Did the Affordable Care Act Medicaid expansion reduce temperature-related mortality?
Maybe not, but more research is probably necessary.
by: Christopher Callahan
June 14th, 2025
Excessively cold and hot temperatures are an enormous hazard to human health, responsible for a substantial fraction of global mortality by some estimates. Given both the strong influence of non-optimal temperature on mortality at present and the projected increase in heat-related mortality due to climate change, decision-makers might be interested in policy interventions that have the potential to break the link between temperature and mortality.
One such intervention may be the provision of health insurance. Having health insurance has been shown to reduce the risk of mortality in many studies, as being able to afford health care likely allows people to see doctors more regularly, take advantage of preventative care, and so on. In the United States, the most recent expansion of health insurance resulted from the 2010 Patient Protection and Affordable Care Act (“ACA” or “Obamacare”), which—among other things—allowed states to opt in to an expansion of the Medicaid health insurance program for low-income individuals. Specifically, states could choose to expand Medicaid coverage to individuals with incomes up to 138% of the federal poverty line, subsidized by federal funding.
Studies of both the ACA Medicaid expansion and previous expansions have found that increasing Medicaid coverage increases health care usage and reduces mortality rates. Because the expansion only occurred in states that opted in, we have a nice quasi-experimental setup, where mortality rates can be compared between states that expanded Medicaid (the treatment group) and states that did not (the control group).
These previous findings and quasi-experimental setting make the Medicaid expansion a natural fit for an analysis that asks how expanding health insurance could affect the temperature-mortality relationship. We took a first cut at this by linking county-level data on monthly all-cause, age-standardized mortality rates across the US with data on temperature and state-level expansion of Medicaid. We regressed log mortality rates on a fourth-order polynomial of daily temperature (following Carleton et al., 2022) and a second-order polynomial of daily precipitation. We interacted the temperature polynomial with an indicator of Medicaid expansion in each county, allowing us to test for a difference in the shape of the temperature-mortality relationship before and after expansion. We control for county-specific seasonality and county-specific annual trends in mortality, isolating plausibly exogenous variation in temperature. (For more technical details, see the end of this post.)
We found no statistically significant difference in the temperature-mortality relationship between the expansion states and non-expansion states. For example, in the plot below, you can see the temperature-mortality relationship without Medicaid expansion (blue) and with expansion (red). We see the typical J-shaped relationship in both cases, and the two curves are nearly identical.
Importantly, of course, Medicaid is focused on low-income people. We don’t have income-specific mortality in our data, but we could limit the analysis to the lowest-income areas to see if the effect shows up there. Here is the result if the analysis is limited to just the top ⅓ of counties in terms of the percent of people living in poverty.
Again, we do not find an effect of the Medicaid expansion on the temperature-mortality relationship, even in the places where we would be most likely to see it. If anything, the with-expansion curve is a little steeper.
This is an extremely simple first cut at the analysis, but it is worth doing because it is the most obvious version of the analysis with a standard dataset and a commonly used regression setup. Taken at face value, then, these results suggest that the expansion of health insurance to greater numbers of low-income people did not alter the effect of temperature on population-wide all-cause mortality.
Why didn’t we find an effect? A few non-exhaustive, non-mutually-exclusive explanations:
Medicaid does not affect mortality. This does not appear consistent with the data. For example, running a simple regression of log mortality rates on the Medicaid expansion in the same quasi-experimental setup suggests that expansion reduced county-level mortality risk by 0.87% (P < 0.0001). This is consistent with more sophisticated other work.
We are “washing out” the treated population in our aggregated data. Medicaid only covers low-income people; after the expansion, this means eligibility starts below 138% of the federal poverty line. Our data on county-level mortality includes everyone regardless of income, meaning that it groups the “actually” treated population with a bunch of people not getting Medicaid even after the expansion. Maybe Medicaid helps, but we just can’t see it because we’re not limiting our analysis to the actually-treated group. The fact that we don’t see an effect even in the lowest-income counties pushes back against this possibility, but it remains the case that more fine-grained data might reveal an effect where more aggregated data can’t see it. This would be consistent with other work showing that Medicaid has disproportionate benefits for the lowest-income populations (as intended).
Some sort of “reverse displacement” effect. If Medicaid allows people to live longer or avoid other causes of death, it might make them “available” to be killed by extreme temperatures. (The converse of the “displacement” effect where extreme temperatures reduce mortality later by killing people who would have died soon anyway.) Credit to my wife Ciara for this suggestion.
The parallel trends assumption is not supported. It’s possible that the temperature-mortality relationship was evolving differently before 2014 in the treated vs. control counties, meaning that simply comparing the relationship before and after the expansion does not statistically identify the effect of the expansion.
If I had to pick an an explanation here, it would be #2. Given the income-specificity of Medicaid, I’m not sure that our data are equipped to find an effect even if it exists in the real world. That’s why I think this is still very much an active area of research, and we cannot yet conclude that expanding Medicaid has no role to play in reducing the effect of environmental hazards on health.
Our lab is extremely interested in understanding the effect of policy interventions on the relationship between environmental stress and health outcomes. If you have useful data, other ideas, or a unique analytical setup, please reach out! We would love to collaborate.
Technical appendix:
Our sample comprised 2141 counties across 36 states. The sample was limited to either states that never expanded Medicaid (968 counties across 10 states) or states that expanded it in 2014 (1173 counties across 26 states). For folks who are up on the recent differences-in-differences literature, this choice helps us avoid potentially weird issues with staggered treatment timing. The data was limited to 2008 to 2019 to create a balanced set of 6 years before and after the 2014 expansion. The final number of observations was 304,744.
Along with the Medicaid expansion interaction, the temperature polynomial was also interacted with county mean temperature, because we know that heterogeneity in average temperature can strongly affect the temperature-mortality relations. When we plot the response functions with and without Medicaid expansion, they are evaluated at the population-weighted average of mean temperature (14.7 °C).
Regressions were weighted by county population and standard errors were clustered at the state-year level.