RISING COSTS DECREASE COVERAGE

The percentage of Americans without health insurance coverage rose more than 17 percent between 1990 and 1998. A study by Michael Chernew, David Cutler, and Patricia Seliger Keenan, funded by the Economic Research Initiative on the Uninsured (ERIU), investigates the factors influencing this increase. They find that over half of decline in coverage was attributable to rising premiums. This is the single largest factor in explaining the reduction in insurance coverage. Assuming medical costs continue their historical growth rate, the number of people uninsured could increase by 1.8 to 6 million in the next decade.

FINDINGS
Effects of Costs on Coverage, 1990 to 1998

• The decline in coverage rates was greater in areas with greater premium cost growth.
• Expansions in public coverage increased overall coverage, despite being associated with a decline in private coverage.
• Each $1,000/year of increased premiums leads to declines in coverage of 2.7 percentage points (p<0.05). The increase in health insurance premiums explains 1.6 percentage points of the 3.1 percentage point decline in coverage (52 percent).

POLICY IMPLICATIONS
Traditional economic theories posit that the ‘correct’ measure of insurance price is the ‘load’ (i.e. the difference between premiums and expected medical payouts). To the extent that variation in premium growth reflects variation in medical utilization and expense, as other literature would suggest, this work suggests that coverage also responds to the medical expense portion of the premium. If growth in health care costs continues to exceed growth in income, we will likely see further declines in health insurance coverage. This creates a challenge for policymakers interested in expanding coverage while also ensuring access to new but cost-increasing technologies. Subsidies to individuals or firms may increase coverage in the short run, but maintaining those coverage increases will likely require ever increasing subsidies.

METHODOLOGY
Premium growth is estimated using a hedonic premium model that includes plan traits and Metropolitan Statistical Area (MSA) dummy variables. The estimated dummy variables are used to create an index of premium growth for each MSA. Probit models are estimated using individual level data from the Current Population Survey (CPS) to assess the relationship between premium growth and coverage. Estimates are presented separately for total coverage, employer sponsored coverage, and public coverage. To account for potential measurement error and endogeneity of premium growth, instrumental variable (IV) models are also estimated using the change in state level spending for the non-elderly and the change in Medicare Part B spending in the MSA as instruments. The IV results suggest an even stronger effect of premiums on coverage.

All models control for demographic and income of the household and family head and a range of market level traits designed to capture explanations put forth in other literature examining coverage and coverage changes. The market level traits include: changes in tax rates, the percentage of working women, insurance market regulation, the share of the population that is foreign born, age 65 or older, or non-white, the share of the child health care expenditures eligible for Medicaid, average MSA income, and MSA unemployment rates.

CAVEATS
The analysis does not examine why premium growth may vary across markets. If premium growth is driven by a fall in coverage, perhaps because of cost shifting from uninsured to insured individuals or differential health status of the insured in markets with low rates of coverage, there will be a reverse causality bias in the base results. Moreover, the proxy premium measures may be subject to measurement error, which may result in an underestimate of the actual premium effects. The measurement error may be due to the relatively small sample size used to estimate premiums in some MSAs or to the inability to control for all relevant benefit traits in the premium estimation regressions. The IV models are designed to correct both the reverse causality and measurement error concerns; however the validity of the IV results depends on the quality of the instruments.

The functional form of the premium variable may not be correct. Specifically, the effect of premiums is measured in dollar changes, and the effect may be proportional (i.e., a $100 increase may have a different effect when premiums are $500 per month than when they are $700 per month.) Moreover, although the analysis controls for income, it does not measure the differential effects by income group. It may be the case that the effects are greater for lower income individuals. Certain other potentially important variables, e.g., the availability of charity care, are poorly measured. Finally, the analysis is based on aggregate coverage changes. There is no attempt to understand if the measured effects are a result of reductions in employer offers or employee take-up and none of the analyses examine the impact of premiums on the share of costs employees must pay.

DATA SOURCE
Current Population Survey (CPS), March 1989 – 2000. National sample of the non-elderly population. Premium data on 7,027 plans are taken from the 1988, 1989, and 1998 Kaiser Family Foundation/Health Research and Educational Trust Surveys of Employer-Sponsored Health Benefits. Medicare Part B expenditure data are from the Office of the Actuary at CMS.

CITATION
Rising Health Care Costs and the Decline in Insurance Coverage
Michael Chernew, University of Michigan; David Cutler and Patricia Seliger Keenan, Harvard University

Conference paper presented at ERIU Research Conference, July 2002.

The final version of the paper appeared as: Chernew, Michael, David M. Cutler and Patricia Seliger Keenan. 2005. "Increasing Health Insurance Costs and the Decline in Insurance Coverage." Health Services Research 40(4)1021-1039.

ERIU Working Paper #8

Back to top