Analysis of Transition Events in Health Insurance Coverage

08/15/2009

Contents

Over time, the number of people lacking health insurance coverage in the U.S. is sustained by a set of dynamic processes. Comparatively few of the uninsured remain in that state indefinitely, but uninsured persons who gain coverage are offset by insured persons who lose their coverage. In good economic times the balance shifts toward the gainers, and uninsured rates tend to decline. In weak times, the balance shifts in the reverse direction, and uninsured rates tend to rise. Migration plays a role as well. New immigrants have much higher uninsured rates than long-term residents, but so do those who leave the population, which dampens the effect of immigration. Achieving a significant reduction in the number of uninsured persons will require reducing the rate at which people lose coverage or increasing the rate at which people (re)gain coverage  or, ideally, both. From a policy perspective, this should focus attention on the factors that contribute to people losing or gaining coverage, yet true longitudinal analyses of these factors are rare.

Recognizing this limitation of our research base, the Office of the Assistant Secretary for Planning and Evaluation (ASPE) in the Department of Health and Human Services contracted with Mathematica Policy Research, Inc. (MPR) to conduct a study focusing explicitly on the dynamics of health insurance coverage. The study had four main components: (1) a literature review; (2) methodological work on the 2001 panel of the Survey of Income and Program Participation (SIPP) to identify and address limitations that represent potential sources of bias in estimates of health insurance dynamics; (3) a descriptive (or tabular) analysis of SIPP panel data to document aspects of the dynamics of health insurance coverage and (4) a multivariate analysis of events associated with transitions in health insurance coverage. The literature review and the methodological findings are presented in appendices to the report and not discussed in this summary.

DATA

The analyses presented in this report use data from the 1996 and, primarily, 2001 panels of the SIPP. The 2001 panel followed a sample of nearly 30,000 households for three years. With its collection of monthly data on health insurance coverage, employment, income, family composition, and a wide range of other potential covariates of health insurance coverage, SIPP is unique in its ability to support analysis of the dynamics of health insurance and the relationship between transitions in coverage and changes in employment, income, and family composition.

[ Go to Contents ]

HEALTH INSURANCE DYNAMICS

The analysis sample for most of the research presented in this report consists of persons who were 19 to 61 in January 2001. In that month, 17.6 percent of this population lacked health insurance coverage. Over the 36-month reference period of the 2001 SIPP panel, roughly twice that fraction (35.0 percent) had some amount of time in which they were not covered by health insurance. How much time? Nearly 60 percent were without coverage for 12 months or more, and more than a third were uninsured for at least two of the three years. These include 4.5 percent who were uninsured the entire time. At the other end of the spectrum, 23 percent were without coverage for four months or less  a very modest amount, to be sure, which may also include some false (misreported) uninsured spells.

The likelihood of being without coverage declined progressively with age. Among persons 19 to 29 in January 2001, 27 percent were uninsured in that month and 54 percent were ever uninsured during the three years. By ages 51 to 61, 11 percent were uninsured in January 2001, and 22 percent were ever uninsured in the three years.

About a third of those who were ever uninsured had more than one uninsured spell during the reference period. Multiple spells were much more common among the young than among the most senior adults. If we exclude persons who were continuously uninsured from those with a single spell, we find that those with two or more spells were without coverage for a longer period of time than those with just a single spell.

Family income relative to the poverty line is the single strongest predictor of insured status when measured at a point in time. The uninsured rate in January 2001 ranged from a high of 42 percent among people below poverty to a low of 5 percent among people above 400 percent of poverty. We find that relative income in the first year of the 2001 panel is very strongly associated with coverage over the duration of the panel as well. The fraction ever uninsured was 68 percent among persons below poverty in the first year and declined to 14 percent among persons above 400 percent of poverty. Nevertheless, the substantially greater numbers of higher- versus lower-income persons yielded a surprising result-namely, that persons with family incomes above 200 percent of poverty in 2001 accounted for just over half of the ever uninsured and just over half of new insured and uninsured spells.

Race and ethnicity are also strong covariates of health insurance coverage in both the cross-section and longitudinally. Similar to the poor, 41 percent of Hispanics were uninsured in January 2001, and 65 percent were ever uninsured in the three years. The corresponding uninsured rates for non-Hispanic whites were 13 percent and 28 percent, respectively. In addition, while Hispanics were just 13 percent of the nonelderly adult population, they accounted for 31 percent of those who were uninsured the entire 36 months.

While the 2001 panel was conducted during a period that included a brief recession followed by a slow recovery, the 1996 SIPP panel was conducted during a period of sustained economic growth. How did the picture of health insurance dynamics differ between the two surveys? Uninsured rates for nonelderly adults were flat during the 2001 panel, whereas they declined by three percentage points in the 1996 panel. More new spells  both uninsured and insured  were started during the 2001 panel than the 1996 panel, but, consistent with the flat trend in uninsured rates, the numbers of the two types of spells were nearly identical. During the 1996 panel, the number of new insured spells exceeded the number of new uninsured spells by 4.5 million. Other differences between the panels were mixed, however, which suggests a complex underlying relationship between economic factors and health insurance transitions. For this reason we did not extend the two-survey comparison to the multivariate analysis of transition events.

The proportion of persons retaining the same type of insurance coverage between one interview and the next (four months later) reflects not only actual retention but reporting accuracy, which appears to vary by source. For coverage from a current employer, 90 percent reported the same coverage four months later; 7 percent reported another type of coverage; and 3 percent reported being uninsured. For private nongroup coverage, 65 percent reported the same coverage four months later; 28 percent reported a different type of coverage; and 7 percent reported being uninsured. For Medicaid, 79 percent reported the same coverage four months later; 8 percent reported a different type of coverage; and 13 percent  the highest among all sources  reported being uninsured. Among the uninsured, 80 percent reported that they were still uninsured four months later while 20 percent reported that they had health insurance.

Of the new uninsured spells that started during the first year of the 2001 panel, 49 percent were preceded by coverage from a current employer or union, 22 percent by public coverage, 14 percent by coverage from a former employer, 13 percent by nongroup or other private coverage, and 2 percent by military-related coverage. Uninsured spells preceded by coverage from a current employer or union or by public coverage had strikingly similar durations. Spells preceded by nongroup coverage ran a little longer while those preceded by coverage from a former employer ran a little shorter. Of the uninsured spells that ended during the final year of the panel, 56 percent were followed by coverage from a current employer or union; 26 percent by public coverage; 13 percent by nongroup or other coverage; 4 percent by coverage from a former employer; and 2 percent by military-related coverage. Here, too, the length of the uninsured spells did not vary between spells followed by coverage from a current employer or union versus public coverage.

[ Go to Contents ]

OBTAINING COVERAGE:
TRANSITIONS OUT OF THE UNINSURED STATE

Previous research on health insurance transitions has used logistic regression models that describe the likelihood of transitioning between two states, such as from being uninsured to being insured. With such models there is a single origin state and a single destination state. Leaving the origin state implies entering the lone destination state. In our work we use a multinomial logistic regression framework to model jointly the transitions from the uninsured state to any of five destination states representing alternative sources of coverage. In this way the full choice set is incorporated into the model. Empirically, we estimate transitions between consecutive waves of the SIPP, with each person contributing an observation for each wave (1 through 8) that he or she was uninsured and remained in the sample through the next wave.

For the uninsured, gaining a job is strongly associated with obtaining coverage through a current employer or union, a private nongroup source, or a former employer. While the association with coverage from a current employer implies that health insurance is provided by the new employer, the association with coverage from a former employer or nongroup source suggests that some new employees find it necessary to acquire coverage elsewhere. Changing jobs, regardless of whether earnings increase or decrease, is almost as strongly associated with obtaining coverage through a current employer or union but no other source. Losing a job, however, is negatively associated with the likelihood of obtaining coverage from a current employer or union or a private nongroup source but positively associated with the likelihood of obtaining coverage from a public source (Medicare, Medicaid or another state program) or even a former employer. Presumably the loss of employment and, with it, earnings increases the odds of qualifying for public coverage. Changes in family earnings are also strongly related to obtaining coverage, mostly through a current employer or union. An increase in family earnings may make current employer or union coverage more affordable whereas a decrease in earnings makes it less affordable.

Net of trigger events, demographic characteristics remain strongly associated with transitions out of the uninsured. As education increases, persons are progressively more likely to leave the uninsured for every source but public coverage. Compared to the childless, persons with children are more likely to acquire coverage through Medicaid but less likely to obtain coverage from a nongroup source, former employer, or the military. Compared to Hispanics, white and black non-Hispanics are more likely to leave the uninsured state for every source of coverage. With increasing family income, individuals are more likely to obtain coverage through a current employer or union or a private nongroup source and are less likely to obtain coverage through a public source.

[ Go to Contents ]

CHANGING AND LOSING COVERAGE:
TRANSITIONS INTO THE UNINSURED STATE

To analyze transitions into the uninsured state, we use a two-step estimation procedure. The first step models the decision to leave the current health insurance coverage type (again, one of five), relative to keeping it, and the second step models the decision to transition to the uninsured state, relative to obtaining an alternative type of insurance coverage, given that the individual leaves the current coverage type. In both steps, a separate logistic regression model is estimated for each coverage type. For example, in the first step, transitions out of coverage from a current employer or union are modeled separately from transitions out of the other four types of coverage. In the second step, transitions into the uninsured state from current employer or union coverage are modeled separately from transitions into the uninsured from another source of coverage. Dividing the transition from covered to uninsured into two steps allows us to model separately the decision to leave the current coverage type and the decision to obtain an alternative coverage type or become uninsured. Conceptually, this provides a richer behavioral framework. As above, we estimate transitions between consecutive waves of the SIPP, with each person contributing an observation for each wave (1 through 8) that he or she was insured and remained in the sample through the next wave.

Losing one's job is very highly correlated with leaving coverage through one's current employer or union and, conditional on leaving, with becoming uninsured rather than entering an alternative source of coverage. Individuals who experience a decrease in earnings without losing their jobs are also more likely to leave coverage through a current employer or union. This is true without regard to whether the reduced earnings are accompanied by a change in family size. The higher likelihood of leaving may be due to a decrease in hours worked that makes the individual ineligible for health insurance benefits. The reduced earnings may also make the coverage too costly if it continues to be offered.

Moving from one job to another, regardless of the impact on earnings, is associated with leaving coverage from every source but the military, but the association is particularly strong for coverage from a current employer or union or a former employer. For these two sources, conditional on leaving, a job change also increases the chances of becoming uninsured. Together, these relationships highlight the prevalence of waiting periods for health insurance coverage at new jobs or provide evidence that individuals are accepting jobs without coverage.

Job gains and earnings increases are associated with leaving public coverage. This is expected, as the earned income obtained from new employment may exceed program eligibility thresholds, or the new job may offer employer-sponsored coverage that is superior to public coverage or less costly than the coverage available from a former employer. In addition, increased earnings, without an increase in family size, make an individual more likely to leave public coverage but, given that they do so, decrease the chances that the individual will become uninsured. We do not find the same association for similar individuals whose increase in earnings is coupled with an increase in family size. This may reflect in part the higher income eligibility thresholds that public programs apply to larger families.

Net of trigger events, key demographic characteristics are strongly associated with the likelihood of leaving current coverage. With increasing age, people are less likely to leave any source of coverage, but among those who do leave, the relationship between age and becoming uninsured varies with the source of coverage that was terminated. White and black non-Hispanics are less likely than Hispanics to leave any source of coverage except, for blacks, nongroup coverage. Conditional on leaving current employer or union coverage, public coverage, or private nongroup coverage, non-Hispanics are less likely than Hispanics to become uninsured. With increasing education, people are less likely to leave any nonpublic source of coverage but more likely to leave public coverage. Regardless of the source they left, the more educated are less likely to become uninsured. Family income behaves similarly to education except that there is no association between income and the likelihood of leaving private nongroup coverage. Conditional on leaving nongroup coverage, however, people with more income are less likely to become uninsured.

[ Go to Contents ]

CONCLUSION

The analyses presented in this report have implications for policy, and they also suggest a number of priorities for continuing research.

Policy Implications

Our findings have implications that policymakers need to understand if they are to develop effective policies for increasing the proportion of the United States population with adequate health insurance coverage.

  • The biggest problem that policymakers must address is how to help people retain coverage once they have it and how to restore coverage more quickly to those who have lost it.
  • As numerous as the persons who are without coverage in a three-year period may be, their numbers provide only a partial measure of the size of the population that was at risk of losing coverage during that period. Policymakers must address the risk to the larger population to minimize the losses that contribute to the numbers of uninsured.
  • Almost half of those who were ever without coverage in a three-year period were under 30. The low cost-effectiveness of health insurance for this group is almost certainly a factor in their low coverage and one that policymakers must address.
  • While the poor and near poor are much more likely than those with higher incomes to experience extended periods without health insurance coverage, losing health insurance coverage is neither exclusively nor primarily a low-income problem. Policymakers must address the needs of both the lower-income and higher-income uninsured.
  • With our focus on persons losing and regaining coverage, we should not overlook the hard core uninsured-those represented by persons who were continuously uninsured in our study; the problem that they present for policy is different than the problem presented by people who are in and out of coverage.

Research Priorities

We suggest several areas where additional analysis building on the findings presented here would help to further our understanding of why people lose coverage, how they gain or regain coverage, and why they change their source of coverage.

  • We found our modeling approach to the multivariate analysis of transition events to be particularly informative about the association between trigger events and transitions out of and into the uninsured state; useful extensions include explicit modeling of the separate coverage provided by married partners and the inclusion of additional trigger events based on employment changes by the spouse.
  • A focused analysis of what distinguishes those persons who appear to remain outside the health insurance system would be useful in helping policymakers to better understand this group.
  • Research is also needed that will help us to understand the mechanism whereby more than half of those who lost health insurance coverage over the length of the 2001 SIPP panel had 2001 calendar year incomes above 200 percent of poverty.
  • We remain concerned that SIPP obtains too many transitions and that a significant number of one-wave spells may be erroneous; further investigation of this issue is needed and should include the 2004 panel, which incorporated a major innovation to the survey instrument to reduce erroneous transitions of all types.