This Article Abstract Submit a response to this article Services E-mail this article to a friend Alert me to new issues of the journal

H E A L T H  T R A C K I N G
T R E N D S  

5 April 2005  
  

It’s The Premiums, Stupid: 
Projections Of The Uninsured 
Through 2013

The number of uninsured Americans is projected to
increase by eleven million in the coming decade.

By Todd Gilmer and Richard Kronick

ABSTRACT:

Increases in the cost of health care from 1979 to 1999 accounted for the decline in health insurance coverage that occurred during that time period, as our earlier work demonstrated. Here we examine whether the model we presented adequately accounts for the observed changes in health insurance coverage from 1999 through 2002, and we show that the model accurately predicted the increase in uninsured people during that time period. Using the model and projections for national health spending, we project that the number of nonelderly uninsured Americans will grow from forty-five million in 2003 to fifty-six million by 2013.

In earlier work we demonstrated that increases in the cost of health care from 1979 to 1999 accounted for the decline in health insurance coverage that occurred during that time period.¹ In this paper we examine whether the model we presented adequately accounts for the observed changes in health insurance coverage from 1999 through 2002, and we use national health spending projections to predict changes in insurance coverage over the next decade.

The Price-Coverage Relationship

The standard economic theory of health insurance posits that the desire for insurance represents an attempt by risk-averse people to maximize utility in the face of uncertainty, subject to a budget constraint: People do not know whether they will be sick or healthy and would often rather take a certain, moderate loss by paying an insurance premium instead of taking the chance of a large financial loss from serious illness.²

Research into the causes of health spending increases points to two primary factors: improvements in medical technologies and increased use of these technologies.³ Either factor could be valued by consumers, who might then be willing to purchase insurance to cover the additional benefits of treatment as well as the increased likelihood of treatment. However, this demand is tempered (or constrained) by the fact that purchasing insurance decreases the amount of money available for the purchase of other goods and that as the share of income required to pay premiums rises, consumers must sacrifice more and more consumption to remain insured. Thus, those who are dropping insurance do so mainly as the result of budgetary pressures. As a result, among low- and moderate-income people, we expect that as health insurance premiums increase, the number of uninsured people will also increase.⁴

Study Data And Methods

Using data from the National Health Accounts constructed by the Office of the Actuary at the Centers for Medicare and Medicaid Services (CMS), we follow the procedure described in earlier work to estimate per capita health spending for insured adult workers from 1979 to 2002.⁵ We have revised our methods slightly by using data collected in 1999 as part of the Medical Expenditure Panel Survey (MEPS) as well as the data used in our earlier work from the 1996 MEPS, the 1987 National Medical Expenditure Survey (NMES), and the 1977 National Medical Care Expenditure Survey (NMCES) to estimate the proportion of total spending accounted for by insured people ages 19–64.⁶

Using data from the March supplements to the Current Population Survey (CPS) from 1980 through 2003 and the revised price series, we calculated the ratio of per capita health spending to median income and the percentage of nonelderly workers covered by health insurance (among those workers who were not covered by a spouse or a public program).⁷ Starting in 2000, an “insurance verification” question was added to the CPS, resulting in a decline in the estimated number of uninsured people. We adjusted the estimates of health coverage in the earlier data to make them consistent with the post-2000 data series.⁸

We reestimated the logistic regression model from our earlier work using data from 1987 through 2002. This regression was modified slightly by expanding information on ethnicity.⁹ We also included a covariate for those owning a home as an indicator of assets that one would wish to protect against large health care claims. The logistic regression model predicts the probability of coverage among nonelderly adult workers as a piecewise linear function (spline) of per capita health spending divided by income; employment characteristics such as firm size, industry, part-time status, and self-employment status; demographic and socioeconomic characteristics such as age, sex, marital status, race, ethnicity, education, and home ownership; and indicators for region and year.¹⁰

We estimated the 2002–2013 rate of growth in per capita health spending for insured adults and the growth rate in personal income using data from the national health spending projections produced by the CMS.¹¹ We then used the parameter estimates from the logistic regression model to estimate the 2013 insurance status of each worker in the March 2003 sample of the CPS, assuming that per capita spending and personal income follow the estimated growth curve.

Finally, we used the historical relationship between the proportion uninsured among workers (not covered as a dependent or by a public program) and the proportion uninsured among the entire under-age-sixty-five population, to estimate the proportion of people in this age group expected to be uninsured in 2013. The percentage uninsured among all adults (ages 18–64) closely parallels the percentage uninsured among adult workers during 1979–2002 (Exhibit 1). The ratio of the percentage uninsured among all adults to the percentage uninsured among adult workers increased from 0.76 in 1979 to 0.82 in 2002. We fitted a least squares regression line to the 1979–2002 experience and estimated that the ratio of the percentage uninsured among adults to the percentage uninsured among adult workers will be 0.84 in 2013.¹² To estimate the percentage of adults who will be uninsured in 2013, we multiplied our projection for the percentage of adult workers who will be uninsured in 2013 by 0.84.

Although the relationship between the percentage uninsured among adults and the percentage uninsured among adult workers (not covered as a dependent or by a public program) was relatively stable over the 1979–2002 period, the relationship between the percentage uninsured among children and the percentage uninsured among adult workers is more variable. The series for the percentage uninsured among children and among workers track each other closely from 1979 through 1987, but as the percentage uninsured among workers increased by four percentage points from 1987 through 1996, among children it increased by only two percentage points; the more modest change in the latter is largely the result of the Medicaid eligibility expansions of the late 1980s.¹³ The effects of the State Children’s Health Insurance Program (SCHIP), enacted in 1997, on increasing coverage among children are apparent in the data, and they result in a decrease in the percentage of children who were uninsured in the 1998–2002 period, at a time when uninsurance was rising among adult workers. Controlling for the effects of SCHIP in a multivariate analysis, we estimated that for each one-percentage-point increase in the percentage uninsured among adult workers over 1979–2002, there is a 0.45 percentage point increase in the percentage uninsured among children.¹⁴ We used this relationship and our projection of the increase in the percentage uninsured among adult workers from 2002 to 2013 to estimate the increase in the percentage uninsured among children in 2013.

Results

Per capita health spending for nonelderly insured adults divided by the median income of nonelderly adult workers increased from 8.8 percent in 1999 to 10.9 percent in 2002 (Exhibit 2). This rapid decline in the “affordability index” resulted from a steep increase of 9.8 percent per year in per capita health spending from 1999 through 2002, while growth in median personal income was notably less—2.2 percent per year. Mirroring the decline in affordability almost exactly, the percentage uninsured among nonelderly workers increased 1.5 percentage points, from 22.3 percent in 1999 to 23.8 percent in 2002.

Consistent with results from our earlier work, we find that changes in the distribution of employees across industries have had very little effect on coverage rates. There has been some shift away from manufacturing jobs in large firms, and this shift has resulted in small declines in coverage. Manufacturing jobs declined from 1987 to 2002 (from 22 percent of workers in 1987 to 16 percent in 2002), and workers in manufacturing jobs are slightly more likely to have employer-sponsored insurance (ESI) than most other workers (controlling for other characteristics, jobs in durable goods are 3.4 percent more likely to have ESI than jobs in transportation, the reference category) (Exhibit 3).¹⁵ However, the net effect of fewer manufacturing jobs is quite modest: We estimate that insurance coverage among workers is 0.2 percent lower because there are fewer manufacturing jobs. And much of the increase in employment is in professional services, which is a relatively high-coverage industry in any case.

Further, while increases in the proportion of Hispanic workers from 1987 to 2002 should have led to declines in coverage, increases in educational levels and increasing proportions of older workers should have led to increased coverage. The combined effects of demographic and employment-related changes from 1987 to 2002 should have been to raise coverage rates by 0.5 percent, a far cry from the 6.2 percent decline that actually occurred.

In Exhibit 4 we provide a plot of the predicted percentage insured against the affordability index (price/income). From 1987 to 2002, the affordability index declined from 5.6 percent to 10.9 percent, and this shift accounts for most of the observed decline in coverage. Declines in coverage have resulted, for the most part, from increases in premiums that were much more rapid than increases in personal incomes.

Although there have been substantial changes in the workforce since 1987, these changes have had little effect on rates of insurance coverage. We expect future changes to have similarly small effects. Therefore, instead of attempting to make projections of changes in industry composition and firm size and using these estimates in our projections for coverage over the next decade, we based our coverage projections on expected changes in health spending and personal income, holding our other control variables constant at their 2002 level (including the year covariates).

Based on CMS spending projections, we estimate that per capita spending for insured adults will grow 7.4 percent per year from 2002 to 2013 and that personal income will grow 4.6 percent per year over the same period.¹⁶ Although health spending is projected to grow more quickly than personal income over this period, the projected difference in growth rates over the next decade is not nearly so large as the actual experience during 1999–2002, when health spending grew 9.8 percent annually, and income, only 2.2 percent.

Based on the projected growth rates for health spending and personal income, we estimate that the rate of uninsured nonelderly workers will increase by 4.0 percentage points to 27.8 percent in 2013. We estimate that the uninsurance rate among all nonelderly Americans will increase by 3.3 percentage points to 20.5 percent in 2013. With an expected population of 271 million people under age sixty-five in 2013, we estimate that there will be fifty-six million uninsured Americans in this age group, an increase of thirteen million over the CPS estimate for 2002. Of this estimated increase, 8.6 million occurs because of the expected increase in the proportion of the population that is uninsured, and 4.4 million because of an increase in population size.

It is difficult to accurately project health spending growth rates, and we performed sensitivity analyses on our results. If health spending rises 4.6 percent per year from 2002 to 2013 instead of the the projected 7.4 percent per year, then the affordability index would be virtually unchanged, and the percentage uninsured would be virtually unchanged from its current level. Conversely, if health spending rises 9.4 percent per year, then we estimate that there would be sixty-three million uninsured people in 2013—a very large increase over the current level.

We estimate that for each 1 percent increase in health spending (relative to personal income), the number of uninsured people will increase by 246,000. This estimate is similar to an earlier estimate from the Lewin Group, using substantially different methods.¹⁷

Discussion

In 2001 we published a paper titled “The Calm Before the Storm,” projecting that the number of uninsured Americans would increase rapidly as a result of health insurance premiums rising more rapidly than personal incomes. It is clear that a full-force storm blew in from 1999 to 2002 and that the actual path of the percentage uninsured, as shown in Exhibit 2, followed the path we expected based on our estimated relationships between affordability and coverage.

We expect the storm to continue during the next decade. However, based on CMS spending projections, its ferocity will lessen. The CMS predicts that health spending for insured people under age sixty-five will grow 2.4 percent per year faster than personal incomes from 2002 to 2013. At this “moderate” rate of spending growth, we expect that the number of uninsured people will increase “only” thirteen million over the next decade from its 2002 level, or eleven million from the estimate of forty-five million uninsured people in 2003.¹⁸ Based on estimates from the Institute of Medicine, this will lead to an increase of 4,500 deaths annually and to an increased annual loss of human capital of $16–$32 billion.¹⁹ If the estimated growth rates for health care prices and income are wrong (as they almost certainly are), it seems much more likely to us that affordability will decline more quickly than expected rather than more slowly, and it seems more likely that the number of uninsured people will increase more quickly than projected rather than more slowly.

A good deal of research and discussion has taken place on the effects of premium increases on employers’ decisions to offer coverage, on the amount of premium contribution required from employees, and on take-up rates for employees and for dependents.²⁰ Understanding the mechanisms through which premium increases translate into coverage decreases is useful for some purposes. However, the strong relationship between the affordability and coverage lines shown in Exhibit 2 and the logistic regression results that support these main findings make clear that the main message is simply this: It’s the premiums, stupid. Regardless of whether health care benefits are being paid out of the employer’s or the employee’s pocket, and without regard to the amount of premium contribution that employees are required to make, there is a remarkably tight relationship between affordability and coverage rates.

Substantial expansions or contractions of public coverage would certainly have an effect on health insurance coverage over the next decade; our projections are contingent on continuation of the status quo for Medicaid and SCHIP. We hope that public policy will change and that as a result, our projections of the number of uninsured Americans will be overstated; however, we do not have enough hubris to attempt to project the direction or magnitude of public policy change.

It is unlikely that we will be able to solve the problem of the uninsured without some form of universal health insurance coverage requiring contributions from some combination of employers, employees, and taxpayers. It is also unlikely that either our current system of employer-sponsored coverage or an alternative system of universal coverage will be sustainable without more effective efforts at cost containment.

This work was supported by a grant from the California HealthCare Foundation.

NOTES

1. T. Gilmer and R. Kronick, “Calm Before the Storm: Expected Increase in the Number of Uninsured Americans,” Health Affairs 20, no. 6 (2001): 207–210; and R. Kronick and T. Gilmer, “Explaining the Decline in Health Insurance Coverage, 1979–1995,” Health Affairs 18, no. 2 (1999): 30–47.
2. See J.A. Nyman, “The Value of Health Insurance: The Access Motive,” Journal of Health Economics 18, no. 2 (1999): 141–152; W.G. Manning and M.S. Marquis, “Health Insurance Tradeoffs Revisited,” Journal of Health Economics 20, no. 2 (2001): 289–293; and J.P Newhouse, The Economics of Medical Care: A Policy Perspective (Reading, Mass.: Addison-Wesley, 1978).
3. See J.P. Newhouse, “Medical Care Costs: How Much Welfare Loss?” Journal of Economic Perspectives 6, no. 3 (1992): 3–21; and D.M. Cutler, Your Money or Your Life: Strong Medicine for America’s Health Care System (New York: Oxford University Press, 2004).
4. This discussion does not address the question of why people don’t simply buy less insurance (that is, a higher deductible or greater coinsurance) instead of moving from being insured to being uninsured when faced with increased premiums. A full consideration of this question is beyond the scope of this paper, but we suggest one possibility. If, as suggested by Nyman, the main purposes of insurance for people with relatively low incomes are to provide better access to care and to protect health, buying a policy with a very large deductible (such as $5,000) or large coinsurance amounts (such as 40–50 percent) will provide very little improvement of access for people who do not have the cash available to pay the deductible or coinsurance amounts.
5. Centers for Medicare and Medicaid Services, “Health Accounts,” 7 March 2005, www.cms.hhs.gov/statistics/nhe (8 March 2005). For each year, we aggregated national health spending for services typically included in an acute care benefit package: hospital and physician services, prescription drugs, home health services, and net administrative costs of insurance.
6. We used data from the 1977 NMCES, the 1987 NMES, and the 1996 and 1999 MEPS to estimate the proportion of spending in each sector accounted for by adults ages 19–64, the proportion of out-of-pocket spending accounted for by this age group, and the ratio of spending on insured people to spending on uninsured people. Data from the 1999 MEPS show that the proportion of private health insurance spending accounted for by the elderly continued to decline (as it has done for the previous decade), while the proportion of out-of-pocket spending accounted for by the elderly increased slightly from its 1996 level. Including these updated data resulted in minor changes to the price series we presented previously.
7. As described in more detail in earlier work, we limited the analysis to workers not covered by a spouse or a public program, for two reasons. First, very few workers covered by a spouse or by a public program purchase coverage on their own or through their employer, so including them in the analysis might bias our estimates of the effect of price on purchase decisions. Second, the CPS questions on health insurance coverage changed substantially in 1988, in 1995, and again in 2001. The 1988 and 1995 changes primarily affected coverage estimates for children, Medicaid, and dependent coverage. See K. Swartz, “Changes in the 1995 Current Population Survey and Estimates of Health Insurance Coverage,” Inquiry 34, no. 1 (1997): 70–79. Limiting the analysis to adult workers not covered by a public program or as a dependent minimizes the effects of the question wording changes on the results. Adjustment for the 2001 question wording change is discussed below.
8. C. Nelson and R. Mills, “The March CPS Health Insurance Verification Question and Its Effect on Estimates of the Uninsured” (Washington: U.S. Bureau of the Census, 2001), shows that the addition of the health insurance verification question in 2001 resulted in an 8 percent decrease in the estimate of the percentage of the population that is uninsured. In the data presented in Exhibit 1, we adjusted our estimates of the percentage uninsured prior to 2000 by multiplying the “raw” CPS estimate by 0.92.
9. Our previous specification included covariates for African American and Latino with non-Latino whites and people of other races/ethnicities as the comparison group. The revised specification also includes a covariate for people reporting themselves as Asian.
10. The logistic regression and supplemental exhibits are available online in Methods Appendix 1, content.healthaffairs.org/cgi/content/full/hlthaff.w5.143/DC2.
11. S. Heffler et al., “Health Spending Projections through 2013,” Health Affairs, 11 February 2004, content.healthaffairs.org/cgi/content/abstract/hlthaff.w4.79 (8 March 2005). The national health spending projections contain data on projected spending by sector (hospital, physician) and by payor (private insurance, out-of-pocket, public). These projections are in the same format as the historical data we used to estimate per capita health spending for insured adults ages 19–64, and we followed identical methods to project per capita spending for this group in 2013. Using this procedure, the CMS data imply projected growth rates of 7.8 percent in 2003 and 7.2 percent in 2004. However, as reported by Jon Gabel and colleagues, premium increases for employer-sponsored insurance increased 13.9 percent in 2003 and 11.2 percent in 2004. J. Gabel et al., “Health Benefits in 2004: Four Years of Double-Digit Premium Increases Take Their Toll on Coverage,” Health Affairs 23, no. 5 (2004): 200–209, The employer premium data fluctuate more widely than the CMS spending data, and we used the historical relationship between the two sources to revise the CMS projections for 2003 and 2004 to incorporate the employer premium information reported by Gabel and colleagues. The CMS spending projections included a projection for per capita gross domestic product (GDP), and we used these projections and the historical relationship between mean per capita income and GDP per capita to estimate mean per capita income in 2013.
12. See online supplemental exhibits, as in Note 10.
13. K. Kronebusch, “Children’s Medicaid Enrollment: The Impacts of Mandates, Welfare Reform, and Policy Delinking,” Journal of Health Policy, Politics and Law 26, no. 6 (2001): 1223–1260; and T. Gilmer, R. Kronick, and T. Rice, “Children Welcome, Adults Need Not Apply: Changes in Public Program Enrollment across States and over Time,” Medical Care Research and Review 62, no. 1 (2005): 56–78.
14. A discussion of the methods used to estimate the relationship between the percentage uninsured among children and the percentage uninsured among adult workers is available online; see Note 10.
15. The 2003 CPS used a revised classification scheme to categorize employment. We mapped these categories to those available from earlier surveys as closely as possible. However, the mapping is not exact, and we used the 2002 rather than the 2003 CPS to estimate the extent to which industry of employment changed over the fifteen-year period.
16. We adjusted the CMS estimate up slightly from 7.0 percent using more recent data on the larger increases in employer premiums as documented in Gabel et al., “Health Benefits in 2004.” We used linear regression to estimate the relationship between premium increases in the CMS spending data and in the employer data provided by Gabel and colleagues from 1988 through 2002, and we used Gabel’s reported premium increases in 2003 (13.9 percent) and 2004 (11.2 percent) to predict national health spending increases in these years. Our predictions were above the CMS estimates: 10.4 percent versus 7.8 percent, respectively, for 2003 and 9.2 percent versus 7.2 percent for 2004. We replaced the CMS projections for these two years with our revised estimates and recalculated the percentage increase in premiums through 2013. We think that this estimate remains conservative and that it is likely there will be larger increases in 2005 and some following years.
17. J. Sheils, P. Hogan, and N. Manolov, “Paying More and Losing Ground: How Employer Cost-Shifting Is Eroding Health Coverage of Working Families,” International Journal of Health Services 29, no. 3 (1999): 485–518; and U.S. Government Accountability Office, Impact of Premium Increases on Number of Covered Individuals Is Uncertain, Pub. no. GAO/HEHS-98-203R (Washington: GAO, 11 June 1999).
18. C. DeNavas-Walt, B. Proctor, and R. Mills, Income, Poverty, and Health Insurance Coverage in the United States: 2003, Current Population Reports, Pub. no. P60-226 (Washington: Census Bureau, August 2004).
19. Institute of Medicine, Hidden Costs, Value Lost: Uninsurance in America (Washington: National Academies Press, 2003).
20. See, for example, H.S. Farber and H. Levy, “Recent Trends in Employer-Sponsored Health insurance Coverage: Are Bad Jobs Getting Worse?” Journal of Health Economics 19, no. 1 (2000): 93–119; M.S. Marquis and S.H. Long, “Employer Health Insurance and Local Labor Market Conditions,” International Journal of Health Care Finance and Economics 1, no. 3–4 (2001): 273–292; and P. Cooper and J. Vistnes, “Worker’s Decisions to Take-Up Offered Health Insurance Coverage: Assessing the Importance of Out-of-Pocket Premium Costs,” Medical Care 41, no. 7 Supp. (2003): SIII35–SIII43.

Todd Gilmer (tgilmer{at}ucsd.edu) is an assistant professor in the Department of Family and Preventive Medicine, School of Medicine, University of California, San Diego (UCSD) in La Jolla. Rick Kronick is a professor in that department and chief of the Division of Health Care Sciences, UCSD.

DOI: 10.1377/hlthaff.w5.143
©2005 Project HOPE–The People-to-People Health Foundation, Inc.