It may seem obvious that health insurance helps health, but very few cause-and-effect conclusions are obvious to economists. For example, suppose that we just compared the health of everyone who has health insurance and everyone who doesn’t. It would be unsurprising to find that those with health insurance are healthier, but the two groups will also differ in many other ways. For example, given that many Americans get health insurance through their employer, the chances are good that those with health insurance are more likely to be employed and on average to have higher incomes. How can we disentangle effect of health insurance from other possible confounding factors?
Or imagine that you compared the health of people before-and-after they had health insurance. This approach has some promise, but again, if getting health insurance is also connected to getting a job with benefits, a higher income, and perhaps a more settled life in other ways, then the task of separating out the effect of health insurance from other confounding factors remains.
Or one can imagine a social experiment in which a large group is randomly divided, with part of the group receiving health insurance and part not. Then you could track the two randomlly selected groups over time, and see what happens. This is essentially the approach used to test the safety and efficacy of new drugs, for example. Thus, social scientists are on the lookout for situations where this kind of random selection in to health insurance happened, but perhaps by accident rather than policy.
In their essay, “The Impact of Health Insurance on Mortality,” Helen Levy and Thomas C. Buchmueller focus in some of these situations in which access to health insurance was determined in a way with a high degree of randomness (Annual Review of Public Health, April 2025).
One of the most clear-cut examples happened in Oregon in 2008. The state wanted to expand eligibility for Medicaid, but didn’t have the money to expand it for everyone. The result, as the authors describe it was”the 2008 Oregon Health Insurance Experiment, which studied ∼75,000 low-income adults under age 65, 40% of whom were selected by lottery to be eligible for Medicaid (the treatment group) with the remaining 60% serving as a control group.” Thus, some randomly received health insurance, and some did not.
Another truly randomized study looked at an “IRS initiative that sent letters in early 2017 with information about HealthCare.gov to a randomly selected sample of 3.9 million households that had been subject to the ACA [Affordable Care Act] individual mandate penalty for failing to have coverage in the previous year. The study finds that the letters led to a small but significant increase in coverage.” In this case, some randomly received a letter that increased the share of that group with health insurance, while others did not.
Yet another approach looked at those admitted to California hospitals who were either just under age 65, and thus not eligible for Medicare, or just over age 65, and thus covered by Medicare. The idea here is that the just-unders and just-overs should be highly comparable groups: after all, the only way they differ was in being born a few months apart. In this “discontinuity” approach (in this example, the discontinuity is age 65), the greater or lesser share of health insurance across groups is quite similar to random.
Other examples involve Medicaid coverage Medicaid is a joint federal-state program, so the program was often introduce in a staggered way, over time, across states. This was true back in the 1960s, when Medicaid first enacted, and it was also true in the 2010s, when states were allowed to expand Medicaid coverage, but over several years, only some did so. A researcher can look at this data and see if, when a group of people become eligible for Medicaid, the pattern of their health outcomes then shifts from previous patterns–and the patterns of health outcomes for groups that did not become eligible at that time. Here, the random ingredient is the staggered time periods in which health insurance was introduced.
My theme here is that there are plausible ways for researchers to study a cause-and-effect relationship between health insurance and health. Of course, not all of these studies cover the same age groups, or find the same outcomes. But my guess is that a number of readers care less about the way the studies are done, and more about how the authors of this review would summarize the overall results. Here, I quote from the abstract of their paper:
A 2008 review in the Annual Review of Public Health considered the question of whether health insurance improves health. The answer was a cautious yes because few studies provided convincing causal evidence. We revisit this question by focusing on a single outcome: mortality. Because of multiple high-quality studies published since 2008, which exploit new sources of quasi-experimental variation as well as new empirical approaches to evaluating older data, our answer is more definitive. Studies using different data sources and research designs provide credible evidence that health insurance coverage reduces mortality. The effects, which tend to be strongest for adults in middle age or older and for children, are generally evident shortly after coverage gains and grow over time. The evidence now unequivocally supports the conclusion that health insurance improves health.
