We’re less than three weeks away from Election Day, and although we’ve already seen an enormous uptick in the number of absentee ballots requested — more than a third of Americans plan to vote by mail because of the coronavirus — many people will still cast their votes in person. This means that people will assemble at polling places, as they did in the primaries held earlier this year and as they’re doing in states that have started early voting. This, of course, raises a difficult question: Could the simple act of voting worsen a public health disaster?
We have spent seven months now living under the threat of the novel coronavirus. And unfortunately, given the increasingly politicized way in which COVID-19 has been handled, voting during the pandemic has itself become a partisan issue. Distrust in science is high, even though now is when such trust is most needed, and we find ourselves in a situation where the discussion around potential effects of the election on the pandemic is transpiring in the absence of a comprehensive assessment of the risks.
Earlier this year, we undertook an analysis of in-person voting in the primaries to better understand the risks of voting in the general election.1 Of course, a comparison between primaries and the general election is imperfect, but our analysis does offer us a window into the impact voting could have on the spread of COVID-19. And what we found should offer some comfort to those who plan to vote in person: We saw no increase in overall COVID-19 mortality due to in-person voting in the primaries.
Modeling the spread of COVID-19 is very challenging. First, the underlying data is of uneven quality. The U.S. has struggled to test enough people since the very beginning of the outbreak; particularly in the early days, there were simply not enough tests to track the number of cases. That’s why our analysis used mortality data collected by The New York Times, instead of the number of reported positive test results, since mortality data remains the most reliable measure for assessing the status of the epidemic.
Second, to better capture the dynamics of the epidemic, we used two different techniques from statistical and epidemiological modeling. (The current preprint of our paper, along with data and code, is available here.)
We should note that our analysis is what statisticians call “ecological.” We examined whether primary elections affected the course of the pandemic in a county, not whether, at the individual level, going to the polls affects an individual’s risk of contracting the infection.
Our first approach relies on a technique called “matching.” In an ideal world (at least from a statistician’s point of view), we would be able to explore a counterfactual — “What if this state had not held its primary during a pandemic?” — and then compare the number of COVID-19 deaths under each scenario.
Of course, we only have access to the world as it actually unfolds. So, we approximate this alternate world by using “matches,” defined for our purposes as counties that did not hold a primary but otherwise have similar characteristics relevant to the incidence of COVID-19 (e.g., population density, the percentage of the population over 65, median income, the share of the vote President Trump won in the 2016 election) and that, most importantly, share a similar cumulative COVID-19 death rate in the 20-day period surrounding the primary.2 Accordingly, matched counties are similar to one other when it comes to the course of the COVID-19 epidemic and risk factors — but one county held its primary and the other has not.
Let’s look at Chippewa County, Wisconsin as an example. Wisconsin’s chaotic April 7 primary was the first major election with in-person voting since states began issuing stay-at-home orders. We matched Chippewa with five3 of the most demographically similar counties that had similar cumulative COVID-19 death rates but had not held their primary, including Sumner County, Kansas; Madison County, Kentucky; and Rensselaer County, New York. These counties each had cumulative death rates similar to Chippewa in the period around the Wisconsin primary, and scored relatively closely on our demographic measures.
Once we had this set of counties, we compared the observed mortality rate after the primary in Chippewa to the average of its five matches over the same (post-primary) period. Performing this procedure across all of the counties that held primaries in our data, we found that there was no overall average increase in the mortality rate in counties that held primary elections compared to their matches.4 In our second approach to the data, we wanted to replicate our results using a method specifically tailored to study the transmission of a virus but that relied on a wholly different set of assumptions.5 In our analysis, we tested the effect of two key dates: the date a state’s governor issued a stay-at-home order and the date of the state’s primary election (if applicable, in either case). We then combined information from across the U.S. into a single model that estimated the overall effect of holding a primary election.
While statewide lockdowns appeared to have reduced the transmission of COVID-19, the primary elections had no effect. Below, you can see the rate at which the virus spreads in Florida and Illinois, two populous states that held primaries in the early phase of the virus, throughout the spring. This is only two of the 34 states we looked at, but the effects on COVID-19 transmission in Florida and Illinois are representative of the overall average that we calculated across all of the states included in our analysis.