It’s not even Election Day yet, but tens of millions of Americans have already gone through the process of waiting in line to cast a vote. Even for those with the options of mail-in and drop-off ballots, early voting lines have stretched for blocks in multiple states.
Theoretically, more early voting means fewer people will try to pack into the polls on Election Day proper — a good thing during a pandemic. But when early voting, itself, results in big crowds — and the voters just keep on coming — it’s worth wondering whether we’re really avoiding as much viral transmission as we’d hoped.
How big a COVID-19 risk is it to vote in person? Here at FiveThirtyEight, we’d normally turn to scientific research to answer a question like that. So we did that. And, well …
There are two studies that tried to quantify how voting in the Wisconsin primary in early April contributed to the spread of COVID-19. One of them concludes that it didn’t. The other says the opposite: the busier and more crowded the polling site, the more local COVID-19 cases occurred afterward.
It’s hugely consequential which study’s conclusions are accurate, given the millions of people voting in person this election season. And experts told me one study is probably getting closer to the truth than the other. But neither study really fully answers the question — a fact that’s also hugely consequential to understanding COVID-19’s risks. Studying a new virus in real time is frustratingly inexact, especially when our country has missed crucial opportunities to test and trace thoroughly.
If you can still think back across the vast eons of time to April of 2020, you’ll recall that the early weeks of the pandemic were a chaotic period for primary voting. In Wisconsin, it wasn’t clear until almost the day of the election whether voting would happen in person and when.
The tumultuous nature of that primary is what attracted researchers to studying it. “I saw in the paper that they were forcing people to show up to vote. I was abhorred by it and was expecting to find lots of infections,” said Lawrence Wein, professor of management science at the Stanford University School of Business. Because of his focus on health care, he wanted to know how mismanagement of an election might contribute to the spread of COVID-19.
Wein’s study calculated the state’s R-naught value — a measurement of how quickly the virus is spreading — in the weeks leading up to and after the April 7 primary election. In that time, the R-naught didn’t seem to change much. And his data showed hospitalizations in the state actually went down in the weeks after the vote, suggesting his calculations weren’t just a fluke caused by inadequate testing. It wasn’t what Wein had expected, but his data suggested the steps election officials had taken — wearing masks and gloves, making hand sanitizer available to voters and facilitating social distancing — helped to keep COVID-19 from spreading at the polls.
(FiveThirtyEight has published separate research by two sociologists that reached a similar finding.)
But wait a second. Because while Wein’s approach tells us something, it’s not the whole picture. There’s no counterfactual, as in no way to tease out what would have happened if the election hadn’t been held. Lacking that, you don’t know if a static R-naught means the election didn’t change anything, or if it means cases would have fallen otherwise and the election just kept them at the same level.
That’s where the other study comes in. Chad Cotti, a professor of economics at the University of Wisconsin-Oshkosh, saw that districts throughout the state weren’t approaching the primary in the same way. Election sites were being closed and consolidated, Cotti said, but it wasn’t uniform. The result was big differences from county to county in the density of voters per polling location. “I live in Appleton and I don’t think we had any” consolidation, he said. “But Green Bay had significant consolidation, and those counties are adjacent to each other.”
Cotti’s paper used those differences to create a sort-of counterfactual. We’ll never know what would have happened to the overall state COVID-19 count if the primary election never happened. But we can know whether counties that crammed more voters into fewer locations had a bigger effect than ones that didn’t.
And here there was a difference: A 10 percent increase in in-person voters at a given location turned out to be associated with an 18.4 percent increase in that county’s COVID-19 cases. More voters, more virus.
Cotti’s paper is probably providing some insight Wein’s lacks, but neither of these papers is perfect. Cotti’s study, for example, is still a preprint that’s being revised as it goes through the peer review process. One thing he told me the reviewers want him to address is the fact that the relationship between increases in voters-per-location and increases in local COVID cases is probably not a linear one. “Adding one person to a polling place doesn’t mean an incremental increase. There’s probably a relatively safe density and then it starts to ramp up quickly after a certain point. So twice as large a group may be more than twice as dangerous,” he said.
But more importantly, other experts told me, neither of these studies can really tell us how risky elections are compared to other behavior. To do that, you would have to randomly assign people to voting and non-voting groups, and see which one was more likely to contract COVID-19, said Tariq Cheema, a pulmonary and critical care physician for Allegheny Health Network who has published research on COVID-19 transmission. And that’s, you know, not really ethical.
In fact, said Jennifer Dowd, a professor of demography and population health at Oxford University in England, this is the kind of challenge that makes sociology-based research a real bear even in the best of situations. Say you want to know how a change in policy — like increasing the minimum wage — changes outcomes in a given city or state. You’ll run into the same kinds of problems, she said. How do you know that the two places you are comparing would have been on the same trajectory had one of them not raised the minimum wage? What if places more likely to raise the wage had something different about them to begin with? How do you separate cause and effect?
“It’s a very persistent problem in social science. And with COVID, it’s been really hard,” Dowd said. For example, we don’t know if the data from the Cotti paper tells us about the risks of voting — or if it tells us that people who vote are more likely to be socializing in groups, generally. The data is going to be messy, Dowd told me. And any research purporting to show the impact of any given event on transmission is going to be imperfect.
That problem gets compounded by the fact that the United States never implemented a consistent and widespread test and trace program. “If you really had a comprehensive program and you can look at clusters, you’d be able to get very descriptive data. You don’t know for sure but you could see the patterns more clearly,” Dowd said.
Instead, if we want to know how risky voting is, we’re basically left to make inferences from other research. For instance, we know group situations are riskier. We know breathing aerosols matters more for transmission than touching surfaces. We know from studies of cases in places with more intensive tracing programs that it’s significantly safer to be outdoors than in. We know mask use and distancing reduces transmission.
And the takeaway from that for voting? “My concerns are the lines,” Cheema said. Actual time voting is short and there aren’t a lot of other people around. But hours-long lines leave a lot of time for big groups of people to be near each other, to eat and drink and take their masks off. If those lines are inside, rather than out, the risks just go up. He expects to see cases of COVID-19 spiking around the country sometime about Nov. 15. He likened it to the risks of flying, in which being in a well-ventilated airplane with a mask on doesn’t seem to be a high-risk activity — but sitting around a crowded airport is.
Dowd had less concern about a voting-related spike but agreed that the waiting is the hardest part. It is also the part voters have the least control over. You can choose to wear a mask. You can choose to stand at a safe distance. You can go away and come back later when the line is shorter. But you can’t generally force your polling place to have people wait outside.
That’s why Cotti is happy the general election has had more options for how to vote than the primary elections offered. “More voting choices, a longer time when you can vote — that should reduce the number of people at the poll,” he said. “It should reduce the risk.”