JUST AS COVID-19 has upended daily life, so has it changed civic rituals. Historically, Americans have mostly voted in person. But in 2020 many states made postal voting easier, to reduce the risk of the virus spreading. The share of ballots cast by mail duly soared to 46%, from 21% in 2016.
Nonetheless, some 85m people still voted in person. Did this contribute to America’s surge of covid-19 cases late last year?
Data from earlier in 2020 are inconclusive. In Wisconsin 450,000 people voted in person in a primary election in April. Two studies later that month did not detect any unusual increase in covid-19 cases; a third, released in May, found a large effect.
The general election in November offered richer data. So far, 20 states have published the number of ballots in each county cast by each method. Overall, places where a high share of votes were cast in person on election day—distinct from both postal ballots and votes submitted in person before the election—also had high covid-19 rates. However, this pattern could arise for reasons besides polling queues.
To help rule out alternative explanations, we studied changes in the incidence of covid-19 within states over time. First, we compared each county’s case rate with its state’s average. Many factors can make the disease more or less common throughout a state, such as super-spreader events or mask mandates. Examining the gap between a county’s numbers and those of its state strips out the impact of such events.
Next, we tracked how these disparities changed between the pre- and post-election periods, a method known as “difference in differences”. Suppose that people who would not have been infected otherwise did catch the virus at polling places. If so, then covid-19 cases in the counties with the most in-person voting in a state should either have risen unusually quickly or declined unusually slowly after the election.
The data display just such a pattern. From mid-October to early November, covid-19 cases in counties with their states’ highest in-person turnout fluctuated similarly to those in areas with the lowest in-person voting rates. But a week after the election, positive tests became more common in places with the most in-person turnout on election day. The gap was biggest after 20-25 days, shortly after official data would include people infected by people who caught the virus while voting.
This divergence does not prove that polling sites were at fault. Places with lots of in-person voting on election day tended to share other attributes as well, such as having relatively low levels of income and education and having voted in 2016 for Donald Trump, a sceptic of masks and social distancing. Such characteristics could also have caused the striking “difference in differences” in the incidence of covid-19.
To isolate the impact of in-person voting, we built a model to predict each county’s post-election change in covid-19 rates, relative to state averages. We tested 22 variables, such as population density and the pre-election growth rate of covid-19 cases.
Many of these factors did affect the spread of the virus. Yet after accounting for all of them, in-person voting still had a statistically significant effect. Holding other variables constant, the gap in in-person voting on election day between the state with the highest rate in our data (Alabama, at 41% of the population) and the lowest (Arizona, at 6%) was associated with an extra 173 cases per 100,000 people. This implies that if no one had voted in person on election day, 220,000 fewer people would have been diagnosed with covid-19.■
Sources: State electoral data; Townhall; Health Resources & Services Administration; US Census Bureau; Survey of the Performance of American Elections; New York Times
This article appeared in the Graphic detail section of the print edition under the headline “Stamped out”