There is more than a little hyperbole about the overhaul of Obamacare proposed by the House and the Senate, and the rhetoric about tens of thousands of deaths is not a bad example. The numbers are inflated, the studies are cherry-picked, the uncertainties are ignored, the context is dismissed, the framework is laughably distorted.
Slightly better than observational studies are natural experiments, where some external factor determines, in this case, which people get health insurance and which don’t. The experiments most commonly cited in defense of the argument that repealing Obamacare will substantially increase mortality are natural experiments: One study compared mortality in states that expanded Medicaid in the early 2000s to states that didn’t; another study compared mortality in Massachusetts counties before and after Romneycare to mortality in control counties in neighboring states. The former study gives rise to the figure of 43,000 deaths; the latter study to both the 24,000 and 36,000 figures (which differ because they use different estimates of the effect of Obamacare repeal on coverage). But of these two, it is the Massachusetts study that has generated the most media coverage. When people talk about the lives saved by Obamacare or the lives threatened by Paul Ryan and Mitch McConnell, they are usually referring to the Massachusetts study. Benjamin D. Sommers, who was involved in both studies, told Vox that the Massachusetts one was more relevant.
There is no doubt that the study is an important work. Benjamin D. Sommers, the lead author, is a professor of health policy at Harvard, and the methodology behind the study is rigorous. But the study, which found that one death was prevented each year for every 830 adults who gained health insurance under Romney’s health-care reform, is not limitlessly applicable to the case of Obamacare repeal. For starters, all of America isn’t Massachusetts, and the writers explicitly note that the result may not hold more broadly. Massachusetts has some of the highest-quality medical care in the United States, for instance, so obtaining insurance may have greater effects on health there than in, say, the rural South. Furthermore, as Megan McArdle notes, the study was conducted between 2007 and 2010 at the height of the recession, which Massachusetts weathered better than neighboring states. This may have distorted the health outcomes in the control counties in ways that the authors can’t account for.
Now, there are some reasons to prefer the results from Massachusetts over the results from Oregon: The Massachusetts study followed considerably more people and over a longer period of time (four years instead of two). And the Oregon sample had some drawbacks. The sample size was too small to confidently measure the impact of Medicaid coverage on mortality, so a decrease in mortality wouldn’t necessarily be inconsistent with the results of the study — although the study also found that outcomes from cardiovascular disease, stroke, and diabetes remained essentially unchanged between those who did and did not have insurance. But there are also reasons for preferring the Oregon study: It was completely randomized, so it isn’t susceptible to the problems that always exist in a natural experiment — whether, for instance, there was some other unaccounted factor that improved outcomes in Massachusetts relative to those in neighboring states during the period of the study. And there is other literature supporting the contention that health care has a relatively small effect on mortality. For instance, people become eligible for Medicare at age 65, so if insurance decreases mortality, we would expect a discontinuity in health outcomes at that age across the broader population. But a 2004 study* found no such discontinuity. Finally, it is plausible that both the Massachusetts and Oregon studies are correct, and that private insurance decreases mortality, but Medicaid does not.
But set aside the Oregon study and assume that Sommers is the only word on the subject. The figures of 24,000 or 36,000 extra deaths a year are still almost certainly too high. The 36,000 figure, in fact, is transparently too high because it assumes that 30 million people will lose health insurance under a repeal of Obamacare, which is higher than even the CBO predicts. So consider the 24,000 figure, which follows from applying the Massachusetts result of one life saved per year for every 830 adults who gain health insurance to the current figure of 20 million Americans who gained health insurance under Obamacare , whether through the exchanges or the Medicaid expansion.
Many of the people who ‘lose coverage’ under Obamacare repeal will be people who only had coverage because they were forced to.
The first problem with this analysis is that it is not necessarily the case that losing health insurance has as much effect on mortality as gaining health insurance, particularly if the population that loses health insurance is non-random. After all, the people who will choose to leave the private exchanges established by Obamacare after the individual mandate is repealed are likely to be relatively healthy, possibly healthier than the uninsured population was in Massachusetts before Romneycare. Even if they are not, it is possible that there are immediate benefits from gaining health insurance that are not comparable to the costs from losing health insurance — a possibility that Sommers himself mentioned to the Washington Post.
The second problem with this analysis is that it is unlikely that 20 million Americans will lose their health insurance. As Avik Roy chronicles, the CBO estimates (which currently predict 23 million more uninsured under the House plan) are far more pessimistic than is justified, since they assume that the individual mandate is improbably powerful and that all states would eventually expand Medicaid under Obamacare. The real impact of the House bill analyzed by the CBO, Roy estimates, might be closer to 5 million than 20 million. And that is the House bill, which almost certainly will look very different from the final bill passed.
The third problem with this analysis is that the entire framework is wrong. The predictions of the CBO are so extreme because they assume, as mentioned above, that the individual mandate is the only thing keeping a large number of Americans in the exchanges. But this means that many of the people who “lose coverage” under Obamacare repeal will be people who only had coverage because they were forced to. They’re not really losing coverage; rather, they’re making a decision to save the money and go without it — a decision that wasn’t available for them under Obamacare. These people are assuming the risk of higher mortality without health care because it makes more sense to them. It makes no sense to think of Obamacare repeal as “killing” these people since they aren’t being forced out of the exchanges.
In short, the only problem with the estimate that Obamacare repeal will kill tens of thousands is that it cherry-picks one study out of several, ignores the limitations of that study, assumes that private insurance and Medicaid are equivalent, assumes that losing health insurance and gaining health insurance are precisely symmetric, uses implausible estimates of coverage loss, and relies on an idiosyncratic definition of the word “kill.” Otherwise, it’s fine.
— Max Bloom is an editorial intern at National Review and a student of mathematics and English literature at the University of Chicago.
*Editor’s Note: This piece originally mistakenly linked to a 2008 study by David Card, Carlos Dobkin, and Nicole Maestas, when the author intended to refer to a 2004 study by the same authors. It has been corrected.