The 2020 Sturgis Motorcycle Rally and COVID-19
TThe 2020 Sturgis Motorcycle Rally and COVID-19 ∗ Yong Cai † Grant Goehring ‡ September 5, 2020
Abstract
The Sturgis Motorcycle Rally that took place from August 7-16 was one of the largest publicgatherings since the start of the COVID-19 outbreak. Over 460,000 visitors from across theUnited States travelled to Sturgis, South Dakota to attend the ten day event. Using anonymouscell phone tracking data we identify the home counties of visitors to the rally and examine theimpact of the rally on the spread of COVID-19. Our baseline estimate suggests a one standarddeviation increase in Sturgis attendance increased COVID-19 case growth by 1.1pp in the weeksafter the rally. ∗ We would like to thank SafeGraph for providing data. † Northwestern University. Email: [email protected] ‡ Boston University. Email: [email protected] a r X i v : . [ q - b i o . P E ] S e p Introduction
In response to the COVID-19 pandemic leaders have adopted a variety of policies to stop the virusfrom spreading. Restricting large gatherings has been one such policy recommended by the Cen-ter for Disease Control (CDC) and adopted by many state and local governments to help reducetransmission. The annual Sturgis Motorcycle Rally that took place in Sturgis, South Dakota fromAugust 7-16 was one of the largest public gatherings in the United States since the start of thepandemic. Over 460,000 people from across the United States travelled to Sturgis to participate inthe ten day event. This paper studies the effect of the Sturgis rally on subsequent COVID-19 casegrowth in the home counties of rally attendees.This study uses anonymous cell phone tracking data from Safegraph to identify individuals thatwere in Sturgis during the rally period. The data also provide information on the individual’s homecensus block group allowing us to identify areas that had relatively more rally attendees. Combiningthis information with county-level COVID-19 case data from the
New York Times we find thatcounties with relatively more rally attendees have higher COVID-19 case growth rates in the weeksfollowing the rally.We contribute to the rapidly growing body of work on the efficacy of social distancing andrelated policies. Whereas many papers have found that social distancing measures were successfulin containing the spread of COVID-19 (e.g. Andersen, 2020; Courtmanche, Garuccio, Le, Pinkstonand Yelowitz, 2020), studies of large-scale public gatherings have yielded seemingly contradictoryresults. For instance, Dave et. al (2020) study whether the Black Lives Matter protests in theaftermath of George Floyd’s death on May 25, 2020 increased COVID-19 cases. By comparing citieswhere protests did or did not occur as well as variation in the start dates they find that protests didnot increase COVID-19 case growth rate. They find evidence that a greater share of people stayedhome due to the resulting curfews and general unrest, possibly offsetting the effect of the protests.A synthetic control study on President Trump’s political rally in Tulsa, Oklahoma on June 20, 2020also found no effect on COVID-19 case growth (Dave, Friedson, Matsuzawa, McNichols, Redpath,Sabia, 2020). The authors suggest that offsetting behavior, as well as smaller-than-expected crowdattendance might be important in limiting the effects of the rally. CDC guidance on in-person gatherings can be found here: https://dot.sd.gov/transportation/highways/traffic The Sturgis Motorcycle Rally is an annual motorcycle rally held in Sturgis, South Dakota. Since2011 the South Dakota Department of Transportation (DOT) has collected data on the total numberof vehicles entering Sturgis during the rally. In 2020 DOT tracked over 460,000 vehicles entering therally compared to a ten year average of 547,882. Rally attendance outperformed many early mediaexpectations predicting the rally to be half the normal size due to the pandemic. Rally attendeesovershadow the native Sturgis population of 7,000, and are an important source of business for localestablishments. In June, the Sturgis City Council voted 8-1 to allow the rally to proceed, and theGovernor of South Dakota, Kristi Noem, enthusiastically urged that the rally take place as usual. While motorcycle riding is a common interest among attendees, the rally in many respects functionsas a large multi-day party. Concerts are held and attendees frequent local bars and drink in campsites set up for the rally.We use cell phone tracking data from SafeGraph to identify rally attendees and their homelocation. For a nationally representative sample of cell phones the data track the census blockgroup containing the device as well as its “home” census block group. SafeGraph determines thehome location of the cell phone by identifying the census block group where the cell phone is mostfrequently located at night over a six week period. Figure 1 shows the distribution of Sturgis See for instance On August 7, 2020 Governor Noem tweeted: “ SafeGraph has made the data available to COVID-19 researchers here: https://docs.safegraph.com/docs/social-distancing-metrics Figure 1: Proportion of Devices at the Sturgis Rally (August 7-16, 2020)
We study effect of Sturgis on the spread of COVID-19, via event study and difference-in-differences.
The event study is motivated by identification concerns. In particular, Sturgis exposure might becorrelated with individual attitudes towards COVID-19, as well as the level of local governmentresponse (see Painter and Qiu, 2020; Allcott, Boxell, Conway, Gentzkow, Thaler and Yang, 2020;among others). This would lead to positive bias in our estimates. Conversely, individuals from The county-level data from the New York Times is available here: https://github.com/nytimes/covid-19-data The data from the COVID Tracking Project is available here: https://covidtracking.com/data Y c,t = / (cid:88) τ = 06 / β τ · Sturgis c · { t = τ } + X (cid:48) c,t γ + ε c,t . where Y c,t is the growth rate of COVID-19, defined as the log difference in cumulative COVID-19cases, in county c at week t . Given the varying norms around case reporting on the weekends, weaggregate our data to the week level. The period under consideration begins on the first week ofJune (week ending on June 7) and ends on the last week of August (week ending on August 30). TheSturgis Rally ran from August 7-16. We take the week ending August 16 as the period of treatmentand exclude it from our analysis. We drop observations from the county containing Sturgis (MeadeCounty) as well as adjacent counties since travel patterns between these counties are likely not dueto rally attendance. While we do not report the specifications, results are robust to different sampleselections. Sturgis c is one of two measures of county exposure to the Sturgis Rally. Our preferred measure is SturgisP rop – the proportion of the county that visited Sturgis. We also consider
SturgisT opHalf – an indicator for being in the top 50% of
SturgisP rop . X c,t is a set of conditioning variables whichincludes lagged median percentage time spent at home, growth in state COVID-19 testing, as wellas county and week fixed effects.We plot the β τ ’s as well as their 95% confidence intervals for the two types of exposure measurein the figure below: Before August 17, the coefficients are generally negative and not significantlydifferent from 0. This suggests that there is little positive selection into Sturgis attendance. AfterAugust 17, the coefficients turn positive and significant, and appear to be increasing over time.According to the estimates, a 1pp increase in SturgisP rop leads to a 7pp increase in the growthrate of COVID-19 cases in the week ending August 23. This effect increases to 11pp in the weekending August 30. Put differently, a one standard deviation increase in
SturgisP rop increasesCOVID-19 case growth by 1.02pp in the week immediately following the event and 1.48pp in the The week ending August 9 is considered to be a pre-treatment period since visitors would not have returned totheir respective communities at this point in time. We exclude: Perkins, Butte, Meade, Ziebach, Lawrence, and Pennington Counties. Results are also qualitativelysimilar when all counties in South Dakota or only Meade and Pennington Counties, which contain Sturgis and RapidCity, are removed. (a)
SturgisP rop : Proportion of county that visitedSturgis (b)
SturgisT opHalf : Indicator for being in the top50% of
SturgisP rop week after. In the same vein, a county in the top half of Sturgis exposure experiences growth inCOVID-19 cases that is higher by 1.02pp relative to a county in the bottom half in the week endingAugust 23, rising to 1.48pp in the week ending August 30. The effect is more precisely estimatedfor
SturgisP rop than
SturgisT opHalf , possibly because the former is a more accurate measure ofSturgis exposure.
Next, we consider the difference-in-differences (DiD): Y c,t = β · Sturgis c · { t > / } + X (cid:48) c,t γ + ε c,t .Y c,t and Sturgis c are defined as before. We consider various specifications, as explained below.Results can be found in table 1.In specifications (1) and (2), X c,t includes lagged median percentage time spent at home, growthin state COVID-19 testing, as well as county and week fixed effects. These specifications are thereforecomparable with the event studies above. Column 1 of table 1 shows that a 1pp increase in Sturgisattendance increased growth in COVID-19 cases by 11pp on average in the weeks following the Rally.Equivalently, a one standard deviation increase in Sturgis attendance increased growth in COVID-19cases by 1.1pp on average. Column 2 shows that a county in the top 50% of Sturgis attendance has6able 1: Results for Difference-in-Differences(1) (2) (3) (4) (5) (6)Coef. 11.014 0.016 12.546 14.654 9.591 2.083S.E. (2.083) (0.007) (1.479) (1.857) (2.401) (57.829)p-value 0.000 0.030 0.000 0.000 0.000 0.971Treatment Prop TopHalf Prop Prop Prop PropCounty Time Trend No No Yes Yes Yes YesExclude College No No No Yes No NoBalanced Panel No No No No Yes No N R About 7% of counties are not present in the week of Jun 7, 2020. To explore selection into datareporting we repeat specification (3) with a balanced panel. This yields specification (5), whichleads to similar results as before.Finally, the last specification explores whether individuals observe their neighbors attendingSturgis, and change their behavior in response. Specification (6) tests this hypothesis by usingmedian percentage time spent at home as the outcome variable. Clearly, individuals are not adjustingtheir behavior. This could account for the relatively larger effect of the Sturgis Rally relative toother public events (e.g. in Dave, D. M., Friedson, A. I., Matsuzawa et al. 2020). College location data can be found here: https://hifld-geoplatform.opendata.arcgis.com/datasets/colleges-and-universities Conclusion
We examine the effect of the Sturgis Rally on COVID-19 case growth in the United States. We findcounties with relatively more rally attendees experienced higher COVID-19 case growth in subse-quent weeks. Other studies that have found large public gatherings do not affect case growth point tooffsetting behavior that reduced possible COVID-19 transmission following the event. We find thatstay-at-home behavior did not change in response to Sturgis exposure suggesting individuals werelikely unaware of Sturgis attendees in their communities and therefore did not take take precautionsto reduce their risk of exposure following the rally.