Health insurance and health of self-employed workers

Perry and Rosen (2001) take as their point of departure the differential tax treatment of employment-based health insurance for self-employed workers compared to wage earners: health insurance premiums received as a fringe benefit of employment are entirely deductible from the worker’s taxable income, while the self employed can deduct only some fraction (currently 60%) of premiums from their taxable income.

This difference, in addition presumably to other differences in the small versus large group markets for insurance, results in much higher rates of insurance coverage among wage earners (81.5% in 1996 for those under age 63) than among the self-employed (69%).

Perry and Rosen investigate whether this difference in insurance coverage leads to any detectable difference in health outcomes. In order for this to be a valid natural experiment, they must demonstrate that self-employment status itself is not affected directly by health status.

For example, if very healthy people who place a low value on health insurance precisely because they are healthy are disproportionately likely to become self-employed because they do not value the tax subsidy to health insurance for wage-earners, then the variation in health insurance is endogenous and the natural experiment is not valid.

Perry and Rosen therefore go to some lengths to document the fact that self-employment status, and transitions into and out of self-employment, do not seem to be driven by the health of either the self-employed individual or the health of his or her children.

Based on the strength of this evidence, it appears that the difference in rates of health insurance coverage for the self-employed compared to wage earners forms the basis for a valid natural experiment, so that it will be possible to draw causal inferences from this situation.

Specifically, any differences in health status between the self-employed and wage-earners can be causally attributed to differences in health insurance coverage.

However, Perry and Rosen fail to find any differences, on average, between the health of the self-employed and the health of wage earners.

Using data from the 1996 Medical Expenditure Panel Survey (n=8,986), they fail to find significant differences in self-reported health status or in the probability of any one of a number of conditions (including viral infections, headaches, cardiac conditions, upper respiratory infections, respiratory disease, skin disease, intestinal disorders, and arthritis).

The authors conclude that the public policy concern over low rates of insurance coverage among the self-employed may be misplaced; or at least that the concern should not be motivated by fear of adverse health outcomes.

Whether this conclusion is warranted depends to some extent on how much of an effect of insurance on health one considers important.

For example, the most broad measure of health Perry and Rosen consider (self-reported health status good or better (versus fair or poor) has a sample mean of 0.93, and the estimated marginal effect of selfinsurance is 0.0118 with a standard error of 0.00706).

Perry and Rosen do not report a 95% confidence interval for this effect, but if one approximates the 95% confidence interval as two standard errors, it includes effects of about 0.026.

Even the mean estimated effect of 1.18% might not be considered small given that only about 7% of the sample rates in health as fair or poor, but clearly the upper bound on the confidence interval is not insignificant given the sample means.

This tends to argue for a more conservative interpretation of the Perry and Rosen results – i.e. that they lack the statistical power to exclude a potentially meaningful effect of health insurance on health. An equally valid interpretation of this study is that for this population (employed adults), the health impact of lower rates of health insurance is not sufficiently large to show up in average differences in a sample of this size.

Importantly, however, this does not exclude a substantial effect of health insurance on health. It is difficult to summarize the results of the quasi-experimental studies since they rely on very different situations and look at very different populations: infants (both American and Canadian), children, the “medically indigent”, HIV patients, veterans, and 65-year olds.

But with the exception of Haas et al. and Perry and Rosen, these studies find evidence of significant improvements (declines) in health outcomes as result of expansions (contractions) of insurance coverage.

Location: Indonesia