U.S. Department of Health and Human Services
May 14, 2007
PDF Version (16 PDF pages)
This report was prepared under contract #HHS-100-03-0022 between the U.S. Department of Health and Human Services (HHS), Office of Disability, Aging and Long-Term Care Policy (DALTCP) and the MEDSTAT Group. For additional information about this subject, you can visit the DALTCP home page at http://aspe.hhs.gov/_/office_specific/daltcp.cfm or contact the ASPE Project Officer, John Drabek, at HHS/ASPE/DALTCP, Room 424E, H.H. Humphrey Building, 200 Independence Avenue, S.W., Washington, D.C. 20201. His e-mail address is: John.Drabek@hhs.gov.
The opinions and views expressed in this report are those of the authors. They do not necessarily reflect the views of the Department of Health and Human Services, the contractor or any other funding organization.
TABLE OF CONTENTS
- LIST OF TABLES
- TABLE 1: Sample Means, SIPP Data on Workers with At Least One Year of Continuous Coverage
- TABLE 2a: Male Age 28, Earning $50,000 Annually, Expecting an 8% Raise
- TABLE 2b: Male Age 45, Earning $80,000 Annually, Expecting a 5% Raise
- TABLE 2c: Female Age 28, Earning $50,000 Annually, Expecting an 8% Raise
- TABLE A1: Logistic Regression of 1 Year Probability of Dropping Small Group Coverage
- TABLE A2: Logistic Regression of 1 Year Probability of Dropping Individual Coverage
ABSTRACT
This paper describes the relationship between the type of insurance coverage a person has in one period and the likelihood of becoming uninsured in the next. Using SIPP data, we find that, for people at the median health status ("very good"), becoming uninsured is most likely for those with individual insurance, less likely for those with small group insurance, and least likely for those with large group insurance. However, we find that for people in poor or fair health, the chances of losing coverage are much greater for people who had small group insurance than for those who had individual insurance. We attribute these results to the offsetting effects of high loadings and guaranteed renewability on the individual market.
INTRODUCTION
Among the various ways Americans can obtain private health insurance coverage, the version usually cited as most problematic is the individual market. Data on this market show some problems relative either to private employment-based group insurance or public insurance: higher premiums relative to benefits because of higher administrative costs and lower or no tax breaks, risk-based underwriting, and higher rates of turnover. Consumers clearly are faced with more immediate challenges in navigating the individual insurance market than in other settings (though we shall argue that there are some serious though less obvious problems in group markets as well).
One unanswered question about different types of health insurance markets, however, is in many ways the most important one: once someone has done what many consider to be the right thing and obtained insurance before becoming a high risk, how likely is he or she to stick with insurance if a high risk condition hits? To what extent can being or becoming a high risk trigger the loss of insurance coverage? We know that the great majority of high risk people uninsured at any point in time had some private insurance sometime in their lives, so one way to diminish the problem of the high risk uninsured would be to emphasize insurance arrangements that have a lower likelihood of people losing coverage (or have higher "persistence"), especially when they transition into high risk states. We also know that the vast majority of people start out their adult lives as low risks, and only become higher risks as they age and chronic conditions pile up. Social concern about losing coverage should apply to people at all risk levels, but we probably are more concerned about the retention of coverage by people who are or become high health risks, since insurance-impacted access to care may have the greatest short run consequences for health for people who are already sickly.
The common negative perception of individual insurance, based largely on its high administrative cost and lack of tax advantages, suggests that it might score poorly on the persistence scale, and especially for higher risks. But there is a feature unique to individual insurance which potentially works in the opposite direction: guaranteed renewability at class average rates. This is a policy provision in which the insurer promises not to increase premiums differentially based on health risk for people seeking to renew. While individual insurance is medically underwritten when newly issued, and while the insurer is free to raise premiums for all in a class, individual insurance is not supposed to be "re-underwritten" if one buyer's risk should change; the initial buyer has an unqualified right to renew at the rate charged to others in the class, regardless of any change in his or her health state. Group insurance, in contrast, is often not guaranteed renewable at premiums independent of risk changes at the group level (depending on state insurance regulations and whether or not the group self-insures), because changes in group membership as workers quit or are hired over time may change group risk or experience. More importantly, there is no guarantee whatsoever of a continued long term offer of coverage at nondiscriminatory rates to individual employees (or their dependents) who leave the firm's employment, whatever the offer of insurance to those who remain in the group. The Health Insurance Portability and Accountability Act "guaranteed issue" rules that protect some workers leaving a group and transitioning to individual coverage do not specify anything about future premiums, and, in particular, do not preclude the possibility that the offering of nongroup coverage to someone who became a high risk is at extremely high premiums.
For group insurance the economically relevant premium is whatever the employees might have been explicitly charged for insurance (the "employee share") plus the portion of the premium the workers paid by devoting a portion of their compensation to a benefit the employer paid directly to the insurer (the "employer share"). Specifically, employees who become high risk and lose their jobs have at best temporary protection at average group premiums under Consolidated Omnibus Budget Reconciliation Act (COBRA) provisions, but they will be left without a guarantee of the opportunity to obtain coverage at the previous group premiums after 18 months.
This paper reports on analysis of panel data intended to see whether guaranteed renewability provisions in individual insurance actually do provide protection against losing coverage as written, and to what extent they provide better protection to high risks compared to group insurance. (We cannot conclusively rule out some insurer re-underwriting because we cannot know the details of how relative premiums changed, but we can see whether anything happened to cause the person to drop coverage.) There has been a fear that some individual insurers are able to avoid carrying out the obligations the policy provisions appear to impose,1 by such strategies as dropping entire classes of insurance. Whatever is reported in a few anecdotes, the more important question is how the protection actually held by a full set of workers with individual insurance with guaranteed renewability differs from that provided to a similar set of workers who have group insurance.
Specifically, we use recent data from the Survey of Income and Program Participation (SIPP), a relatively large panel survey, to see what happens over time to people who initially obtained different kinds of insurance coverage. We examine differences in the absolute and relative probabilities of becoming uninsured at insurance renewal time for workers at different risk levels who had individual insurance, small group insurance, and large group insurance, for two different lengths of time with the initial form of insurance.
Our analysis is complementary to the recent study by Ziller et al.2 who also explored SIPP data on insurance coverage. They only looked at people with individual insurance, and looked at those who terminated spells of this type of insurance for a wide variety of alternatives (group insurance, public insurance, and becoming uninsured), not at those who persisted. In contrast, we compare people who started with group insurance as well as individual insurance, and look at determinants of the absolute and relative probability of becoming uninsured. Specifically, we ask what are the odds that someone at a given level of health status who had either individual or group coverage already in the previous year becomes uninsured in the next time period (the difference in absolute probability). We also ask how the odds of becoming uninsured vary with the risk level for people with each type of coverage (the difference in relative probability). However, we only look at the outcome of becoming uninsured, and do not ask whether the coverage that continues is the initial coverage the person had or some substitute (another type of private coverage or Medicaid coverage) that he or she was able to obtain.
A CONCEPTUAL MODEL
We suppose that people are risk averse, so they potentially value insurance, compared to the alternative of taking a chance on various levels of out of pocket payment. However, whether they will choose coverage in any given time period will depend both on the premium they would have to pay and on the expected benefits they would expect to collect. That is, a person is most likely to obtain insurance if the premium is low but the risk of seeking medical care (and therefore the benefits to be collected on average from a given policy) is high.
A measure that combines both of these factors is the "net loading" for an insurance policy: the difference between the net-of-taxes premium the person pays and the average benefits they might collect; this is usually measured as a proportion of the premium. The higher the loading, the less likely the person is to buy insurance in the first place. But what about the willingness to continue with coverage? Clearly that will be lower should the loading increase. But it also seems plausible to assume that, the higher a constant-over-time loading (and therefore the lower the net benefit) from insurance that one had bought, the greater the likelihood of failing to renew. A fall in disposable income, the onset of some other financial emergency, or even just lower marketing of or attention to the need for coverage is more likely to lead to neglect to renew when the price of insurance relative to its benefits was not very advantageous in their first place.
We can use this model to predict when there will be greater likelihood of dropping or losing coverage. On average, insurance net-of-tax loading is known to be lowest for large group insurance, higher for small group insurance, and highest of all for individual insurance.3 The loading in group insurance is reduced both by the lower resource cost of administering such insurance and the fact that, compared to individual insurance, the potential tax advantages are greater. The worker who receives compensation in the form of a partially employer paid premium can avoid income and payroll taxes on that amount of income, and the adoption of a Section 125 or cafeteria plan can shield any explicit employee premium from income taxes. In contrast, a person who buys individual insurance cannot reduce payroll taxes, and can reduce income taxes only if self employed and itemizing deductions, and only toward the end of the observation period. These observations imply that the likelihood of continuation of coverage for a person of average risk at a given income level should be greatest in large group insurance and lowest in individual insurance, with small group insurance being intermediate.
The more complex question has to do with the level of and variation in risk. As a benchmark, if premiums were always proportional to initial risk (perfect risk rating and therefore constant percentage loading) we would in theory expect initial purchases to be independent of the level of risk as long as household income was high enough that the premium itself was "affordable" in one of several senses.4
If risk were then to increase with the onset of a costly chronic condition, whether coverage would continue would depend on what would happen to the net premium the person would have access to. Here is where the answer to the question of developing plausible hypotheses is somewhat murky. For individual insurance with guaranteed renewability, the most plausible hypothesis is at least somewhat clear: if the premium is guaranteed not to jump when risk increases, the person who becomes a higher risk should if anything actually be more likely to continue coverage than the person who remains a low or average risk. For group insurance, a key determinant of future premiums and coverage is employment status. If workers can hang onto the job, they can continue to obtain group coverage at a total net of tax premium (taking account both of explicit premiums and wage offsets) that remains low and constant. Moreover, as Madrian5 has shown, people who become higher risks who can keep a given job with coverage tend to do so: there is "job lock" for higher risks especially. But if reasonably good health is needed to retain employment, the person who becomes a higher risk and can no longer hold a job will surely face a very large increase in potential premium: the group coverage will disappear as an option (at least by the end of any COBRA continuation) and the only options in most states will be individual insurance at very high medically underwritten premiums, or a high risk pool at high premiums (if the pool is accepting more members). Moreover, in many states a small group's premium can be subject to a large increase if a worker or dependent becomes higher risk.6 Thus the impact on the likelihood of retaining coverage under group insurance depends on a host of conflicting influences: the low future premium for the person who keeps their job, the stronger incentive to hold onto that job, but the much higher premium for the person who loses their job. It may also matter whether the "owner" of the insurance is the working family member who is a high risk; if the unhealthy person is covered under a small group family plan owned by another household member, there will be an incentive to hang onto group coverage, but also the possibility of a large increase in experience rated premiums. What will actually happen in the face of these conflicting incentives and circumstances is an empirical question.
DATA AND MODEL SPECIFICATION
We use the nationally representative SIPP data which covers the period 2000-2004. Participants are interviewed every four months. The sample is split into four groups, so that every month, one fourth of the participants are interviewed about the preceding four months. If households, or adults (defined as individuals 15 or older) within a household, move they are followed to their new address. The data in the SIPP is split into two groups -- core data (including employment and health insurance status) which is asked at every interview, and topical data, which is asked yearly (or less frequently).7 Our sample consists of people who reported being employed at the observation of initial insurance coverage; it includes both men and women workers, and among those with individual coverage are both self employed workers and other workers who bought individual rather than group coverage, generally because the latter was not offered. Workers may be covered either as the primary insured or owner of the policy or as a dependent on a family policy. We wanted to include workers who were dependents since their health status would affect risk-rated family premiums for individual insurance, at least initially. We combine the periodic waves into annual intervals, since most insurance policies provide coverage for 12 months. (There are some explicitly temporary individual policies which can be taken for various intervals of time and do not carry guaranteed renewability, but their market share is small.)
While individuals in the SIPP can in principle be followed for some time, both attrition and complexity grow the longer we try to follow a given individual. Attrition is especially a problem for the individual market which, even in this large sample data set, starts with a relatively small number of observations and goes down from there. Accordingly, we are able to look at insurance purchasers who were documented as being insured only for one year (that is, observed for two years); however, this is long enough for renewal to be an issue. We then ask: given that we have observed a person with a given type of insurance for one year, what determines the probability that the person will be uninsured in the next period?
In addition to the initial type of insurance, we also have measures of household income, change in income, sex, and the level of the worker's health status in the initial period. Over a two-year time frame there are too few changes in health status to justify including change in worker health status as a regressor, and in preliminary specifications the change in health status was not statistically significant. However, many of the workers who are in worse health states made the transition from lower (if not low) risk in the recent past.
Table 1 shows means for the sample we study, the sample of people who had a given type of coverage for at least one year. Many of the people in this sample probably had that coverage for longer than one year, but we can only track their experience from the first time we observe them.
The most interesting finding in Table 1 is the proportion in each insurance setting who became uninsured at the end of the indicated period. As expected, that proportion is lowest for people in large groups and highest for those with individual coverage. The average household income is considerably higher for those who work in large groups, and the individual coverage sample is disproportionately female.
The large group sample had so low a rate of dropping coverage that formal analysis would not be meaningful. This reinforces a well known proposition: large group coverage is good if you can get it (and good [high-paying] large firm jobs carry good insurance). We therefore concentrated on comparing persistence and its determinants in the small group versus the individual samples.
To understand the determinants of dropping coverage in each sample, we estimated logit regressions that relate the (log) odds of dropping or losing coverage to various covariates, including health status as a measure of risk level. Health state was transformed into a continuous variable (with "poor" being one and "excellent" being five.
All regressions were statistically significant according to the Wald test. The multivariate analysis indicated that income, sex, and age are significantly related to the odds of dropping coverage. Higher income and older workers in both the small group and individual samples were less likely than younger workers to drop. Those with a larger increase in wages were less likely to drop all coverage in the group sample while there is no effect of risk in the individual sample.
Foreshadowing our general results, being in worse health substantially increases the odds of dropping coverage for small group insurance for both samples. That is, the worse is health status, the greater is the likelihood of losing all coverage in the small group market (higher relative probability). Regression coefficients from the logit regressions are shown in Appendix A.
We also explored another specification in which health state was a binary, with those with "fair" or "poor" health in one category, and those with "good," "very good," and "excellent" in the other. The measures of goodness of fit were less good for this specification but the overall pattern was the same as that shown with the continuous measure: people initially with small group coverage who were in fair or poor health were more likely than those in good to excellent health to lose coverage, whereas there was no effect of health status for those with individual coverage.
Because the relationships of variables are complex in the logit functional form, and the numerical magnitudes of regression coefficients hard to interpret, we present some sample results derived from these regressions by a simulation. We simulate the probability of losing coverage as a function of health status for three hypothetical workers: a middle aged male, a young male, and a young female worker.
In the small group sample, male workers were very rarely covered under a spouse's plan, so their relationship between risk and dropping coverage will be affected by ability to work. Table 2a and Table 2b show the estimated probabilities that a younger and older male who had insurance for at least one year dropped all insurance coverage in the subsequent period. Table 2c shows the results for young women, who have a lower probability of dropping then young men but who display the same health effect in the small group market. At the median health status ("very good"), as expected, the dropout rates for small group coverage were consistently lower than those for individual coverage. However, the most striking finding is that this order is reversed for workers with health status "fair" or "poor." Sicker workers initially with group coverage are more likely to lose any and all coverage compared to healthier workers in the same setting. They were also more likely to lose coverage compared to similarly sick workers with individual coverage. That is, in small group insurance, less healthy workers are relatively more likely to lose coverage than healthier workers and are also absolutely more likely to lose coverage when they are in the small group setting than when they are in the individual setting.
In the simulations as in the regressions, the individual coverage dropout rates do not vary strongly with health status. There is an apparent "protective" effect of lower health status, but this is based on a coefficient that is not statistically significant.
The interpretation of the relationship of health state to losing coverage for those with group insurance is reasonably clear for those (almost exclusively male) workers who were the primary source of insurance coverage: the most likely reason why these workers lose coverage when they are in poor health is that they cannot continue working. However, the relationship might be different for those workers who were covered as dependents under family coverage, primarily working wives in two worker households, since poor health of a working spouse covered as a dependent does not directly impact the primary insured's job tenure (though some workers may need to quit to care for a sick spouse). We therefore further investigated the relationship between health status and continuation of coverage for the subsample of workers initially with small group coverage who were not the "owners" of insurance coverage. We found effects very similar in sign to the full sample; the magnitudes of the impact of health state were a little lower (but still significant). However, female workers with small group coverage were less likely than males to drop or lose coverage at all health state levels.
DISCUSSION
The results reported above are consistent with two frequently debated and somewhat conflicting empirical propositions: that individual insurance is more costly than group insurance, but that individual insurance provides better protection for people who become high risks than does group insurance. Specifically, for an initially insured person of average or better risk, dropping back (or out) to no health insurance coverage is more likely if the coverage was expensive individual insurance than if it was cheaper and tax subsidized group insurance. But group insurance has a tear in its net of protection: it leaves a person who becomes sick, possibly too sick to work at the same job, more vulnerable to losing any and all coverage.
The point estimate that a young male high risk who had small group coverage faces a 44% chance of becoming uninsured in the next period, a risk nearly twice times as great as if he had individual insurance, might be large enough to get policy attention. Somewhat ironically, the usual blame for such a person lacking coverage will be laid at the door of the medically underwriting individual insurer which quotes a high premium, rather than referred to the group insurance system that plunged this person into such a vulnerable situation in the first place.
In summary, neither the individual nor the group insurance system is strictly preferable to the other; they both have advantages and disadvantages. It would be possible to design a group policy that did provide the protection it now lacks: build in guaranteed renewability at low risk premiums for small group insureds upon conversion to individual coverage. However desirable this feature might be from a social perspective, and despite the fact that group-to-individual conversion is offered by some insurers, it might be a hard sell to many employers. Employers may want to offer insurance to attract employees if and only if they continue to work for the employer; the employer may balk at appearing to pay more up front to provide protection for employees who quit or get fired. Perhaps wise employers might see that by offering workers a less risky framework for insurance purchasing they could get better workers for the same money wage, or could get workers to sacrifice even more alternative compensation for such an attractive benefit.
At least in the short run, leveling the tax playing field between individual and group coverage might be help with (though not solve) this problem, as with many others. At a minimum, some of the negative perception of individual insurance may be undeserved.
TABLES
| TABLE 1: Sample Means, SIPP Data on Workers with At Least One Year of Continuous Coverage | |||
| One Year of Continuous Coverage | Individual Coverage | Small Group Coverage | Large Group Coverage |
| Count | 620 | 1,675 | 14,011 |
| Average Age | 41.56 | 39.18 | 40.84 |
| Average Health Status* | 2.04 | 1.98 | 2.01 |
| Average Household Income | $59,411 | $65,186 | $74,898 |
| Percent Male | 0.41 | 0.55 | 0.51 |
| Percent Dropping Coverage After One Year | 0.17 | 0.13 | 0.06 |
| * 1 = Excellent, 2 = Very Good, 3 = Good, 4 = Fair, 5 = Poor | |||
| TABLE 2a: Male Age 28, Earning $50,000 Annually, Expecting an 8% Raise(Probably of Uninsurance After One Year of Continuous Coverage) | ||
| Health Status | Individual Insurance | Small Group Insurance |
| Excellent | 0.23 | 0.13 |
| Very Good | 0.23 | 0.18 |
| Good | 0.24 | 0.25 |
| Fair | 0.24 | 0.33 |
| Poor | 0.24 | 0.43 |
| TABLE 2b: Male Age 45, Earning $80,000 Annually, Expecting a 5% Raise(Probably of Uninsurance After One Year of Continuous Coverage) | ||
| Health Status | Individual Insurance | Small Group Insurance |
| Excellent | 0.16 | 0.05 |
| Very Good | 0.16 | 0.07 |
| Good | 0.16 | 0.10 |
| Fair | 0.16 | 0.15 |
| Poor | 0.17 | 0.20 |
| TABLE 2c: Female Age 28, Earning $50,000 Annually, Expecting an 8% Raise(Probably of Uninsurance After One Year of Continuous Coverage) | ||
| Health Status | Individual Insurance | Small Group Insurance |
| Excellent | 0.17 | 0.09 |
| Very Good | 0.17 | 0.13 |
| Good | 0.17 | 0.18 |
| Fair | 0.17 | 0.25 |
| Poor | 0.17 | 0.34 |
APPENDIX A
| TABLE A1: Logistic Regression of 1 Year Probability of Dropping Small Group Coverage | ||||||
| Number of observations | = | 1,675 | ||||
| Wald X2 (5)(p-value) | = | 102.04(<0.001) | ||||
| Log pseudolikelihood | = | -591.723 | ||||
| Pseudo-R2 | = | 0.101 | ||||
| Odds Ratio | Robust Standard Error | z-score | p-value | 95% Confidence Interval | ||
| Lower Bound | Upper Bound | |||||
| Age | 0.962 | 0.008 | -4.71 | <0.001 | 0.947 | 0.978 |
| Health Status | 1.508 | 0.128 | 4.85 | <0.001 | 1.278 | 1.781 |
| Log wage | 0.393 | 0.057 | -6.45 | <0.001 | 0.296 | 0.522 |
| Log wage change | 0.542 | 0.082 | -4.06 | <0.001 | 0.403 | 0.729 |
| Sex | 1.486 | 0.244 | 2.41 | 0.016 | 1.077 | 2.051 |
| TABLE A2: Logistic Regression of 1 Year Probability of Dropping Individual Coverage | ||||||
| Number of observations | = | 620 | ||||
| Wald X2 (5)(p-value) | = | 16.03(0.007) | ||||
| Log pseudolikelihood | = | -276.837 | ||||
| Pseudo-R2 | = | 0.031 | ||||
| Odds Ratio | Robust Standard Error | z-score | p-value | 95% Confidence Interval | ||
| Lower Bound | Upper Bound | |||||
| Age | 0.981 | 0.009 | -1.98 | 0.048 | 0.963 | 1.000 |
| Health Status | 1.013 | 0.121 | 0.11 | 0.911 | 0.802 | 1.280 |
| Log wage | 0.738 | 0.083 | -2.72 | 0.007 | 0.593 | 0.919 |
| Log wage change | 0.999 | 0.088 | -0.01 | 0.993 | 0.841 | 1.187 |
| Sex | 1.517 | 0.355 | 1.78 | 0.075 | 0.959 | 2.402 |
NOTES
-
M.A. Hall, "Perspective: Of Magic Wands and Kaleidoscopes: Fixing Problems in the Individual Market," Health Affairs 21 (2002): w353-w358 (published online October 23, 2002; 10.1377/hlthaff.w2.353).
-
E.C. Ziller, A.F. Coburn, T.D. McBride, and C. Andrews, "Patterns of Individual Health Insurance Coverage, 1996-2000," Health Affairs 23, no. 6 (2004): 210-221.
-
M.V. Pauly and L.M. Nichols, "10 Myths of the Uninsured," AHIP Coverage 45, no. 3 (2004): 16-22.
-
M.K. Bundorf and M.V. Pauly, "Is Health Insurance Affordable for the Uninsured?" Journal of Health Economics 25, no. 4 (2006): 650-673.
-
B.C. Madrian, "Employment-Based Health Insurance and Job Mobility: Is There Evidence of Job-Lock?" Quarterly Journal of Economics 109, no. 1 (1994): 27-54.
-
M. Freudenheim, "When Even One Illness Can Push Insurance Costs Up," The New York Times (May 5, 2007), C1.
-
U.S. Department of Commerce, Economics and Statistics Administration, U.S. Census Bureau, "Survey of Income and Program Participation User's Guide (Third Edition)," 2001, http://www.sipp.census.gov/sipp/usrguide/sipp2001.pdf (accessed March 27, 2007).