B. Douglas Bernheim Nina Buchmann Zach Freitas-Groff Sebasti´an Otero* October 30, 2020

1 Introduction As of this writing, more than 8.7 million Americans have contracted COVID 19, resulting in more than 225,000 deaths (Dong et al., 2020). The CDC has advised that large in-person events, particularly in settings where participants do not wear masks or practice social distancing, pose a substantial risk of further contagion (Centers for Disease Control and Prevention, 2020). There is reason to fear that such gatherings can serve as “superspreader events,” severely undermining efforts to control the pandemic (Dave et al., 2020).

The purpose of this study is to shed light on these issues by studying the impact of election rallies held by President Donald Trump’s campaign between June 20th and September 30th, 2020. Trump rallies have several distinguishing features that lend themselves to this inquiry.

First, they involved large numbers of attendees. Though data on attendance is poor, it appears that the number of attendees was generally in the thousands and sometimes in the tens of thousands. Because the available information about the incidence of COVID-19 is at the county level, the effects of smaller meetings would be more difficult to detect using our methods.

Second, the set of major Trump campaign events is easily identified. We know whether and when the Trump campaign held a rally in each county. This property allows us to distinguish between “treated” and “untreated” counties.

Third, the events occurred on identifiable days. They neither recurred within a given county nor stretched across several days. This feature allows us to evaluate the effects of individual gatherings.

Fourth, rallies were not geographically ubiquitous. As a result, we always have a rich set of untreated counties we can use as comparators.

Fifth, at least through September 2020, the degree of compliance with guidelines concerning the use of masks and social distancing was low (Sanchez, 2020), in part because the Trump campaign downplayed the risk of infection (Bella, 2020). This feature heightens the risk that a rally could become a “superspreader event.”
Despite these favorable characteristics, the task of evaluating the effects of Trump rallies on the spread of COVID-19 remains challenging for the reasons detailed in Section 3. Briefly, our approach involves a separate analysis for each of eighteen Trump rallies. We identify a set of counties that are comparable to the event county at the pertinent point in time, based at least in part on the trajectory of confirmed COVID-19 cases prior to the rally date. We then estimate the statistical relationship among those counties between subsequent COVID-19 cases and various conditions, such as pre-existing COVID-19 prevalence and pandemic-related restrictions, along with demographic characteristics. We use this relationship to predict the post-event incidence of new confirmed COVID-19 cases for the event county.