Research Update · February 16, 2021

Fear, Lockdown, and Diversion: Comparing Drivers of Pandemic Economic Decline


In Goolsbee and Syverson (2021), we used data from January through May 2020 to measure the economic impact of lockdowns during the early part of the COVID pandemic. Using updated data from SafeGraph on consumer visits to more than 2.2 million businesses in 110 different industries, we have revisited these findings to further test for lockdowns’ effects since the end of our earlier sample. We report the results of this exercise in this research update.

The results, which use data from the three-month period from May 18 to August 16, 2020 (when many states ended their lockdowns) and also from the three-month period from October 5, 2020 to January 3, 2021 (when a number of states reimposed sheltering orders or advisories), almost exactly replicate the original findings from the spring. Sheltering orders had only a modest impact on consumer behavior. The magnitudes are virtually identical in each period. The updated estimates suggest the economic gains from exiting lockdowns in the summer and the economic cost of reimposing lockdowns in the fall were each around 8% (compared to around 7% in the original findings).

Moreover, the severity of the COVID outbreak locally continues to be a significant driver of consumer activity.


We have the updated SafeGraph information on consumer visits to the same businesses and industries as in our previous sample.

There are two periods of policy change that we explore. The first is a three-month period following the end of the previous sample. This spans the weeks from May 18, 2020 to August 16, 2020. This period saw many states repeal their Shelter-in-Place (SIP) orders. The second is the final three months available in the sample. This spans the weeks starting October 5, 2020 through January 3, 2021. During this period, the virus resurged to its highest levels in the U.S., causing several states to reissue sheltering orders or advisories.

The policy data are not as detailed as in the previous paper because we do not have nationally comprehensive information on county level SIPs as in Goolsbee et al. (2020).

Instead, we construct samples of businesses in all counties that are part of commuting zones where there is a policy border with variation in sheltering policy based on the state-level policy information in New York Times (2020) and Raifman et al. (2020). This included 303 counties in the summer time period.[1]

For the reimposition of sheltering orders, we use the state level information in New York Times (2020) and Spiegel and Tookes (2020) to find the commuting zones where there is policy variation. We classified a policy as a reimposition if the state issued a “Stay at Home” order or advisory or imposed multiple restrictive measures simultaneously. These states included California, Pennsylvania, Delaware, Nevada, New Mexico, Oregon and Colorado.[2] This yielded a sample of 221 counties.


In Table 1, we repeat the standard specification of Goolsbee and Syverson (2021) using weekly data on the log number of visits per day to a store regressed on store-level dummies, commuting zone-by-week dummies (to control for aggregate conditions and the overall level of fear), a variable equal to one if the county the store is located in has a sheltering order operating that week, the log of the cumulative number of COVID deaths in the county (using a hyperbolic sine transformation to account for zeros), and the same measure but for deaths only in the previous two weeks (in case the more recent figures have an outsized impact).

Column 1 looks at the summer sample (May 18 – August 16), when states exited their lockdowns. The coefficient indicates that there was a positive, statistically significant rebound of the economy in counties that repealed their orders compared to other counties in the same commuting zone that did not. However, the magnitude (about 7.7%) is modest and almost identical to the coefficient estimated in the spring (which itself included some exits from sheltering orders).

Column 2 looks at the fall/winter sample (October 5 – January 3), when several states reimposed sheltering orders or advisories. They were not exactly like the SIP orders from the spring, being less strict in language, but their estimated coefficient shows they had an economic impact of an almost identical magnitude (8.9%) as the SIPs did in the spring (7.7%).

In both cases, local deaths in the county had a significant negative effect on consumer activity. In the summer, this was driven by more recent deaths. In the later period, the cumulative death toll mattered more.


In sum, the additional data from these two other time periods confirm that, just as in the early response to the COVID pandemic, the economic impact of policy orders is modest. Fear of the disease remains a significant driver of consumer behavior.


View Working Paper    View Finding     View Related Data





Goolsbee, Austan and Chad Syverson, 2021. “Fear, lockdown, and diversion: Comparing drivers of pandemic economic decline 2020” Journal of Public Economics Volume 193, January 2021, 104311
Goolsbee, Austan, Luo, Nicole Bei, Nesbitt, Roxanne, Syverson, Chad, 2020. “COVID-19 Lockdown Policies at the State and Local Level.” BFI Working Paper 2020-116. August.
Keystone Strategy, 2020. “Covid-19 Intervention Data.” Available at: Strategy/covid19-intervention-data, accessed February 6, 2020
Raifman, Julia, Kristen Nocka, David Jones, Jacob Bor, Sarah Lipson, Jonathan Jay, and P. Chan, 2020. “COVID-19 US state policy database.” Available at:, accessed January 20, 2020
Spiegel, Matthew, and Heather Tookes. “Business restrictions and Covid-19 fatalities.” Covid Economics (2020): 20. Database available at:, assessed January 20, 2020
The New York Times, 2020. “See Coronavirus Restrictions and Mask Mandates for All 50 States.” Available at:, accessed January 20, 2020
[1] We cross-checked the policy dates for each of these counties with the information in Keystone Strategy (2020) and Spiegel and Tookes (2020), and we did manual searches of the local news in the county where the dates differed. There were some counties (primarily in California) that allowed businesses to re-open before their SIP orders expired. We tried including a separate policy variable in the regressions to allow for a different response in those places, but it did not make any difference to the results.
[2] We matched to county-level information from local news searches for states where the reimpositions could differ regionally, as in California and Colorado. We did not include Wisconsin in the list of reimposition states. Its original SIP order was struck down by the state supreme court, so we could not judge whether the reimposition orders were binding there. Including Wisconsin made the estimated impact of policy even smaller.