|
Insuring Low-Income Adults: Does Public Coverage Crowd Out Private?
During the mid-1990s Minnesota, Washington State, Oregon, and Tennessee implemented programs to provide subsidized health insurance for low-income persons who were not previously eligible for Medicaid. We estimate the effects of these programs on the health insurance status of low-income adults in these states. We find that among persons with family incomes below 100 percent of the federal poverty level, subsidized public coverage reduced the number of uninsured persons with very little effect on private coverage rates. Among persons with income between 100 percent and 200 percent of FPL, public coverage reduced the number of uninsured persons and crowded out some private insurance. The partial successes achieved by these programs should be kept in perspective: Even after program implementation, approximately 30 percent of low-income adults in the four states were uninsured.
During the mid-1990s a handful of states—Minnesota, Washington, Tennessee, and Oregon—implemented programs to provide subsidized health insurance for low-income persons who were previously not eligible for Medicaid.1 These states enrolled large numbers of low-income persons, particularly among those whose required premiums were less than 3 percent of family income.2 However, we know very little about whether these programs actually reduced the numbers of uninsured persons. The programs might have primarily enrolled persons who would have been covered by private insurance if the programs had not existed—that is, they might have crowded out private insurance. Alternatively, they might have enrolled persons who otherwise would have been covered by Medicaid. To the extent that the programs crowded out either Medicaid or private insurance, they could have enrolled substantial numbers of low-income persons without doing much to reduce the number of uninsured persons. The purpose of this paper is to assess the effects of subsidized insurance programs on the health insurance status of low-income adults. To what extent did these programs reduce the number of uninsured? To what extent did they crowd out private insurance or Medicaid coverage? Did the effects vary by state or by enrollees’ income level? Factors that influence the extent of crowding out. Stimulated by David Cutler and Jonathan Gruber’s provocative 1996 work, a number of analysts have estimated whether the Medicaid expansions of the late 1980s and early 1990s crowded out private health insurance coverage.3 Most of the studies conclude that approximately 20 percent of the increase in Medicaid enrollment that occurred following the expansions resulted from crowding out of private insurance, while the remaining increase resulted from a reduction in the number of uninsured persons.4
Although the literature on crowding out has reached a consensus on the effects of Medicaid expansions on private insurance coverage for children, this work has limited applicability to understanding the effects of subsidized insurance programs for low-income workers. The state programs we studied expanded eligibility to many persons with incomes of 100–200 percent of the federal poverty level, while the Medicaid expansions primarily expanded eligibility to children below 100 percent of poverty (see Exhibit 1
We expect more crowding out as the programs expanded to persons with higher incomes, primarily because there is more private insurance available to crowd out. However, other program features should have pushed in the direction of reducing crowding out. First, the Medicaid expansions provided coverage with no premium payments required, while the programs of subsidized insurance required most new enrollees to pay some premium.5 This should reduce the extent of crowding out, since some people who are tempted to drop private coverage in favor of free public coverage may be less willing to do so if a premium is required.6 Second, Medicaid is an entitlement program, and Medicaid eligibility expansions were likely perceived to be "permanent." In sharp contrast, the state programs of subsidized insurance were less stable, which should have reduced the extent of crowding out, since employers and employees would be less willing to give up private insurance in favor of public coverage if that coverage might not be available in the future. Finally, Minnesota and Tennessee adopted explicit protections against crowding out: Applicants were not eligible for MinnesotaCare if they were privately insured during the four months prior to application, or if employer-sponsored insurance was available during the previous eighteen months. TennCare restricted eligibility to persons who did not have an offer of employer coverage. The net effect of all of these factors is unclear.
We present results from two types of analyses. The first is a simple pre-post analysis in which we compared insurance status in the four states before and after the subsidized coverage programs were implemented. We used a "difference-of-difference" approach in which we compared the change in insurance status in the states having subsidized insurance programs with that in other states in the region. The second approach estimates a multivariate logistic regression to predict insurance status as a function of individual demographic and employment characteristics, as well as fixed effects for each state and each year. We think that the second analysis produces more accurate estimates of program effects, but the simpler analysis enables one to assess the sensitivity of the results to alternative analytic methods. Data sources. We used the March supplements to the Current Population Survey (CPS) from 1988 to 1999 to measure insurance status as well as individual employment and demographic characteristics. The CPS provides a large database, with detailed demographic and employment information and relatively consistent questions on health insurance status over the twelve-year period.7 We also obtained data on total state program enrollment among adults in December of each year from administrators of the four programs.8 Measuring insurance status—specifying the dependent variables. Starting in 1995 the CPS added a set of questions to identify recipients in state-sponsored insurance programs.9 After the standard set of questions asking about private insurance, Medicare, and Medicaid coverage, respondents were asked whether they were covered by any other form of insurance, such as MinnesotaCare (for Minnesota respondents), Basic Health Plan (for Washington respondents), and other state-specific programs. Respondents in our study states were much more likely to report coverage in these programs than were respondents from other states.10 Although these questions are useful in identifying enrollees in state-specific programs, they do have limitations. Most notably, in the March 1996–March 1999 surveys, "expansion" enrollment in TennCare and the Oregon Health Plan (OHP) cannot be separated from regular Medicaid enrollment. That is, respondents who were enrolled in TennCare or the OHP as a result of the eligibility expansions are treated in the CPS as regular Medicaid enrollees. Because of this, we combined enrollment in Medicaid and in state-sponsored programs into a "public coverage" outcome. We combined employer-sponsored and nongroup coverage into a "private coverage" outcome. We analyzed the effects of state programs on three outcome variables: the probabilities of being privately covered, publicly covered, and uninsured.11 We excluded from the analysis the few respondents with Medicare or Civilian Health and Medical Program of the Uniformed Services (CHAMPUS) coverage. Our analysis was restricted to adults ages nineteen to sixty-four with incomes below 200 percent of poverty. We focused on adults for three reasons. First, much of the existing literature on crowding out deals with the effects on children, not adults. Second, many states moved to expand coverage to children in the mid-1990s, and specifying and measuring the size of these programs is beyond the scope of this work. Third, enactment of the State Children’s Health Insurance Program (SCHIP) largely resolved the policy question of whether to extend subsidized insurance to low-income children, while coverage for adults is still very much the subject of debate in state and federal legislatures. We restricted the main analyses to persons with family income below 200 percent of poverty because few persons with incomes above that level are enrolled in the programs. As discussed below, however, some CPS respondents with incomes above 200 percent of poverty did report being enrolled in the state programs, and we report the results of supplementary analyses including persons at 200–300 percent of poverty. "Difference-of-difference" analysis. We provide a simplified view of the effects of these programs with a difference-of-difference analysis, in which we compared the insurance status in each state for the five years after program implementation with that for the seven years before implementation, using the change in insurance status in other states in the region as controls. Although the pace and timing of implementation varied somewhat among our four states, few people were enrolled prior to 1994 in any state. To simplify the analysis, we compared insurance status in the 1994–1998 time period with that in the 1987–1993 time period in each state. This analysis produced an estimate of the effects of each program on the insurance status of low-income adults. As we see below, program size varied greatly across the four states. For example, in 1994 approximately 33 percent of adults with incomes below 200 percent of poverty in Tennessee were enrolled in TennCare, while the penetration rate of MinnesotaCare among low-income adults was under 4 percent. To estimate the effects of a 10 percent program penetration rate on insurance status, we divided the difference-of-difference estimates of program effects on insurance status by the estimated average program penetration rate from 1994 to 1998. To estimate program penetration, we used responses to the CPS questions on "other government insurance" to allocate total program enrollment to adults under 100 percent, 100–200 percent, and above 200 percent of poverty.12 On average, across the four states, we estimated that approximately 80 percent of adult program enrollees had family incomes under 200 percent of poverty. We divided the estimated number of adult enrollees at this poverty level by the total number of low-income adults to estimate the program penetration rates among low-income adults in each state and year. Multivariate analysis. The difference-of-difference estimates are instructive, but they do not take full advantage of the information available to estimate program effects. The estimates assume that the changes in insurance status in the states we studied would have been similar to changes in the rest of the region. However, changes in economic conditions or demographic characteristics in our program states may have caused insurance coverage to evolve differently there. To control for such factors, we estimated a multivariate logistic regression in which we predicted a person’s insurance status as a function of demographic and employment characteristics, indicator variables for state and year, and a set of eight dummy variables indicating the presence of the programs. There are two program variables for each of the four states: one for adults below 100 percent of poverty, and a second for adults with family income of 100–200 percent of poverty.13 To concisely summarize the results of the multivariate logistic regressions, we used the parameter estimates from the model to predict the probabilities of private coverage, public coverage, and no coverage for each low-income respondent in our four states in the 1995–1999 CPS samples. We made two predictions for each respondent: first, assuming that the program did not exist (that is, setting the indicator variable equal to zero), and second, assuming that the program did exist. The average of the difference in probabilities across all respondents in the state provides our estimate of the effects of the state program on the probability of private coverage, public coverage, and no coverage. We calculated standard errors using the delta method.14 As in the difference-of-difference analysis, we divided the estimated program effects in each state by the estimated average program penetration in the state over the 1994–1998 time period, to estimate the effect of a 10 percent program penetration rate on the probability of public coverage, private coverage, and no coverage. In addition to summarizing the results by state, we summarize the results separately for respondents under 100 percent and at 100–200 percent of poverty. In the summary by income level we combined data across the four states to increase the stability of the results. We divided the estimated effect of the state programs by the estimated program penetration rate at both income levels, to estimate the effects of enrolling 10 percent of the population at a given income level into a state program. Supplementary analysis. In each of the four states, a small number of CPS respondents with incomes of 200–300 percent of poverty reported that they had non-Medicaid public coverage, despite the fact that few persons with incomes above 200 percent of poverty were eligible for subsidized coverage. In a supplementary analysiswe replicated the multivariate analysis described above, but we included respondents with incomes below 300 percent of poverty and an additional indicator variable in each state for persons at 200–300 percent of poverty.
Penetration rates. The subsidized insurance programs enrolled substantial numbers of low-income adults (Exhibit 2  ). TennCare is the largest program, enrolling close to one-third of Tennessee adults with incomes below 200 percent of poverty in its first full year of operation; enrollment declined after the program was closed to most new adult enrollees and then rose in 1998 when enrollment was reopened for specified groups. The OHP enrolled approximately 15 percent of low-income adult Oregonians during 1995–1998. The penetration rate for Washington’s Basic Health Plan increased throughout the 1994–1998 period, rising to 17 percent of the low-income population by 1998. MinnesotaCare, primarily a program for children, enrolled relatively few adults—averaging only 7 percent of low-income adults over the 1994–1998 period.
Difference-of-difference analysis. In all four states public coverage among adults below 200 percent of poverty increased by significantly more than it did in the comparison states in the region (Exhibit 3 ). In Oregon, for example, public coverage increased by an average of 9.3 percentage points in 1994–1998 compared with the public coverage level in 1987–1993. In the rest of the western region public coverage increased by only 0.7 percentage points. As a result, the difference-of-difference estimate is that public coverage in Oregon increased by 8.6 percentage points more than it did in the rest of the region after 1994. The largest increase in public coverage occurred in Tennessee, where public coverage rose by 13.1 percentage points more than it did in the rest of the South.
The difference-of-difference estimates suggest that there was virtually no crowding out of private insurance in Oregon and relatively little in Washington, where approximately 80 percent of the increase in public coverage is accounted for by a decrease in the uninsured. The remaining 20 percent resulted from a decrease in the percentage of low-income adults with private insurance (although the change in private insurance is not statistically significant).15 In contrast, in Tennessee the difference-of-difference estimates suggest that 42 percent of the increase in public coverage resulted from crowding out of private insurance. And in Minnesota the entire increase in public program enrollment is accounted for by a decline in private coverage. The point estimate is that the percentage of the low-income population in Minnesota that was uninsured increased more than in the rest of the Midwest, although the change is not significantly different from zero.
To estimate the effects of a 10 percent program penetration rate on coverage rates, we divided the difference-of-difference estimates of the program effect by the estimated average program penetration rate among adults below 200 percent of poverty over the 1994–1998 period (Exhibit 3
Logistic regression analysis.
The results from the multivariate logistic regressions largely confirm the findings from the simple difference-of-difference analysis (Exhibits 3
Analysis by income level. As expected, among persons below 100 percent of poverty the main effect of the state programs is to reduce the number of uninsured persons (Exhibit 5 ). Among persons at 100–200 percent of poverty approximately 55 percent of the increase in public program enrollment is estimated to have resulted from a reduction in the number of uninsured, while the remainder is associated with a decline in private insurance coverage.
In supplementary work we have extended the logistic regression analysis to include persons at 200–300 percent of poverty. In this income group approximately 65 percent of increased public program enrollment is associated with a decline in private coverage (data not shown). As noted above, persons above 200 percent of poverty were not, in general, eligible for subsidized coverage, and the estimated program penetration rate among persons in this income range is only 3.8 percent; we are cautious about drawing strong conclusions based on this limited group.
We have shown that in two states, Oregon and Washington, expansion of public coverage resulted in a decline in the number of uninsured and very little crowding out of private insurance. In Tennessee an expansion of public coverage was associated with a decrease in the number of both uninsured persons and privately insured persons. In Minnesota the implementation of MinnesotaCare was accompanied by a decline in the number of privately insured persons and virtually no change in that of uninsured persons.17 We do not have a convincing explanation for these differences in results across states. The Oregon program is targeted at persons below 100 percent of poverty, which might account, in part, for its relative success. But the Washington Basic Health Plan, which provides subsidies to persons up to 200 percent of poverty, resulted in little crowding out as well. Both TennCare and MinnesotaCare have greater explicit protections against crowding out than the Oregon and Washington programs; despite these protections, they appear to have experienced greater amounts of crowding out. Estimated program effects vary widely by income group: Crowding out increases as adults with higher incomes enroll in the programs. Combining all four states together, we estimate that among persons below 100 percent of poverty, a 10 percent program penetration rate resulted in a 4.7 percent increase in public program enrollment, a 3.8 percent decrease in the percentage uninsured, and little change in the percentage with private insurance. Among adults at 100–200 percent of poverty, a 10 percent program penetration rate is estimated to have resulted in a 9.3 percent increase in public program enrollment, with approximately 45 percent of this increase coming from the crowding out of private coverage. It is clear from our results that the programs in these four states led to little crowding out among adults with incomes below 100 percent of poverty, but crowding out did occur among persons with incomes 100–200 percent of poverty.18 However, there is much uncertainty about the amount of crowding out we might expect if similar programs were implemented in other states. There are two main sources of uncertainty. First, as we have shown, the effects of the state programs on the probability of having private coverage vary across our four states, and we were unable to convincingly connect variations in program characteristics with variations in program effects. This makes it difficult to know whether programs in other states might be expected to have effects closer to those in Oregon, where there was very little crowding out, or Tennessee, where there was more crowding out. Second, while our point estimate is that 45percentofthe increase in public coverage among persons with incomes of 100–200 percent of poverty results from a reduction in private coverage, the 95 percent confidence intervals for the estimated effects on private insurance and on uninsurance are consistent with a wide range of crowding-out estimates. Research on the effects of coverage expansions for low-income adults is in its infancy, and we expect that additional work using different data sources and studying other programs will be needed if we are to gain a more precise understanding of program effects. One aspect of our results is somewhat puzzling: namely, the incomplete response in the CPS data to the programs we studied. For example, in Oregon we estimated that for every hundred adults who enroll in the program, sixty-six additional CPS respondents will report themselves as having public insurance and sixty-five fewer will report themselves as being uninsured, with virtually no change in the number reporting private insurance. What happened to the other thirty-four program enrollees? This is even more puzzling in Minnesota and Washington, where there is even less reflection in the CPS data of the enrollment in the states’ subsidized programs. One possible explanation is that some people who enrolled in the public programs would have been enrolled in Medicaid if those programs hadn’t existed. Although federal participation in Medicaid gives states a strong financial incentives to maximize Medicaid enrollment, some persons may have preferred the state-specific programs to Medicaid. This explanation is supported by our finding that the CPS response to program enrollment is incomplete among persons below 100 percent of poverty, where "Medicaid crowding out" is most likely, but not among persons at 100–200 percent of poverty, where Medicaid crowding out is less likely.
A second possible explanation is that the CPS undercounts the number of persons enrolled in these public programs. It is well known that the CPS undercounts Medicaid coverage; apparently some who are covered by Medicaid do not report this coverage to the CPS interviewer.19 It is likely that the CPS undercounts coverage in states’ publicly subsidized programs as well. To the extent that the CPS undercounts program enrollment, then our estimates of program effects on the number of uninsured persons are biased downward; the true effect of the program on the number of uninsured persons is even larger than the effects we estimate in Exhibit 4
Projections of the effects ofa national program.
If subsidized insurance for low-income workers were implemented nationwide, and the program were similar to the programs we have studied here, we might expect from the results in Exhibit 2
There are approximately thirty-one million uninsured adults; a reduction of five million would be important but would still leave a sizable problem.21 As seen in Exhibit 1 If expanded programs of subsidized insurance for low-income adults are to greatly reduce the numbers of uninsured persons, they must be designed, implemented, financed, and marketed more successfully than were the programs we studied. Such programs have the potential to reduce private coverage, particularly as they are extended to persons with incomes above the federal poverty level. As has been discussed elsewhere, crowding out of private coverage will result in welfare-improving enhancements for low-income persons but does reduce the program’s target efficiency. Given the pressing problems created by the existence of close to forty million uninsured persons, we think that designing programs to maximize participation should be an overriding policy goal. However, in a voluntary market programs that are attractive enough to enroll large numbers of uninsured persons inevitably will be attractive enough to enroll large numbers of persons who would have had private insurance in the program’s absence.
Richard Kronick is an associate professor in the Division of Health Care Sciences, Department of Family and Preventive Medicine, University of California, San Diego in La Jolla. Todd Gilmer is an assistant adjunct professor there. This research was supported by a grant from the California Program on Access to Care of the California Policy Research Center. The authors thank Sandra Hunt at PricewaterhouseCoopers for providing information on the Oregon Health Plan and TennCare, Carolyn Watts and Vicki Wilson for sharing insights on the Washington Basic Health Plan, and Gestur Davidson for information on MinnesotaCare.
This article has been cited by other articles:
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Top
Data And Methods
Study Results
