Travel within the U.S. declined substantially during the early stages of the COVID-19 pandemic. How and why did people change their travel behavior? And to what extent did the change in travel patterns reduce exposure and slow the spread of COVID-19? These and related questions are addressed in Jeffrey Brinkman and Kyle Mangum’s paper, “The Geography of Travel Behavior in the Early Phase of the COVID-19 Pandemic.”
The authors studied changes in travel behavior in the spring using county-level location data, obtained from the movement of smartphones, across more than 2,000 U.S. counties.1 They observed the number of smartphones registered in a county each day and the fraction of those devices that were registered in each of the other counties within the prior 14 days. (The data were compiled over a two-week period to account for potential exposure based on widely reported COVID-19 incubation periods.) Using these data, the authors computed a mobility index that allowed them to observe travel patterns in and out of counties — and when going out, where people were visiting. They also performed statistical analyses to identify how people’s travel behavior was specifically affected by (i) existing COVID-19 case counts and (ii) government restrictions in place (that is, stay-at-home orders, closures of “nonessential” services and businesses, and limits on mass gatherings).
Brinkman and Mangum found that mobility overall fell by one-half from February (prior to the pandemic) to mid-April.2 Some counties faced declines greater than 60 percent, while others registered declines of less than 40 percent. They determined that travel dropped as a result of government restrictions but also because people decided on their own to avoid areas with high infection rates. The authors also found that individual decisions to avoid outbreaks, especially larger outbreaks, was the dominant reason for the decline in mobility. One example of this is that during the initial runup of infections in the first half of March, mobility declined by 20 percent (as cases rose by 500 percent), and this was before any stay-at-home order was enacted.
The authors show that people not only traveled less overall, but they avoided locations that had a high number of cases. These observed travel patterns imply that people considered available information on outbreaks when making their travel plans. The authors note that, “The data do not reveal precisely why people avoid places with high case counts. It could be a fear of being exposed to the virus in high caseload areas. It could be that reduced activity in the destination produces less of a reason to travel there … (beyond the government’s prescription).”
Using a nonlocal exposure metric for each county (computed as the sum of the travel flows between counties, weighted by the confirmed case counts in the counties visited), the authors show that people’s decisions to avoid travel to areas with high infection rates were instrumental in limiting viral spread back in their home counties.3 Using counterfactual analyses, the authors estimate that the median county was only one-half as exposed to the virus in the spring as it would have been if people had not reduced their mobility from prepandemic levels. One-third of this decline in exposure occurred because people changed their travel destinations as opposed to avoiding travel altogether.
The implied change in out-of-county exposure likely reduced total cases because nonlocal exposure led to local case growth. Brinkman and Mangum estimate that a 1 percent rise in a county’s nonlocal exposure index led to a 0.12 to 0.20 percent rise in COVID-19 cases.4 Counties with more exposure tended to have greater contact with other counties having high caseloads, although they did not necessarily have higher mobility levels. A case in point is Philadelphia and Pittsburgh, both of which were under the same state stay-at-home order. Yet in the spring, Philadelphia had a higher exposure rate (and hence more COVID-19 cases), which the authors attribute to Philadelphia's relatively high rate of contact with counties having high case counts along the Northeast corridor, including hard-hit New York City. As the authors remark, it follows then that “outside exposure can undo some of the suppression effects of stay-at-home orders.”
Using a model of spatial dynamics of the outbreak, the authors further show that greater overall connectiveness in terms of higher cross-county visitation rates increases the speed of the viral spread in the near term and also perpetuates the outbreak in the longer term. This means, for example, that if a city is strongly connected to another city in terms of travel flows, both cities are interdependently at risk of rising infections. If, in the extreme case, there is no travel, the virus eventually dies out. But, with mobility, the virus jumps from one county to another, keeping the virus alive. The authors write, “Such connectedness means the presence of the virus anywhere in the system is a threat everywhere else in the system.”
Brinkman and Mangum's valuable and timely analysis contributes to our understanding that the amount that people travel and the geographical location of that travel play an important role in determining the spread of COVID-19 infections. Furthermore, given that many people avoided travel voluntarily, the authors point out that the effectiveness of government shutdown orders on travel should not be overestimated. Their findings shed light on the importance of multiregional coordination and of government officials providing clear and timely local-level information to the public on outbreaks to curtail future infection rates.
1 The authors used measures constructed by Couture et al. using cell phone data, aggregated and anonymized, provided by PlaceIQ. (Victor Couture, Jonathan I. Dingel, Allison E. Greene, et al., “Measuring Movement and Social Contact with Smartphone Data: A Real-Time Application to COVID-19,” National Bureau of Economic Research Working Paper 27560 .)
2 By late May, mobility partially recovered to 20 percent below prepandemic levels.
3 Based on this exposure metric, a county’s exposure is lower (and, in turn, there are fewer cases) when people travel less from home and when they specifically avoid traveling to counties with high case counts, while exposure is higher when people travel more and fail to avoid areas of outbreak.
4 As the authors explain, “This is a direct effect estimated in the early stages and does not reflect the long-run dynamics of the pandemic.”