By pooling employees from a variety of small firms, policymakers hope

Transcription

1 DataWatch Implicit Pooling Of Workers From Large And Small Firms by Alan C. Monheit and Jessica Primoff Vistnes Abstract: Risk pools for small employers have become an integral part of proposals for national health care reform and have been implemented by a number of states. These explicit attempts to pool small employers are occurring at the same time that many smallfirm employees obtain health insurance through implicit pooling arrangements as the dependent of a policyholder insured by a large firm. We use data from the 1987 National Medical Expenditure Survey to document the extent of implicit pooling arrangements, to examine whether smallfirm employees and their dependents are adverse health risks, and to assess the cost implications of pooling small and largefirm employees and dependents. By pooling employees from a variety of small firms, policymakers hope for two things: to avoid a number of exclusionary practices that characterize the smallgroup market for health insurance, and at the same time to spread health care risks over a sufficiently large base of enrollees. 1 The ultimate goal of risk pooling is to achieve access to health insurance coverage for employees of small firms at premiums comparable to those available to employees of large firms Explicit attempts to pool groups of smallfirm employees are occurring at the same time that a substantial number of insured smallfirm employees already obtain health insurance through implicit pooling arrangements: as dependents of a worker in a large firm. By our computations, some 36 percent of smallfirm employees (those in firms of twentyfive or fewer workers) covered by employmentrelated health insurance obtain insurance in this manner. This has important implications for reform. Under the Clinton plan, with its proposed communityrated regional health alliances to pool diverse groups of workers from large and small firms, such an implicit subsidy from large to small employers raises important questions regarding the potential impact of drawing smallfirm employees into an insurance pool. In particular, does the inclusion of smallfirm workers and their dependents raise the per capita health care costs of largefirm enrollees and, ultimately, the level of premiums paid by large employers? Put Alan Monheit is senior research manager and Jessica Vistnes is a service fellow at the Agency for Health Care Policy and Research, Center for General Health Services Intramural Research, Division of Medical Expenditure Studies, in Rockville, Maryland.

2 302 HEALTH AFFAIRS Spring (I) 1994 differently, are smallfirm employees and their dependents adverse health risks who can be expected to raise the average level of expenditures for all employees in the pool? What are the implications for health care spending and health insurance premiums should uninsured smallfirm employees and their dependents obtain coverage? Finally, a discussion of implicit pooling also can shed light on equity issues related to an employer mandate. To the extent that small employers are indirectly subsidized by large firms for the health insurance of the former's employees, a stronger argument can be made to require all employers to contribute to health insurance purchases. Our objectives in this DataWatch are to measure the extent of implicit pooling of small and largefirm employees and dependents; to examine whether implicit pooling raises the cost of insurance to largefirm employees by considering whether smallfirm workers and their dependents are adverse health risks; and to examine the cost of providing all smallfirm employees and their dependents with coverage available at large firms. Our findings indicate that implicit pooling arrangements affect a substantial number of smallfirm employees. We also find little evidence to suggest that current implicit pooling arrangements raise the per capita medical care expenditures of largefirm employees or their dependents. Finally, our results suggest that expanded pooling would not raise per capita medical care costs above those incurred by persons with largefirm insurance. Data and methods. The data used in this analysis are from the household component of the 1987 National Medical Expenditure Survey (NMES). The sample used in this paper includes 23,685 persons under age sixtyfive who were present in the last round of NMES and either were employed or were nonworking dependents of employees under age sixtyfive. To identify dependents of employees as well as to assign household characteristics to individuals, households were defined as consisting of an employee and (if present) a working or nonworking spouse, any unmarried children under age nineteen, and any unmarried fulltime students between ages nineteen and twentythree. In classifying workers as employees of small or large firms, firms that were freestanding establishments with fewer than twentysix employees were identified as small firms, and all others were classified as large firms. Persons were classified as employees or dependents in oneworker or twoworker households where the workers were employed by either small or large firms. Employmentrelated insurance was defined in the last round of NMES as insurance obtained through an employee's job. Note that insurance obtained through retirement or a previous job is included in "other private insurance." Finally, unless otherwise indicated, all differences discussed in the text are significant at least at the.05 level for Zscores having asymptotic normal properties.

3 DATAWATCH 303 Sources Of Health Insurance To provide a basis for our discussion of implicit pooling, Exhibit 1 presents the health insurance status of employed persons and their dependents according to household type and size of firm for each employee. As expected, broad differences are revealed in the type of coverage held by employees and dependents in small and large firms. Employees of small firms and their nonworking dependents are far less likely to hold employmentrelated health insurance and are far more likely to be uninsured than are their counterparts in large firms. Only 27.4 percent of employees and 27.3 percent of dependents in smallfirm/singleworker households have employmentrelated coverage, compared with 71.3 percent of employees and 65.9 percent of their dependents in largefirm/singleworker house Exhibit 1 Health Insurance Status Of Employed Persons And Dependents, By Household Type And Firm Size, 31 December 1987 Employment'related Household Population type/firm size (thousands) Total Total employees and dependents 180, % One worker, small firm Employee 15, Dependent 12, One worker, large firm Employee 39, Dependent 32, Two workers, both in small firm Employee 6,773 Dependent 3,921 Two workers, both in large firm Employee 26,907 Dependent 15, Two workers, one small/one large firm Employee Small 8,454 Large 8,367 Dependent 10, Policyholder 33.1% Dependent only 33.4% Other private 14.7% Public only 4.0% b 0.9 b Uninsured 14.7% Source: National Medical Expenditure Survey (NMES), Note: "Other private" insurance includes policyholders or dependents of insurance that is purchased directly from an insurance company or through a group that is not employment related as well as insurance where the policyholder is outside the NMES sample or outside the household, where the policyholder is over age sixtyfive, and where the policyholder holds employmentrelated insurance but is classified as a nonworking dependent (for example, a student age nineteen to twentythree) or where the insurance is through a retirement job or a job held prior to the last round of the NMES survey. The insurance categories are mutually exclusive. a Not applicable. b Relative standard error exceeds 30 percent.

4 304 HEALTH AFFAIRS Spring (I) 1994 holds. Similarly, just over half of employees and dependents in smallfirm/twoworker households are covered by employmentbased coverage, compared with 90.1 percent of employees and 82.4 percent of dependents in largefirm/twoworker households. Although smallfirm households obtain other private coverage at more than twice the rate of persons in large firms, these purchases fail to compensate for the differences in coverage obtained through the workplace. As a result, employees and dependents in smallfirm/oneworker households are more than twice as likely to be uninsured than their counterparts in large firms, while workers and dependents in smallfirm/twoworker households are uninsured at rates three to four times those of their counterparts in large firms. These data indicate that 24.1 million smallfirm employees and dependents lack employmentrelated coverage and are potential candidates for inclusion in risk pools. Since 7.7 million of these persons are covered by privately purchased insurance, such persons might enter voluntary pooling arrangements if outofpocket premiums were commensurate to those now paid. 2 However, the cost implications of including smallfirm workers and dependents in a risk pool along with those insured by large firms must be considered. Before doing so, we first examine the degree to which employees and dependents of small firms are in "implicit" pooling arrangements through which they obtain coverage from large firms. Extent of implicit risk pooling. Small firms are far less likely than large firms to provide their workers with health insurance, and when coverage is provided, it is likely to be more costly than comparable coverage at large firms. Thus, one would expect twoworker households in which the workers are employed by large and small firms to obtain coverage from the large employer. Exhibit 2 examines sources of employmentrelated coverage in twoworker households. We find that almost threequarters of smallfirm employees in households with a largefirm worker obtain coverage from the large firm. In contrast, only about 17 percent of employees of large firms obtain coverage from small firms (10.7 percent as dependents of smallfirm employees and 6.4 percent with coverage from both large and small firms). Also, 78.1 percent of nonworking dependents with access to coverage from both small and large firms obtain insurance exclusively from large employers. Finally, when we consider the source of coverage for all smallfirm employees and dependents with employmentrelated insurance, we find that over onethird of smallfirm employees (36.4 percent or 5.1 million persons) obtain coverage through their working spouses in large firms, and 46 percent of smallfirm dependents (5.7 million persons) obtain coverage from large firms. In total, 40.8 percent of smallfirm employees and dependents (10.8 million persons) with employmentrelated insurance obtain such coverage

5 DATAWATCH 305 Exhibit 2 Sources Of EmploymentRelated Insurance For TwoWorker Households, By Firm Size, 31 December 1987 Household type/firm size Population (thousands) Both employed by small firm Policyholder Dependent Both employed by large firm Policyholder Dependent Both firms Small firm Employee Dependent 3,554 2, % 43.3% %b 3.2 b Large firm Employee Dependent 24,230 13, % 26.7% Small/large firm Employee Small Large Dependent 6,421 6,645 6, Source: National Medical Expenditure Survey (NMES), Notes: Computations are as follows. First, the number of smallfirm employees with employmentrelated insurance is million (derived from Exhibit 1). Second, million smallfirm employees receive insurance from a large firm (36.4 percent, or 5.164/14.169). Third, of million dependents of smallfirm workers with employmentrelated insurance, million are insured by large firms (45.9 percent, or 5.677/12.370). The result is that 40.8 percent of insured smallfirm workers and dependents are insured by large firms: million/( million million). Insurance categories are mutually exclusive. a Not applicable, b Relative standard error exceeds 30 percent. from large employers, and these persons represent 10.4 percent of all employees and dependents with largefirm coverage. These tabulations suggest that large employers may "subsidize" small employers to a substantial degree and that there is currently a large amount of implicit health insurance pooling between employees of large and small firms. 3 They also suggest that concerns regarding equity in the financing of employmentrelated health insurance are warranted. Health Status By Firm Size The extent of implicit health insurance transfers from large to small employers raises questions about potential cost implications of pooling. To elucidate this issue, we compare health status indicators for employees and their dependents according to the worker's firm size and the source of health insurance for the worker and the dependent. Employees. Comparisons of employee health status by firm size alone fail to reveal broad differences (Exhibit 3). Only total health care expenditures ($1,033 for employees in large firms and $890 for those in small firms) vary widely. This disparity, however, most likely reflects the differential rates of insurance coverage between employees in small and large firms (see

6 306 HEALTH AFFAIRS Spring (I) 1994 Exhibit 3 Health Status Of Employees, By Firm Size And Health Insurance Status, 1987 Size of firm/ insurance status Employed by a Large firm Small firm Population (thousands) 74,407 29,860 Largefirm insurance large firm 51,428 small firm 4,677 Smallfirm insurance small firm 8,831 No smallfirm insurance small firm 13,267 Dual coverage from large firms Largefirm employee 6,595 Uninsured large firm 7,745 small firm 7,539 Other private insurance large firm 2,273 small firm 4,785 Public coverage large firm 1,545 small firm 943 Average number of bed days b Percent fair or poor health 10.8% Percent chronic health conditions 29.9% Functional health Total expenditures $1, Source: National Medical Expenditure Survey (NMES), Notes: "Other private insurance" includes policyholders or dependents of insurance that is purchased directly from an insurance company or through a group that is not employment related. Adult chronic health conditions include arthritis, heart disease, high blood pressure, cancer, gall bladder disease, diabetes, rheumatism, emphysema, arteriosclerosis, and other conditions; children's chronic conditions include asthma, allergies, chronic otitis media, digestive problems, and heart problems. Functional health is a scaled variable summarizing limitations in mobility and restrictions in daily activities. Increased values of this variable correspond with worse health. Insurance categories are mutually exclusive. a Total includes some employees in categories not shown in Exhibit 3. b Relative standard error exceeds 30 percent. 1, , , ,794 1,152 Exhibit 1). Using the coefficient of variation (the ratio of the standard deviation of expenditures to mean expenditures) to obtain an indication of the variation in health expenditures, and hence the "riskiness" of insuring specific groups of employees, we find no evidence that smallfirm employees exhibit more variability in their expenditures than largefirm employees

7 DATAWATCH 307 (data not shown). When both firm size and insurance status are considered, however, some striking results emerge, with direct implications for the cost of pooling workers from small and large firms. First, we find no statistically significant differences between the health status measures (including expenditures) for large and smallfirm employees covered by largefirm insurance (the difference between percentage with chronic conditions is significant only at the.10 level). Thus, implicit pooling of large and smallfirm employees, based on these dimensions of health status, does not raise the average health care expenditures of employees in large firms. In addition, the variation in spending of smallfirm employees covered by largefirm insurance is about onefifth less than that of largefirm employees, consistent with the notion that the former are no more risky to insure than the latter. Finally, to draw further inferences regarding the cost implications of including all smallfirm employees in a risk pool with insured largefirm employees (as might occur under the regional health alliances proposed by the Clinton plan), we compare the health status of the latter to the remaining groups of smallfirm employees. We find that employees insured by small firms appear to be more favorable health risks than largefirm employees with largefirm coverage: Average bed days and the percentage reporting fair/poor health are lower for the former, and there are no differences in the remaining health status measures or in spending levels. 4 These smallfirm employees also show less variation in spending and are less likely than largefirm employees with largefirm insurance to have arthritis or emphysema (the latter significant at the.10 level; data for specific conditions are not shown). Other comparisons of health status between insured largefirm employees and the remaining groups of smallfirm employees in Exhibit 3 also show that the latter are in no worse health than those employed and insured by large firms. Nonworking dependents. Consideration of the health status of nonworking dependents of small and largefirm employees, disaggregated into adults and children under age eighteen, yields findings similar to those obtained for employees (Exhibit 4). We find no significant differences in any of the health status measures or in any specific chronic condition between adult dependents whose spouses are employed in either small or large firms. Among smallfirm adult dependents, those with smallfirm insurance, those with other private coverage, and those who are uninsured are in no worse health than are largefirm dependents with largefirm insurance. 5 Finally, in data not presented, we find no consistent pattern to support the hypothesis that implicit pooling raises the average health care costs of children. These comparisons suggest that current implicit pooling, or potentially including all smallfirm employees and dependents in a risk

8 308 HEALTH AFFAIRS Spring (I) 1994 Exhibit 4 Health Status Of Nonworking Dependents Age Eighteen Or Older, By Firm Size And Health Insurance Status, 1987 Size of firm/ Population insurance status (thousands) Dependent of a Largefirm employee (LFE) 14,352 Smallfirm employee (SFE) 5,610 Average number of bed days Percent fair or poor health 16.9% 17.0 Percent chronic health conditions 31.1% 32.8 Functional health Total expenditures $1,752 1,779 Largefirm insurance Dependent of LFE 9,444 Smallfirm insurance Dependent of SFE 1,568 No smallfirm insurance Dependent of SFE c 2, b ,795 2,057 1,145 Uninsured Dependent of LFE 1,350 Dependent of SFE 1, b , Other private insurance Dependent of LFE 634 Dependent of SFE 1, b 3.1 b l,515 b 1,108 Public coverage Dependent of LFE b ,706 Source: National Medical Expenditure Survey (NMES), Notes: "Other private insurance" includes policyholders or dependents of insurance that is purchased directly from an insurance company or through a group that is not employment related. Insurance categories are mutually exclusive. a Total includes some employees in categories not shown in Exhibit 4. b Relative standard error exceeds 30 percent. c Includes dependents of smallfirm employees with public coverage not shown separately. pool with those insured by large firms, would not raise the per capita medical care costs of the group. Adverse health risks and smallgroup market segmentation. A number of observers have alleged that the risk selection practices of insurers segment the smallgroup market so that only persons who are favorable health risks obtain employmentrelated insurance. 6 We examine this possibility in two ways. First, we consider whether employees and dependents with coverage from their small firms are in better health than are those who have not obtained such coverage. Next, we examine whether employees and dependents insured by their small firms are in better health than are smallfirm employees and dependents who have purchased private insurance when workrelated coverage has not been available (thus revealing strong preferences for coverage, which may be characteristic of adverse health risks). 7 Results from these comparisons support the hypothesis of smallgroup market segmentation. We find smallfirm employees with employment

9 DATAWATCH 309 related coverage to be in better health than those without smallfirm insurance (Exhibit 3). The former were less likely to report their health as fair or poor; had fewer bed days; had better functional health status; and were less likely to have arthritis (significant at.10 level), cancer (significant at.10 level), arteriosclerosis, rheumatism, and emphysema. These are common preexisting conditions that may preclude enrollment in health plans. 8 We also find that smallfirm adult dependents with employmentrelated coverage may be more favorable health risks than those without such coverage: The former were less likely to be in fair or poor health, had better functional health, and were less likely to have diabetes and rheumatism. We observe no differences in health status between children with and without smallfirm insurance. The second set of comparisons also reveals that smallfirm employees covered by smallfirm insurance were less likely than smallfirm workers who purchased private coverage to be in fair or poor health. Although there were no differences among the other health status measures in Exhibit 3, we find that smallfirm employees who purchased private coverage show much more variation in their spending levels than do those who obtained coverage directly from small firms, so that the former appear more risky to insure. Employees insured by their small firms also were less likely than those who purchased private coverage to have arthritis, cancer, arteriosclerosis (significant at.10 level), rheumatism, and emphysema. 9 In contrast, smallfirm adult dependents with coverage through small firms do not appear to be in better health than are those with privately purchased coverage. The only differences found were that the former were less likely than the latter to have a chronic health condition (Exhibit 4) and to have arthritis (both significant at.10 level). Comparisons between children insured by small firms and those with other private coverage do not reveal any differences in health status. 10 In sum, the differences in health status between those insured by small firms and other smallfirm employees and dependents support allegations that the smallgroup market may be segmented so that employees who obtain smallfirm insurance are viewed as more favorable health risks. 11 Cost Of Pooling Exhibit 5 examines how pooling additional individuals with those insured by largefirm insurance might affect average medical spending for the group. The predictions of per capita medical expenditures for these additional enrollees were obtained by applying a twopart econometric model of medical expenditures estimated on the sample of current largefirm enrollees. The first part of the model estimates the probability of incurring a

10 310 HEALTH AFFAIRS Spring (I) 1994 Exhibit 5 Predicted Expenditures By Current Insurance Status, Employment Status, And Dependent Status, 1987 Employees Population Adults (age 18 and older) (thousands) Covered by largefirm insurance 63,087 Smallfirm employee 5,114 Largefirm employee 57,973 Smallfirm employees and dependents a Uninsured 7,398 Covered by other private insurance 5,204 Covered by public insurance 1,893 Covered by small firm insurance 9,410 Largefirm employee Uninsured Children (age 17 and younger) Covered by largefirm insurance Smallfirm dependents a Uninsured Covered by other private insurance Covered by public insurance Covered by small firm insurance 7,574 Predicted expenditure $1,094(15.25) 1,021 (34.28) 1,100(16.02) 790(34.11) 948(55.81) 1,084(62.78) 837 (23.33) 876 (37.65) _ Dependents Population (thousands) 10,229 1,528 1, ,568 30,150 3,222 1,565 1,197 3,837 Predicted expenditure $2,032(119.53) 1,343(170.79) 2,202 (402.69) 3,070 (595.87) 1,856(210.00) 680(12.87) 501 (28.72) 797 (82.75) 637(123.62) 644 (23.09) Source: National Medical Expenditure Survey (NMES), Notes: Standard errors are in parentheses. "Other private insurance" includes policyholders or dependents of insurance that is purchased directly from an insurance company or through a group that is not employment related. The insurance categories are not mutually exclusive. a Dependents of oneworker and twoworker smallemployer households. medical expense via a probit equation, and the second part estimates medical expenditures for the sample of individuals with medical expenses via an ordinary least squares regression. 12 The model is estimated separately for individuals age eighteen or older and for those under eighteen. Predicted expenditures in Exhibit 5 assume that all individuals have largefirm insurance, are insured all year, and have no other insurance. Overall, the results indicate that including additional groups of smallfirm employees and dependents in the current pool of largefirm enrollees would not increase the per capita spending (and hence the premiums) of the latter. Predicted expenditures for 1987 average $1,066 for enrollees with largefirm insurance and $933 for newly enrolled smallfirm employees and dependents, and yield an overall average of $1,025. Disaggregating, we find that predicted expenditures for small and largefirm employees with largefirm insurance ($1,021 and $1,100, respectively) are not significantly different, which is consistent with earlier findings of no

11 DATAWATCH 311 differences in their health status. Smallfirm employees with public insurance would face expenditures no different from employees currently insured by large firms ($1,084 compared with $1,094), and their predicted expenditures show less variability than do those for employees with largefirm insurance. Smallfirm employees who are uninsured or have other private insurance would face lower expenditures than all employees who currently have largefirm coverage ($790 and $948 compared with $1,094), as would those with smallfirm insurance ($837). In addition, uninsured largefirm employees would experience lower expenditures once covered than those employees who currently have largefirm insurance ($876 compared with $1,094). The lower predicted expenditures for uninsured employees are consistent with the work of Susan Marquis and Ellen Harrison, who found that within the RAND Health Insurance Experiment, annual spending for the previously uninsured was less than that for persons previously covered by their own employmentrelated plan (except for mental health spending). Adult nonworking dependents, who are disproportionately female, have predicted expenditures that are higher and more variable than those of their employed counterparts. The predicted expenditures of smallfirm dependents who are currently uninsured ($1,343) are lower than all other categories (the comparison with smallfirm dependents covered by smallfirm insurance is significant only at the.10 level). Smallfirm dependents with other private insurance or with smallfirm insurance would have expenditures that are not statistically different from those of dependents who currently have largefirm insurance ($2,202 and $1,856, compared with $2,032). Those with public insurance would have higher expenditures than dependents with largefirm coverage ($3,070 compared with $2,032), but this difference was significant only at the.10 level. Finally, adding children to the pool of largefirm enrollees either would decrease or would have no effect on per capita medical expenditures within that age category. Children of smallfirm workers who are currently uninsured have lower predicted expenditures ($501) than those who are currently insured by large firms ($680). We find no differences between predicted mean expenditures for children of smallfirm workers with smallfirm coverage, other private coverage, or public coverage and those covered by largefirm insurance. However, children with smallfirm insurance, like their adult counterparts, show less variation in predicted expenditures than do children with largefirm insurance. In contrast to adults, predicted expenditures for children covered by public insurance are more variable than are predicted expenditures for children in other insurance categories.

12 312 HEALTH AFFAIRS Spring (I) 1994 Policy Implications Reform of the U.S. health care system is likely to rely upon a variety of riskpooling arrangements to ensure that individuals and families have access to affordable health insurance. Since employees of small firms and their dependents make up a significant proportion of the uninsured population, their inclusion is of particular importance to the success of any pooling mechanism. However, these persons often are perceived to be adverse health risks. Consequently, there is concern that their inclusion in the risk pool will raise per capita health care costs, which, at best, will be borne by all members of the risk pool via higher premiums or, at worst, will threaten the longrun stability of the pooling arrangement. Analyses of data from the 1987 NMES indicate that a fair degree of implicit risk pooling between employees and dependents of small and large firms already exists. We find that 36.4 percent of insured smallfirm employees with employmentrelated insurance and 40.8 percent of such employees and their dependents obtain their coverage from a large firm. The extent of such implicit pooling some 10.8 million smallfirm employees and dependents representing 10.4 percent of all persons with largefirm insurance and the accompanying shift in responsibility for health insurance from small to large employers suggests that a mandate requiring all employers to contribute to the purchase of health insurance could yield greater equity in the financing of health care. Our results indicate that smallfirm employees and dependents subject to implicit pooling are in no worse health than are their counterparts in large firms and therefore do not increase the average health care costs of the largefirm employees or dependents who form the basis of the pool. In the context of a more general pooling arrangement between largefirm employees and all smallfirm workers, our multivariate analyses reveal that if smallfirm employees and dependents obtained largefirm insurance, their medical expenditures would be no greater than those of employees and dependents of large firms. Thus, on both counts, we find little support for the hypothesis that employees and dependents of small firms are likely to be adverse health risks should they obtain coverage comparable to that held by largefirm employees. Our empirical work also points to imperfections in the market for smallgroup insurance. We find that smallfirm employees who obtain coverage from their employers appear to be among the most favorable health risks. This result, together with our finding that smallfirm employees without such coverage may be adverse health risks, is consistent with allegations that insurers may have segmented the smallgroup market and selected only those firms whose employees are expected to have low health care costs.

13 DATAWATCH 313 Our findings and those reported by David Helms and colleagues offer preliminary evidence that including employees of small firms in a health insurance pool with those of large firms, as in the regional health alliances proposed by the Clinton administration, need not have deleterious consequences for health care expenditures and, therefore, for health insurance premiums. 13 Additional research of this nature can make an important contribution to the health care reform debate by clarifying the cost implications of using risk pooling to expand health insurance coverage. An earlier version of this paper was presented at the Annual Meeting of the Eastern Economic Association, 17 March 1993, in Washington, D.C. The views expressed in this DataWatch are those of the authors, and no official endorsement by the Agency for Health Care Policy and Research (AHCPR) or the Department of Health and Human Services is intended or should be inferred. We thank Bill Custer of the Employee Benefit Research Institute and anonymous referees for helpful comments, and Steve Cohen of AHCPR for advice on statistical issues. Gary Moore of Social and Scientific Systems provided superb programming assistance. NOTES 1. See The President's Comprehensive Health Reform Program ( 6 February 1992 ) of the Bush administration; P.M. Ellwood, A.C. Enthoven, and L.M. Etheredge, "The Jackson Hole Initiatives for a Twentyfirst Century American Health Care System," Health Economics (October 1992): ; and U.S. General Accounting Office, Access to Health Insurance: State Efforts to Assist Small Businesses, GAO/HRD9290 (Washington: U.S. Government Printing Office, May 1992). Space limitations preclude full discussion of the empirical literature and bibliographic references here. This is available from the authors at the Agency for Health Care Policy and Research, Center for General Health Services Intramural Research, 2101 East Jefferson Street, Suite 500, Rockville, Maryland The 24.1 million figure is derived by applying percentages from the "public only" and "uninsured" columns of Exhibit 1 to rows corresponding to smallfirm employees and dependents and small/largefirm employees and dependents. The figure also includes smallfirm employees and dependents with other private coverage (7.7 million persons) obtained from Exhibits 3 and 4 and from unpublished data on children with such coverage. 3. An explicit estimate of the subsidy from large to smallfirm employers requires knowledge of the extent to which largefirm employers pay the marginal cost of family coverage. 4. This comparison does not account for differences in benefits between small and large firms. Smallfirm employees might incur greater expenditures if their coverage was commensurate with that of large firms. We address this issue in our simulations. 5. Adult dependents insured by smallfirm workers had better functional health (significant at the.10 level) and were less likely to have diabetes than adult dependents with largefirm coverage. There were no differences in any other health status measure or in average expenditures. 6. J. Kosterlitz, "Unrisky Business," National Journal (6 April 1991): 796; M.A. Hall, "Reforming the Health Insurance Market for Small Businesses," The New England Journal of Medicine (20 February 1992): ; and G. Kolata, "New Insurance Practices: Dividing the Sick from the Well," The New York Times, 4 March 1992, Al,

14 314 HEALTH AFFAIRS Spring (I) 1994 A In Exhibits 3 and 4 "other private" coverage refers only to insurance that is obtained directly from an insurance company or through a group that is not employment related. Note also that the uninsured also may be adverse health risks. Some may undervalue or not be able to afford coverage. Those with preexisting health conditions may be denied coverage. 8. We found no difference in rates of diabetes, heart disease, hypertension, or gall bladder disease between these groups. For a list of preexisting conditions, see U.S. Congress Office of Technology Assessment, Medical Testing and Health Insurance, OTAH384 (Washington: U.S. GPO, August 1988). 9. We also applied a more rigorous Scheffe test for multiple comparisons to examine whether the differences in these chronic conditions were significant when tested simultaneously across groups of smallfirm employees. With this test, only differences in rates of arthritis and rheumatism retained significance. However, the Scheffe test does not negate our earlier comparisons between smallfirm employees insured by their firms and all other smallfirm employees. We did not find compelling evidence that uninsured smallfirm employees are adverse health risks compared with smallfirm employees with employmentrelated coverage. The few employees in small firms covered by public insurance were found to be in worse health than their counterparts covered by employmentrelated insurance. 10. We did not find uninsured adult dependents of smallfirm employees to be in worse health than those with smallfirm coverage. We found that the few largefirm dependents with public insurance were in somewhat worse health than dependents with largefirm insurance: The latter were more likely to report their health as fair or poor, had worse functional health, and had more bed days on average and higher total expenditures (both significant at the.10 level). 11. Our results are consistent with M.S. Marquis and E.R. Harrison, "Health Status and Health Care Use of Uninsured Workers," in Health Benefits and the Workforce (Washington: U.S. Department of Labor, 1992), Per capita medical expenditures were estimated for additional enrollees not covered by largefirm insurance by applying their characteristics to the estimating equations. Since persons with largefirm insurance could have obtained such coverage in anticipation of high health care use, applying coefficients derived from the behavior of largefirm employees to other groups could lead to biased predictions. We address such potential selection bias by directly including measures of an employee's attitude toward risk and health insurance as well as measures of health status as explanatory variables in the model. Finally, our estimated twopart model does not account for possible transitory responses to improved coverage (from pentup demand for medical care) by the uninsured or by those whose coverage is less generous than largefirm insurance. We justify this approach by noting that Marquis and Harrison conclude that such responses by the newly insured are likely to be small (other than for mental health care use) and that such persons are likely to use fewer services than the currently insured. 12. Probit estimates of the probability of incurring medical expenditures are combined with ordinary least squares estimates of expenditures to yield predicted expenditures for each person in the sample. The estimating equations control for age, sex, race, education, income, marital status, location, health insurance source, employee or dependent status, measures of health status, and attitudes toward risk. Estimates of the twopart model are available from the authors upon request. 13. W.D. Helms, A.K. Gauthier, and D.M. Campion, "Mending the Flaws in the Small Group Market," Health Affairs (Summer 1992): 727.