The Marketplace Pulse series provides expert insights on timely policy topics related to the health insurance marketplaces. The series, authored by RWJF Senior Policy Adviser Katherine Hempstead, analyzes changes in the individual market; shifting carrier trends; nationwide insurance data; and more to help states, researchers, and policymakers better understand the pulse of the marketplace.
The economic dislocation caused by COVID-19 has many manifestations. Millions have filed for unemployment, and food banks are overwhelmed by an unprecedented level of demand. Housing insecurity is expected to mount as eviction moratoria expire. In the context of this massive income loss and financial hardship and given the health impact of COVID on individuals and families, understanding the potential impact on insurance coverage is critical.
The relationship between unemployment and uninsurance is complex. It depends not only on the extent of job loss, but also on the coverage status of those losing jobs, as well as transitions into alternative sources of coverage such as a family members’ employer plans, Medicaid, or the individual market. A number of estimates have been published in the last few months, but they differ substantially in approach and findings, as has been recently explained. Due to their many differences, these studies are not strictly comparable, but a simplified version of their main conclusions is as follows:
- A recent microsimulation study by the Urban Institute estimates a net increase in uninsurance of 2.9 million over the last three quarters of 2020.
- An earlier econometric model from the Urban Institute estimates an increase in uninsurance of between 5.1 and 8.5 million assuming an unemployment rate of 15 percent.
- An estimate by the Kaiser Family Foundation finds that 27 million would lose employer coverage and be at risk of uninsurance, and suggests that 80 percent would be eligible for alternative coverage, although not necessarily enrolled.
- A report recently released by Families USA suggests that the number of uninsured workers had already increased by 5.4 million between February and May.
Of these four estimates, only the first results from a microsimulation model. While many estimates or studies might refer to some underlying model that generates their results, a microsimulation model is a specific type of analytic tool built from large representative samples of individual and household data. Microsimulation models are designed to produce detailed distributional analyses, which is one of their main distinguishing features from other approaches. These models are powerful analytic tools for simulating proposed policies and analyzing impacts across all segments of the population, but they are costly to build and maintain. Microsimulation models are typically built from multiple data sets and must be updated regularly. This step is time-consuming but critical, since the underlying model must capture the current baseline against which proposed policy impacts are measured.
Given the complexity, it is not surprising that there are currently only three microsimulation models of U.S. health insurance coverage that are regularly used to analyze policies and have publicly-released information about the methodology and data sources. The Congressional Budget Office (CBO) uses a microsimulation approach to estimate the impact of proposed policy changes on health insurance coverage. The Urban Institute has the Health Insurance Policy Simulation Model (HIPSM), and the RAND Corporation has a microsimulation model, COMPARE. The research organizations and governmental agencies that maintain these models make their estimates public, and are also highly informative about their methods, assumptions, and data sources. While there are other models being used in the private sector—for example, Milliman and Health Management Associates both have models—much less is known about them, since they are proprietary and mostly used to create estimates for clients. To be credible, microsimulation models must be explicit about their data and methods, and also their track-record in predicting outcomes.
The Urban Institute's HIPSM model uses data on six million individuals and parameters about coverage behavior based in research and recent evidence to create a functioning synthetic version of the health insurance system. By incorporating data about the characteristics of those losing jobs, the model is able to create a more accurate estimate of the potential impact on coverage by modelling the likelihood of various transitions, leveraging information about the coverage status of individuals within families, eligibility for alternative sources of coverage, and the likelihood of enrollment. It is also designed to incorporate data from the most recent administrative data on Medicaid, marketplace enrollment, and premiums, in addition to employment changes released monthly by the Bureau of Labor Statistics. These attributes allow the HIPSM model to make a more considered estimate of the impact of the current recession on coverage, an estimate that is able to incorporate all of the complexities in our health insurance coverage system, as well as the particularities of the current recession.
Given current data, the Urban model’s projected impact of job loss on uninsurance is a net increase of close to three million—an increase in the size of the uninsured population of slightly more than ten percent. Of the ten million people predicted to lose employer coverage, they estimate that about two thirds are able to transition to a family member's employer coverage or become eligible for and enroll in Medicaid or a Marketplace plan. But it is notable that of 48 million people estimated to live in a household with someone that loses a job, only about 20 percent will lose their pre-pandemic employer coverage as a result. This speaks to the unequal impact of the current economic dislocation. This recession, which is causing unprecedented levels of food and housing insecurity, may affect health insurance coverage a little less precisely because it disproportionately affects those who have the least to begin with, and are therefore less likely to have had employer-based coverage to lose.
Needless to say this story is far from over. The extent of economic dislocation is yet unknown. Further job loss may lead to more reductions in employer sponsored insurance, and depending upon how the recession evolves, those job losses could spread to a wider range of industries and occupations than those most affected so far. Such changes could lead to higher growth in both uninsurance and Medicaid enrollment than estimated thus far. The most accurate assessment of the coverage impacts of the crisis will have to await the release of the National Health Interview Survey (NHIS), but several recent surveys and reported managed care organization (MCO) enrollment data are consistent with the general contours of the HIPSM estimates. While a microsimulation model is arguably the best tool for this particular job, by no means is it infallible. As the saying goes, “All models are wrong. Some are useful.”