |X.sub.F~) (6)

Our model thus far consists of equations 2 through 6. To determine the profit-maximizing HMO enrollment and premium, we substitute equations 3 through 6 into equation 2 to obtain an equation for expected profit in terms of one decision variable, the HMO's enrollment share, H:

E(|Pi~) = PROB{|P.sub.H~||P.sub.F~(H), H~} * H * {|P.sub.H~||P.sub.F~(H),H~ - |C.sub.H~} (7)

For simplicity and in order to highlight the pathways by which H influences expected profit, we have omitted the exogenous variables from this equation. The marginal condition for maximizing expected profit is:

dE(|Pi~)/dH = dPROB/dH * H * (|P.sub.H~ - |C.sub.H~) + PROB * |(|P.sub.H~ - |C.sub.H~) + H * (d|P.sub.H~/dH - d|C.sub.H~/dH)~ = 0 (8)

The derivative dPROB/dH in equation 8 is positive, because the HMO can increase both its enrollment and the chance of being offered by cutting its premium. Next, we note that the HMO's premium must exceed its average cost if the HMO wants to be offered. Therefore, the first term in equation 8 is positive. It follows that the second term must be negative when the HMO is maximizing profits. The fact that this term is the difference between marginal revenue and marginal cost means that the HMO reduces its premium compared with the premium it would charge if there was no competition to enter the firm.

The solution to equation 8 determines the HMO's profit-maximizing enrollment share in the firm:

H = H(LOYALTY, DIVERSITY, HMO CHARACTERISTICS, |X.sub.F~, |X.sub.H~) (9)

Measures of employer shopping are omitted from this equation because we assume that the employer is equally aggressive in dealing with all health plans. For example, an employer that negotiates with FFS plans over administrative costs is also assumed to negotiate with HMOs over their administrative costs. This assumption implies that the difference in the employee's out-of-pocket premiums for each pair of health plan alternatives is unaffected by aggressive shopping. Thus, the HMO's enrollment share is not affected by employer shopping.

To summarize, the purpose of this article is to evaluate inequality 1, which is a function of the HMO and FFS premiums (|P.sub.H~ and |P.sub.F~) and the HMO enrollment share (H). The HMO will want to be offered if it can make a profit in the firm, and it is assumed to choose enrollment to maximize expected profits. Since a higher premium reduces the probability of being offered, the HMO will underprice its product, compared with a situation in which entry is guaranteed. The extent of underpricing depends on the firm's aggressiveness in shopping for other health plans. Nevertheless, in equilibrium the HMO premium will still exceed the competitive level, unless either the employer's or the employees' demand function is perfectly price-elastic.

In order to evaluate inequality 1 we need to estimate values of |P.sub.H~, |P.sub.F~, and H conditional on the firm offering and not offering HMOs.

For example, we need to estimate the value of |P.sub.H~ if a firm not currently offering an HMO begins to do so. We must also estimate values of |P.sub.F~ and we cannot assume that the equation predicting |P.sub.F~ in firms that offer HMOs will yield accurate predictions of |P.sub.F~ for firms presently not offering HMOs. These estimation problems are discussed in the next section.

II. ESTIMATION METHODS

Evaluation of inequality 1 requires estimates of |P.sub.H~, |P.sub.F~, and H. Estimation of these equations is complicated by two problems. First, since we assume that the HMO is a profit maximizer facing two demand curves, |P.sub.H~ and H are jointly determined and estimating their values invokes the usual problem of simultaneous equations bias. The FFS premium, |P.sub.F~, also is endogenous because it is influenced by |P.sub.H~ and H.

An additional problem arises from the conditional nature of the three equations. The HMO premium and market share equations are observed only if the firm offers an HMO. The |P.sub.F~ equation is observed for all firms, but it may be affected by unobserved variables that also affect the firm's decision to offer an HMO. For example, workers in firms offering an HMO may be more or less healthy than in firms without HMOs. Consequently, FFS premiums in firms now offering an HMO may be a poor predictor of |P.sub.F~ in firms not offering HMOs, or |P.sub.F~ for a randomly selected firm.

We used Heckman's (1976) two-step selection correction to obtain unbiased estimates of these equations. The first step utilizes probit analysis to predict whether or not an HMO will be offered. Selection terms constructed from the probit equation are included in the premium and HMO enrollment share equations. This method produces unbiased coefficient estimates for a randomly selected firm. From these selectivity corrected equations we calculate the components of inequality 1 that will reveal the effect of HMOs on health insurance premiums in the firm.

III. DATA

Our data were taken from a survey of Minnesota employers conducted during 1986. The universe for the survey was 106,908 firms that paid unemployment tax in Minnesota. We restricted the survey to firms with five or more employees. Firms smaller than this are often highly transient or seasonal and employ only 5.6 percent of the total work force covered by unemployment insurance. Even within the remaining population, the distribution of firm sizes is highly skewed. Therefore, we selected random samples within the following strata: 1,500 surveys were mailed to firms with 5-49 employees, 1,500 surveys were mailed to firms with 50-249 employees, and all 826 firms with more than 250 employees were surveyed.

The survey was mailed to firms with an endorsement letter from the governor of Minnesota. A reminder letter was mailed two weeks later. Altogether, 1,275 usable responses were received, for a response rate of 33 percent. Although a 33 percent response rate is not ideal, it is substantially higher than rates from the national employee benefits surveys conducted by the National Federation of Independent Business (NFIB) and the Small Business Administration (SBA).(6) Our survey also had an advantage that would be very difficult to duplicate in a national survey: a large proportion of Minnesota employers offer more than one health plan. Of the responding firms, 315 offered multiple health plans (these firms offered a total of 933 different plans). This is a much higher proportion of multiple-plan firms than one would expect to find in a national survey (Jensen, Morrisey, and Marcus 1987).

Nonresponse tests indicated slight overreporting among the largest firms and slight underreporting among the smallest firms. Government employers tended to respond more often than private firms, but otherwise the distribution of survey respondents was similar to the known distribution of Minnesota firms by industry type. Some of the government agencies were independent school districts, which responded well to the survey.

To the extent that Minnesota data could be compared against the national NFIB and SBA surveys, Minnesota employers offering insurance were as satisfied with their health insurance plans as national employers were. Minnesota employers that did not offer health insurance often cited high premiums as the reason, as well as coverage for their employees under a spouse's policy. These reasons were also cited by firms in the NFIB survey.

The firms that did not offer health insurance (N = 185) were dropped from the study. These firms were smaller than firms with health insurance. They were more likely to be engaged in certain industries (construction, transportation, and trade and services) and located in nonmetropolitan areas. They employed more part-time workers and were less likely to be unionized. Our findings concerning the effects of firm size and part-time workers on provision of health insurance are consistent with an earlier national survey (Rossiter and Taylor 1982).

Some of the remaining firms fell into categories in which premium competition is difficult to describe: 92 firms offered only one HMO; 28 offered only multiple HMOs; and 37 offered only multiple fee-for-service plans. These firms were dropped from the analysis, as were 11 others where we were unable to identify the plans offered as either FFS or HMOs. The remaining usable data set consisted of 683 firms that offered only one FFS plan, and 239 that offered a FFS plan or plans and one or more HMOs. Descriptive statistics for the plans used in our analysis are shown in Table 1.

Although the usable data set may be subject to some response biases (with overrepresentation of large employers and government units), this will not bias the results unless the reasons for nonresponse are correlated with dependent variables of interest. On the other hand, our data set is representative of national findings concerning health insurance offering and employers' attitudes. In addition, because the data were collected within a single state, we did not have to control for interstate differences in health plan regulations and mandated benefits laws.

IV. EMPIRICAL SPECIFICATION OF THE MODEL

HMO Offering Equation. The first task is to estimate an equation where the dependent variable equals 1 if an HMO is offered and 0 otherwise. Equation 3 is the structural form of this equation. However, since equation 3 includes the endogenous HMO premium (|P.sub.H~), we estimated the reduced-form HMO offering equation in which the HMO premium is replaced by exogenous variables. The variables in this reduced form equation are now discussed together with their rationale.

First, firms with more than 25 Minnesota employees should be more likely to offer HMOs than firms with fewer than 25 employees. One HMO can insist that it be offered by a larger firm, whereas among smaller firms both the employer and the HMO must be willing to offer the HMO. A closely related variable indicates whether the greater number of the firm's employees work in the Minneapolis St. Paul metropolitan area (dummy variable = 1 if yes and 0 if no). There are more HMOs in the Twin Cities than in the rest of Minnesota, and if we assume that HMOs have different levels of efficiency, it TABULAR DATA OMITTED is more likely that at least one of them will see an opportunity to profit from entering a Twin Cities firm.(7) Twin Cities location also may affect employers' demand for HMO services. There are more HMO model types in the Twin Cities than in the rest of the state, and HMOs may have a better reputation among employers in the Twin Cities. Thus, it is likely that Twin Cities employers will seek out HMOs to offer their employees.

The firm's primary type of business in Minnesota (represented by eight dummy variables: for trade, construction, manufacturing, mining, transportation, government, agriculture, and finance/real estate relative to service firms) is a proxy for workers' health status. HMOs may be more willing to enter firms with "sicker" workers if these firms have relatively high FFS health care costs and, therefore, high FFS premiums. However, HMOs will shun these firms if poor employee health status primarily affects the HMO's average cost of covered services.

Next, we selected several variables that measure diversity of employees' preferences for different health plans. One of these variables is the firm's total number of employees in Minnesota. Large firms are more likely than small firms to have heterogeneous work forces. The large firm can cater to workers of different sex, age, and income by offering multiple health plans. Transaction costs of offering HMOs should also be lower for large firms. Empirical work by Jensen (1986) has shown that large firms are more likely to offer HMOs. Another indicator of diverse preferences is a dummy variable equal to 1 if retired employees can participate in the firm's health benefits program. Firms with retirees eligible for health insurance should be more likely to offer an HMO. This variable also measures the average health status of all individuals (not just current workers) covered by the health plan.

Several variables in the HMO offering equation measure workers' "loyalty" to particular health plans. An indirect measure of loyalty is the percentage of the firm's Minnesota employees in labor unions (three dummy variables for 1-25 percent, 26-75 percent, and 76 percent or more unionized workers compared to no union). Our hypothesis is that unions prefer the type of coverage, with low cost sharing, offered by HMOs. Jensen (1986) showed that unionization increased the likelihood of offering multiple health plans.

Our survey had four questions dealing directly with employees' loyalty to particular health plans, as it was perceived by the employer. First, the employer was asked to rate the importance of employee preferences for a particular plan in determining the employer's choice of plans to offer. Maintaining employees' ability to choose their own doctor and hospital was the second employee loyalty variable. The importance of low premiums and broad coverage when selecting health plans rounded out the list of employee preference variables. These variables were all coded on a 1-8 scale, with 1 as the most important of eight ranked factors in determining the choice of health plans to offer, and 8 as the least important.

Dowd, Feldman, and Klein (1987) showed that active employer shopping for health plans can lead to lower health plan premiums.(8) Therefore, the next set of variables are three indicators of employer shopping for health plans, a factor that is hypothesized to discourage an HMO from seeking to enter the firm. The variables are these: that the firm has declined to renew a health plan contract within the last five years; the firm shops actively prior to the expiration of the current contract; and the firm negotiates for lower premiums after information about premiums is known. Employer shopping variables were all coded as 1 if the firm exhibited that type of shopping and 0 if it did not shop.

The last variables in the HMO offering equation are exogenous characteristics of the FFS insurance policy. A detailed description of these variables is deferred until we get to the FFS premium equation. We assume, however, that characteristics of the FFS insurance policy are exogenous when the HMO requests entry into the firm. This assumption may not be true in the long run when, for example, the firm changes its FFS plan in response to HMO entry. However, we assume that such changes are not preemptive, but that they occur over time after the HMO enters the firm. Thus, they are exogenous during the period under consideration in this model. We also exclude HMO plan characteristics from the HMO offering equation. These variables are observed if HMOs have actually entered the firm. They are missing for firms that do not have an HMO. These missing variables might be observed if we had data on the best nonselected HMO in the firm's employment area, but this information is missing in our data set and would be very difficult to collect under ideal conditions. Missing variables may bias the coefficients in the HMO offering equation. However, they should not bias the coefficients in the HMO premium equation, which is estimated for the firms where HMO characteristics are observed.

HMO Enrollment Share Equation. The second task is to estimate equation 9, the reduced form equation for HMO enrollment share, using the firms that offer HMOs. The dependent variable is the number of employees enrolled in HMOs divided by the total number of employees covered by all health plans offered by the firm.(9) As noted above, we used Heckman's (1976) two-step method to obtain unbiased estimates of the coefficients in the HMO enrollment share equation. A selection variable, constructed from the HMO offering equation, is included in the HMO enrollment share equation.

With several exceptions, the variables in the HMO offering equation also appear in the HMO enrollment share equation. The exceptions are the dummy variable for more than 25 employees and indicators of employer shopping. Having more than 25 employees means that the firm must offer an HMO if one wants to be offered, but it does not mean that more employees in the "mandated" firm will choose an HMO. Employer shopping is omitted from the HMO enrollment share equation because we assume that it affects the FFS and HMO premiums equally and that, therefore, it should not influence employees' choice of an HMO. The HMO enrollment share equation is identified by these exclusions. However, we also estimate a version of the enrollment equation that includes all of the variables in the HMO offering equation. This version of the enrollment equation is identified only by the nonlinearity of the selection term. We test the hypothesis that the identifying variables are jointly insignificant.

Fee-for-Service Premium Equation. There has been little analysis of the effect of HMOs on fee-for-service (FFS) health plan premiums in firms. The most thorough study to date, by Jensen, Feldman, and Dowd (1984), is based on the experience of large employers in Minnesota. They regressed FFS premiums on measures of benefits, factors related to loading changes, and a dummy variable for whether the firm offered an HMO. They found that offering an HMO was associated with an increase in FFS premiums.

In this research we followed the empirical specification used by Jensen, Feldman, and Dowd. We regressed individual and family plan FFS premiums on measures of benefits, employer shopping, loading charges, industry dummy variables, an instrumental variable for HMO enrollment share (if the firm offered an HMO), and a selection variable. Separate equations were estimated for firms with and without an HMO. The independent variables discussed next were used in these equations.

Several measures of employee cost sharing for covered medical care services were included in the FFS premium equation: the annual deductible, the coinsurance rate (in percent), and the annual maximum out-of-pocket expense. FFS health plans with larger annual deductibles should have lower monthly premiums because enrollees use fewer covered services when they face a larger deductible. The coinsurance rate is the fraction of each dollar of covered services paid by the enrollee after the deductible has been met. Plans with higher coinsurance rates should have lower monthly premiums. Plans with higher annual maximum out-of-pocket expense also should have lower monthly premiums. The family plan premium equations used family deductibles and maximum out-of-pocket expense; the single-coverage equations used the same variables for single coverage. Both equations used the same coinsurance rate variable, since this does not vary by type of coverage in most FFS plans.

A variable closely related to employee cost sharing is whether the plan has a preferred provider arrangement (1 = yes, 0 = no). PPOs use incentives to steer enrollees toward a limited group of providers, who typically agree to accept discounted reimbursement rates. Consumers not using preferred providers pay a higher coinsurance rate on the plan's approved charge and are fully responsible for bills that exceed the approved charge. We expect that plans with a PPO arrangement will have lower premiums than plans without this arrangement.

Measures of active employer shopping for health plans were included in the FFS premium equation to test the hypothesis that firms can obtain lower premiums by shopping for health plan alternatives. These variables, described previously, are shopping actively for health plans before the current contract expires, negotiating with the health plan after receiving its final bid, and switching health plans during the past five years.

We included four variables that may be related to the loading charge for the FFS health insurance premium. The first of these is a dummy variable denoting whether the firm is self-insured (1 = yes, 0 = no). Self-insured firms pay for covered medical care directly, without using an insurance company. Many experts believe that self-insurance should reduce loading charges. Second, there is a dummy variable for firms that offer FFS plans sponsored by Blue Cross (1 = yes, 0 = no). Jensen, Feldman, and Dowd (1984) found that Blue Cross premiums were significantly higher than those of other FFS plans. This may reflect Blue Cross's generous provider reimbursement policies. Third, we included a dummy variable for firms that participate in a group buying arrangement for FFS health insurance (1 = yes, 0 = no). A group purchasing arrangement should enable the firm to buy health insurance at a lower loading charge. Fourth, the size of the group for whom insurance is purchased has been found to exert a strong negative effect on premiums (Phelps 1973). This may reflect economies of scale in offering health insurance. Large firms may also have bargaining power to negotiate lower loading charges. Therefore, we included firm size in the FFS premium equation.

As mentioned earlier, industry dummy variables are proxies for worker health status. We included these industry dummy variables in the FFS premium equation. Last, we included an instrumental variable for HMO enrollment share (for those firms that offer an HMO), and a selection variable to correct for the possible omission of relevant variables related to offering an HMO and the FFS premium.

HMO Premium Equation. Family and single coverage HMO premium equations were estimated for the firms that offer an HMO. Most of these firms offer multiple HMOs. Thus, the dependent variable was the enrollment-weighted average HMO premium. Independent variables include the annual HMO deductible, coinsurance rate, and maximum out-of-pocket charge.(10) HMO plan characteristics, except for the coinsurance rate, were tailored to the type of coverage being analyzed (e.g., the family deductible was used in the family coverage premium equation). In addition, we used measures of active employer shopping for health plans, firm size and location, workers' loyalty to health plans, diversity of preferences, and industry dummy variables in this equation.

V. RESULTS

HMO Offering Equation. Table 2 shows the probit estimates for offering an HMO (sample size 922 firms, of which 239 offer an HMO). Several results support the predictions of our theoretical model. Firms in the Twin Cities and those with more Minnesota employees are more likely to offer HMOs than are other firms. The dummy variable representing firms with more than 25 employees also was statistically significant. This result indicates that HMOs have taken advantage of the federal regulation that requires firms of this size to offer an HMO if one requests to be offered. Skipping the coefficients of the industry dummy variables (which have no clear interpretation and generally are insignificant), we next see that unionized firms are more likely than nonunion firms to offer an HMO. The statistical significance of the union effect increases as the percent of unionized workers increases. Next, firms in which retired employees are eligible for health insurance are more likely to offer an HMO. This supports the argument that diversity of employee demographics in a firm increases the likelihood of offering an HMO.

Two of the "loyalty" variables are statistically significant at the .10 confidence level or less. First, the importance of employee preferences for particular plans is positively related to HMO offering (smaller values of this variable indicate greater importance attached to employee preferences). This finding supports our theoretical model. Second, firms stating that broad coverage is important when selecting plans are less likely to offer HMOs. This counterintuitive finding does not support the theory that such firms should be more likely to offer HMOs.

Shopping actively prior to the expiration of the current health plan contract is negatively related to offering an HMO, but firms that negotiate after the health plan's premium is known are more likely to offer HMOs. The first result supports our theory, whereas the second does not. This finding has another interpretation, however. Firms that offer HMOs may negotiate lower premiums from the HMO, after receiving the HMO's bid. Thus, employer shopping behavior may arise in response to HMO entry.

Several FFS plan characteristics are related to offering an HMO. Firms with higher coinsurance rates are less likely to offer HMOs, as are firms with higher individual deductibles and maximum out-of-pocket expense limits.(11) These findings strongly support our prediction that a "lean" FFS plan does not provide an attractive environment for TABULAR DATA OMITTED an HMO. Likewise, participation in a group purchasing arrangement to obtain lower FFS administrative charges inhibits HMO entry into the firm. Contrary to our expectations, PPOs and HMOs tend to be offered together. The most probable explanation for this finding is that firms redesign their FFS plan in response to HMO entry. Including a preferred provider arrangement in the FFS plan might be a way to overcome the disadvantages related to selection of the HMO by healthier employees. Finally, an HMO is less likely to enter firms where the FFS plan is sponsored by Blue Cross. This finding runs counter to our hypothesis, but it may indicate that employees view Blue Cross, with its generous coverage, as a close substitute for independent practice association (IPA) model HMOs.

HMO Enrollment Share Equation. The HMO enrollment share equation is based on 239 firms that offer an HMO. A joint F-test indicated that the dummy variable for firms with more than 25 employees and measures of employer shopping were statistically significant. Closer inspection showed that joint significance was due solely to the effect of the employer size variable. Therefore, we retained this variable in the final estimates presented in Table 2, but the shopping variables were excluded.

We did not find evidence of selection bias (the model does not omit variables that affect both offering an HMO and HMO enrollment). However, several health plan characteristics were related to HMO enrollment share. Self-insured firms have less HMO enrollment, as expected, as do firms with higher HMO out-of-pocket expense. Higher FFS out-of-pocket expense works in the opposite direction by encouraging HMO enrollment, but this effect is significant only at the 11 percent confidence level. These coefficients support our hypothesis that employees tend to shun health plans with fewer desirable characteristics. Higher FFS deductibles seem to discourage HMO enrollment, however.

HMOs have a significantly lower market share in firms that offer a FFS plan sponsored by Blue Cross, possibly because employees view Blue Cross as a substitute for IPA plans. HMOs have a larger enrollment share in firms that attach greater importance to broad coverage in selecting health plans and firms where the FFS plan has a preferred provider arrangement. The latter finding further supports the argument that PPOs represent a competitive response to HMO entry in a firm. Further studies of this competitive response, in which the PPO variable is treated as endogenous, might be warranted.

Last, firms with more than 25 employees have a high HMO enrollment share. We did not expect that employees' preferences for HMO coverage would be related to the federal mandate to offer an HMO. This finding may be caused by correlation between the dummy variable for more than 25 employees and firm size, measured by the firm's total number of Minnesota employees. It also may be due to the fact that only three small firms in our sample offer HMOs. Therefore, the "small employer effect" reported here is almost equivalent to an intercept for three specific firms. Nevertheless, as we will explain in Section VI, estimates of HMO market share should not exclude this variable.

FFS Premium Equations. We estimated selection-corrected FFS premium equations separately for 683 firms without an HMO and 239 firms with HMOs. Table 3 shows the results for family coverage FFS plans. First, we look at the results for firms without HMOs. The selection coefficient for these firms is not statistically significant, indicating that firms without HMOs do not have omitted characteristics that also affect the FFS premium.

Higher deductibles are associated with lower FFS premiums in firms without HMOs. An additional $100 annual deductible would decrease the average monthly FFS premium by $2.33 in these firms. On an annual basis, the premium savings per additional $100 deductible is $27.96. This result was predicted by our theory. Higher maximum out-of-pocket expense, however, is associated with a higher premium. This runs counter to our theory, although plans with high maximum out-of-pocket expense may also have unmeasured provisions, e.g., unlimited lifetime benefits, that tend to increase premiums.

Self-insurance reduces family FFS premiums by $14.38 per month in firms without HMOs. This important finding indicates that self-insurance saves money for these firms. None of the other variables related to FFS loading charges (group buying, the Blue Cross dummy variable, or the preferred provider dummy variable) was statistically significant in the equation for FFS-only firms. Twin Cities location was associated with significantly higher costs, $12.67 per month, possibly due to high health care prices or the style of medical practice of Twin Cities health care providers.

Next, we look at the results in Table 3 for firms that offer HMOs. The selection variable is insignificant, again indicating that there are no omitted variables common to offering an HMO and to the equation of interest. There is some evidence of adverse selection against the FFS plan in these firms. As predicted by Feldman and Dowd (1982), the FFS premium increases as the percentage of the firm's workers enrolled in HMOs increases. The FFS premium at the mean HMO penetration, 52 percent, is $25 per month higher than in firms without an TABULAR DATA OMITTED HMO, holding other factors constant. However, this effect is not statistically significant at conventional confidence levels (t = 1.327).

Several other interesting results are observed for firms that offer HMOs. Twin Cities firms that offer HMOs have substantially lower FFS premiums than similar firms located in other parts of the state. We estimate this saving to be over $24 per month. In contrast, Twin Cities firms without HMOs had higher FFS costs than similar firms in other parts of the state. Self-insurance and negotiating with health plans are associated with premium savings. The self-insurance strategy reduces family plan premiums by more than $16 per month in firms that offer HMOs.

We also estimated premium equations for single-coverage FFS insurance plans. These tables are omitted to conserve space but we found that single-coverage FFS premiums in firms without HMOs are less expensive if the plan has a higher deductible; firms with HMOs can save money by self-insuring the FFS plan. We did not find evidence of selectivity bias in the single-coverage FFS premium equations.

HMO Premium Equations. Table 4 shows family- and single-coverage HMO premium equations for 239 firms with HMOs. There are three interesting coefficients in the family-coverage HMO premium equation. First, the sign of the selectivity coefficient is negative, TABULAR DATA OMITTED contrary to our expectation that HMOs will target firms where they can charge exceptionally high premiums. However, this coefficient is barely significant at the 10 percent confidence level, and it was sensitive to the specification of the model. One possible explanation is that omission of HMO policy characteristics from the HMO offering equation may cause spurious correlation that is absorbed by the selectivity coefficient. Second, and as predicted by our model, the demand curve for HMO family-coverage enrollment slopes downward. An HMO must cut its monthly premium by $15.53 to achieve the mean enrollment share of 52 percent in a firm. However, this effect lies just outside the 10 percent statistical confidence level. Third, firms that attach greater importance to broad coverage in selecting health plans have higher HMO premiums than other firms.

The single-coverage HMO premium is sensitive to the premium of competing FFS health insurance, with a positive sign as predicted by our theory. However, the slope of the demand curve for single-coverage HMO plans is not significantly different from zero. Also, neither the family- nor single-coverage HMO premium equation is estimated with much statistical precision.

VI. THE COST OF OFFERING HMOS

Calculations. We calculated the cost of offering HMOs for three groups of firms. First, we compared the mean predicted FFS premium in firms without HMOs to the mean predicted weighted average premium in firms with HMOs, using all 922 firms in the sample. Second, we repeated the calculation for 771 large firms (more than 25 employees). Third, we calculated the effect of offering HMOs for the 465 firms in the Twin Cities metropolitan area. All calculations were based on the assumption of equal coverage in the FFS and HMO plans. To obtain equal coverage, we set FFS deductibles, coinsurance, and maximum out-of-pocket expense equal to the levels found in HMOs. This prevented the calculations from being biased against the HMO because it offers lower cost sharing than the FFS plan. Because equation 1 is nonlinear, all calculations were done on a firm-by-firm basis and averaged over firms.

If HMOs entered all the firms in our data set or a random sample of firms with the same measured characteristics as those in our data, we predict that they would obtain a 33 percent enrollment share, on average. This contrasts to the observed share of 52 percent in the 239 firms that have experienced HMO entry. This difference suggests that HMOs have chosen to enter firms where their enrollment share is higher than that expected in the non-entered firms.

The expected FFS family premium would be $164.85 per month if an HMO was not offered, or $189.89 per month if HMOs were offered. The average HMO premium would be $197.11 per month. We calculate from equation 1 that entry of HMOs raises premiums by $25.14 per month, on average. This prediction is statistically significant at the .01 confidence level.(12)

As noted in Section V, HMOs have a greater enrollment share in large firms than in small firms. Even though this result is based on only three small firms with HMOs, it is important not to exclude the dummy variable for "more than 25 employees" from the calculations. The reason is that our study includes 151 small firms. In order to predict the effect of offering HMOs for all firms in the study, it is necessary to retain the distinction between small and large firms. However, our estimates for all firms may be less reliable than those for large firms, since the regression coefficients essentially are based on large firms' experience with HMOs. Therefore, we selected the 771 large firms and repeated the calculation. We found that HMOs' average enrollment share would be 38 percent in the large firms. This is five percentage points higher than the average enrollment share in all firms. Nevertheless, both the HMO and FFS premiums were very similar in large firms and all firms. Consequently, the cost of offering HMOs for large firms, at $28.17 per month, is similar to the cost for all firms.

A different cost is calculated for the 465 Twin Cities firms, however. In Minneapolis-St. Paul, FFS premiums are somewhat higher without HMOs, HMO premiums are somewhat lower, and HMO market share is somewhat higher, compared with the rest of the state. But the big difference is found in the predicted FFS premium with HMOs. This is only $178.11 per month, which is substantially lower than the calculation based on all firms. As a result, the weighted average premium is only $11.02 per month higher in Twin Cities firms.

A similar pattern is observed for the single-coverage cost calculations. Offering an HMO increases the average cost of health insurance by $3.68 per month for all firms and by a slightly larger amount for large firms. Twin Cities location, however, is associated with a much smaller HMO effect on premiums.

VII. DISCUSSION

Our research indicates that offering HMOs increases the cost of employer-sponsored health insurance. This occurs because HMOs are more costly than FFS plans and because offering an HMO raises the cost of the FFS health insurance plan. This finding confirms the earlier study by Jensen, Feldman, and Dowd (1984), that offering an HMO raises the cost of FFS insurance. The effect of HMOs on cost was larger, in both absolute and proportionate terms, for family-coverage premiums compared with single coverage. The average cost for all 922 firms was $25.14 per month in the family plan. Because the firms with HMOs typically were large (more than 25 employees), we also predicted the cost of offering HMOs for this group. We found little difference between the experience of all firms with HMOs and that of large firms. However, firms in the Twin Cities have had much better luck with HMOs. The predicted cost of offering family-coverage HMOs is cut by approximately 56 percent in Twin Cities firms. This occurs primarily because adverse selection against the FFS plan is smaller for these firms. HMO premiums are also somewhat lower in the Twin Cities than elsewhere in Minnesota. These effects both deserve further study. They may indicate that selection of sicker employees into FFS plans decreases in importance over time. In addition, the Twin Cities HMO market may be more competitive than elsewhere in the state.

Our findings are still subject to other interpretations, among which the most likely is differences in the scope of benefits between HMOs and FFS plans. However, Minnesota's FFS plans are subject to some of the most stringent mandated benefits in the nation (Gabel and Jensen 1989). These mandates force a uniform level of benefits on FFS plans which, in some cases, is more generous than the benefits offered by HMOs. Many firms self-insure in order to escape these state mandates. Our finding that FFS premiums are lower in self-insured firms may be due to the avoidance of state mandates by these firms.

The probit analysis shows that federal regulations have had a significant effect on the probability of offering an HMO. Firms with more than 25 employees are significantly more likely to offer an HMO, even when firm size is explicitly controlled. This finding supports the policy analysis of Dowd and Feldman (1985), which suggested that "involuntary" entry of HMOs into firms is a barrier to competition among all health plans.

Our research suggests ways in which firms can obtain lower premiums for their traditional FFS plan. Among these is point-of-purchase cost sharing. We found that an additional $100 annual deductible would decrease the annual family plan premium by $27.96 in firms without HMOs. Other findings indicate that a FFS plan with "lean" policy characteristics (higher coinsurance rates, individual deductibles, and individual maximum out-of-pocket expense) does not provide an attractive environment for an HMO.

ACKNOWLEDGMENTS

We acknowledge the helpful comments from two anonymous referees.

NOTES

1. Our equations use an asterisk to indicate multiplication. If necessary, the multiplicands are separated by (), |~, or {}, in ascending order of inclusion. Functions and operators (e.g., expected value) are denoted by (), |~, or {} in ascending order of inclusion. For example, f = f(x) means "f is a function of x," f = f|g(y)~ means "f is a function of g(y), and x * (y - z) means "multiply x times the difference of y minus z."

2. See Dowd and Feldman (1985) for a detailed discussion of these regulations and their implications for employers.

3. Kirkman-Liff, Christianson, and Hillman (1985) used a similar model of expected profit maximization in their study of competitive bidding for Medicaid contracts. In their model, the bidder calculates expected profit as the probability of winning a contract times the profits if the contract is won. They also included the fixed cost of preparing a bid.

4. A review of studies of HMO selection bias in employment-based health plans found favorable HMO selection in six of eight cases with conclusive results (Wilensky and Rossiter 1986). Buchanan and Cretin (1986) showed that HMOs offered by a large aerospace corporation attracted employees with lower annual health expenditures.

5. Employers can require that FFS plans use experience rating because they can obtain data on health care use and cost for their employees covered by the FFS plan. Federal regulations, however, require that HMOs use "community rating" to set their premiums. These regulations prevent the firm from discovering the HMO's marginal cost of production. A detailed discussion of this point can be found in Dowd and Feldman (1985).

6. According to a personal communication to the first author, the response rate to the SBA survey was 23 percent. The NFIB survey achieved a response rate of 19 percent (National Federation of Independent Business, Research, and Education Foundation 1985).

7. An anonymous referee suggested that HMOs may not be available in all parts of the state. According to the Minnesota Department of Health (1987), HMOs were operating in all parts of Minnesota at the time of our survey. Although HMO market shares were largest in the Twin Cities and the northeast region (around Duluth), every region of the state had at least three HMOs operating in 1986.

8. However, further analysis by Gifford et al. (1991) failed to show that employer shopping was related to health plan premiums. The later study used a latent variable to measure employer shopping, and treated shopping as one of several endogenous variables in a model of health plan premiums and employer cost-control activities.

9. Missing values for enrollment were imputed by randomly drawing from a normal distribution with the same mean and variance as the reported data. This procedure ensures that the expected value of imputed enrollment is equal to the mean of the reported data. However, unlike the commonly used method of assigning the mean value to all missing cases, it does not bias the estimated t-statistics away from zero. Similar imputations were used for all variables in the model.

10. Although HMOs generally are regarded as prepaid plans (with no point-of-purchase cost sharing), this is not always the case. Some HMOs use cost sharing for covered services, particularly in plans sold to small employers.

11. We also estimated the HMO offering and market share equations with FFS family plan deductibles and out-of-pocket expense. Adding these characteristics did not improve the fit of either equation, according to likelihood ratio and joint F-tests. We suspect that family plan FFS characteristics may be measured with error (e.g., family plan deductibles often depend on family size). In addition, family- and single-coverage plan characteristics are positively correlated. Thus, we believe that single-coverage characteristics are sufficient to measure FFS plan "generosity," and we do not need to include family plan characteristics in the HMO offering and market share equations. For similar reasons, HMO family plan characteristics were excluded from the HMO enrollment share equation.

12. Standard errors of the predicted cost effect, reported in Table 5, are based on the mean values of the exogenous variables. Standard errors would be larger if the predictions were based on a different set of exogenous variables.

REFERENCES

Buchanan, J. L., and S. Cretin. "Risk Selection of Families Electing HMO Membership." Medical Care 24, no. 1 (1986): 39-51.

Dowd, B., and R. Feldman. "Biased Selection in Twin Cities Health Plans." In Advances in Health Economics and Health Services Research. Vol. 6. Edited by R. M. Scheffler and L. F. Rossiter. Greenwich, CT: JAI Press, 1985.

-----. "Voluntary Reduction in Health Insurance Coverage: A Theoretical Analysis." Eastern Economic Journal 13, no. 3 (July-September 1987) 215-32.

Dowd, B., R. Feldman, and J. Klein. "What Do Employers Really Want in a Health Plan?" Business and Health 4, no. 3 (January 1987): 44-48.

Dowd, B., R. Feldman, S. Cassou, and M. Finch. "Health Plan Choice and the Utilization of Health Care Services." Review of Economics and Statistics 73, no. 1 (February 1991): 85-93.

Feldman, R. D., and B. E. Dowd. "Simulation of a Health Insurance Market with Adverse Selection." Operations Research 30, no. 6 (November-December 1982): 1027-42.

Feldman, R., M. Finch, B. Dowd, and S. Cassou. "The Demand for Employment-based Health Insurance Plans." Journal of Human Resources 24, no. 1 (Winter 1989): 115-42.

Gabel, J. R., and G. A. Jensen. "The Price of State Mandated Benefits." Inquiry 26, no. 4 (Winter 1989): 419-31.

Gifford, G., R. Feldman, B. Dowd, and M. Finch. "A Simultaneous Equations Model of Employer Strategies for Controlling Health Benefit Costs." Inquiry 28, no. 1 (Spring 1991): 56-66.

Heckman, J. "The Common Structure of Statistical Models of Truncation, Sample Selection and Limited Dependent Variables and a Simple Estimator for Such Models." Annals of Economic and Social Measurement 5, no. 4 (1976): 475-92.

"HMO Savings Depend on Who Enrolls." Wall Street Journal (11 August 1988).
COPYRIGHT 1993 Health Research and Educational Trust
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1993 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:health maintenance organizations
Author:Feldman, Roger; Dowd, Bryan; Gifford, Gregory
Publication:Health Services Research
Date:Feb 1, 1993
Words:9936
Previous Article:The relationship between adjusted hospital mortality and the results of peer review.
Next Article:Relationship of physician ratings of severity of illness and difficulty of clinical management to length of stay.
Topics:

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters