COVID-19
By Charles Courtemanche, Joseph Garuccio, Anh Le, Joshua Pinkston, and Aaron Yelowitz
doi: 10.1377/
Strong Social Distancing hlthaff.2020.00608HEALTH AFFAIRS 39,NO. 7 (2020): ?
?2020 Project HOPE?
Measures In The United The People-to-People HealthFoundation, Inc.
States Reduced The COVID-19
Growth Rate
Charles Courtemanche
ABSTRACT State and local governments imposed social distancing (courtemanche@uky.edu) is an
measures in March and April of 2020 to contain the spread of novel associate professor of
economics at the University
coronavirus disease 2019 (COVID-19). These included large event bans, of Kentucky, in Lexington,
school closures, closures of entertainment venues, gyms, bars, and Kentucky.
restaurant dining areas, and shelter-in-place orders (SIPOs). We evaluated Joseph Garuccio is a PhD
the impact of these measures on the growth rate of confirmed COVID-19 student in economics at
Georgia State University, in
cases across US counties between March 1, 2020 and April 27, 2020. An Atlanta, Georgia.
event-study design allowed each policy?s impact on COVID-19 case growth
Anh Le is a PhD student in
to evolve over time. Adoption of government-imposed social distancing economics at the University
measures reduced the daily growth rate by 5.4 percentage points after of Kentucky.
1?5 days, 6.8 after 6?10 days, 8.2 after 11?15 days, and 9.1 after 16?20 Joshua Pinkston is an
days. Holding the amount of voluntary social distancing constant, associate professor of
economics at the University
these results imply 10 times greater spread by April 27 without SIPOs of Louisville, in Louisville,
(10 million cases) and more than 35 times greater spread without any of Kentucky.
the four measures (35 million). Our paper illustrates the potential danger Aaron Yelowitz is a professor
of exponential spread in the absence of interventions, providing relevant of economics at the
University of Kentucky.
information to strategies for restarting economic activity. [Editor?s Note:
This Fast Track Ahead Of Print article is the accepted version of the
peer-reviewed manuscript. The final edited version will appear in an
upcoming issue of Health Affairs.]
critical questionduring theCOVID- cial distancing orders appears to be related
A 19 pandemic is the effectiveness to local income, partisanship, and political be-of the social distancing policies liefs in the US; and compliance with self-quaran-adopted by US states and localities tines is related to potential losses in income inin bending the curve. Although Israel.2?4
these policies take a variety of forms?such as Some epidemiological models forecast the
imposing shelter-in-place orders (SIPOs); re- eventual number of COVID-19 cases and fatali-
stricting dine-in at restaurants; closing other ties based on untested assumptions about the
non-essential business such as bars, entertain- impact of social distancing policies in contem-
ment venues, and gyms; banning large social porary society. The widely cited Imperial College
gatherings; and closing public schools?their ef- model assumes contact outside the home, school
fectiveness depends critically on the cooperation or workplace declines by 75 percent, school con-
of the public. For example, although California?s tact rates are unchanged, workplace contact
first-in-the-nation SIPO carries threats of fines rates fall by 25 percent, and household contact
and incarceration, its effectiveness fundamental- rates rise by 25 percent.5 Another study assumes
ly relies on social pressure.1 Compliance with so- social distancing measures will reduce the aver-
July 2020 39:7 Health Affairs 1
Downloaded from HealthAffairs.org on June 01, 2020.
Copyright Project HOPE?The People-to-People Health Foundation, Inc.
For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org.
COVID-19
age contact rate by 38percent, basedon evidence ords report New York City as a whole rather than
from the 1918 influenza pandemic.6 dividing it into five counties, reducing this num-
In the US the literature on models of social ber to 3,138. Our dataset tracked counties over
distancing during the COVID-19 pandemic is 58 days from March 1, 2020 to April 27, 2020,
evolving rapidly, and at the time of our writing, leading to a sample size of 182,004. We chose
we were aware of several working papers that March 1 as the start date because no new cases
examined the consequences of social distancing were reported in the entire U.S. on most days in
policies. Recent work found significant effects January and February. The April 27 end date was
of stronger measures (like SIPOs) on movement chosen to coincide with the first removal of one
using difference-in-differences methods and of four types of restrictions we analyzed (the re-
state-level data from Google.7 Similar findings opening of restaurants and other entertainment
have been obtained in a study with SafeGraph facilities in Georgia).15 Each county observation
mobility data,8 although a different study using was weighted by population using 2018 esti-
PlaceIQ and SafeGraph data found strong mates from the United States Department of
measures were not important.9 Another paper Agriculture?s Economic Research Service.16
used synthetic control methods to show that Outcome Of Interest We examined the daily
California?s SIPO significantly reduced COVID- growth rate in confirmed COVID-19 cases at the
19 cases.1 A study of SIPOs across the U.S. also county level, which originated from the 2019
found a reduction in cases, aswell as higher rates Novel Coronavirus COVID-19 Data Repository
of staying home full-time.10 Other authors used provided by the Johns Hopkins Center for Sys-
interrupted time-series methods and found that tems Science and Engineering. This repository
early statewide social distancing measures were collected data on COVID-19 cases worldwide
associated with decreases in states? COVID-19 from a range of sources including government
growth rates, but later SIPOs did not lead to and independent health institutions.17
further reductions.11 The daily exponential growth rate was calcu-
At issue is not whether isolationworks to limit lated as the natural log of cumulative daily
the spread of disease, but rather whether the COVID-19 cases minus the log of cumulative dai-
particular government restrictions designed to ly COVID-19 cases on the prior day.We chose this
encourage social distancing in the US reduced functional formbecause epidemiologicalmodels
spread relative to simply providing information predict exponential growth in the absence of
and recommendations. Individuals may volun- intervention. Percent growth in cases is identical
tarily engage in avoidance behavior, such as to percent growth in cases per capita since re-
hand washing or wearing masks, once they fully ported county populations did not vary during
perceive the risks of contagion.12,13 Critics of the sample period. The growth rate was multi-
more stringent government measures highlight plied by 100 and can be read as percentage point
Sweden?s less intrusive response to COVID-19, changes. In computing the growth rate, we fol-
although Sweden?s strategy is increasingly ques- lowed a recent COVID-19 study and added one to
tioned.14 Rigorous empirical research is needed the case counts to avoid dropping counties that
to determine the impacts of the various aspects started with zero cases.18
of state and local governments? responses in Covariates The data on the timing of state
the U.S. and local government social distancing interven-
Our work?which leveraged both state and tions was gathered from a host of sources and
county policy variation and used a flexible event- made available by Johns Hopkins University.19
study method that allowed for effects to vary Part A of the online appendix explains a few
across measures and over time?estimated the corrections we made to the dates and provides
impacts of four types of social distancing mea- a list of state- and county-level policies used in
sures on confirmed COVID-19 case growth rates the analysis.20
through April 27, 2020. The reduced-form ap- We focused on four government-imposed in-
proach captures any potential pathways driven terventions: SIPOs, public school closures, bans
by these mandates, including complementary on large social gatherings, and closures of enter-
avoidance behaviors that the public may engage tainment-related businesses. For large gather-
in if these orders provide an informational shock ings, we used the date of the first prohibition
in addition to increasing social distancing. that was at least as restrictive as 500 people.
Most of the bans were much more restrictive:
95percent of the time (inourpopulation-weight-
Study Data And Methods ed sample) the prohibition extended to 50 peo-
Study Data The unit of observationwas dailyUS ple. For entertainment-related businesses, we
county/county equivalents. Although there are used the date of the first closure of either restau-
3,142 counties in the US, official COVID-19 rec- rant dining areas (including bars) or gyms/
2 Health Affairs July 2020 39:7
Downloaded from HealthAffairs.org on June 01, 2020.
Copyright Project HOPE?The People-to-People Health Foundation, Inc.
For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org.
entertainment centers. 96 percent of the time, if Each policy was implemented at least 10 days
one such prohibition was in place, the other was after the start of the sample period and at least
in place as well. 20 days prior to the end. Therefore, each policy
We included control variables related to the contributes to the identifying variation for all
availability of COVID-19 tests. The same Data coefficients except those for more than 20 days
Repository that provides cases also includes ago and 10 or more days from now. Since the
daily counts of positive, negative, and pending estimated policy effects at those two ?catch-all?
tests in each state on each day, which we added time periods could partially reflect composi-
together.17 To mirror our measure of cases, we tional changes, they should therefore be inter-
converted this testing variable to the exponential preted with more caution than the estimates for
daily growth rate of cumulative tests performed. the other time intervals.
Since COVID-19 test results are generally not In addition to the testing controls discussed
available immediately, we also included the above, the model also included fixed effects for
one-day lag of this growth rate. Further lags geography and time. County fixed effects ac-
(out to 10 days) were considered but always sta- counted for the likelihood that, even aside from
tistically insignificant, so we did not include differences in policies, case growth rates may
them. Most states did not report any pending have varied due to a number of county character-
tests, meaning that they did not officially record istics. These characteristics include population
tests until the results were obtained. This likely density and residents? education, political orien-
explains the lack of a longer lag between testing tation, and age.3,4 Fixed effects for each day in
growth and case growth. each of thenineU.S. CensusDivisions (522 fixed
Methods We estimated the relationship be- effects in total) allowed for flexible underlying
tween social distancing policies and the expo- trends ingrowth rates that could vary indifferent
nential growth rate of confirmedCOVID-19 cases parts of the country, helping to account for the
using an event-study regression with multiple staggered nature of the outbreak across loca-
treatments. Statistical analysis was conducted tions.23 We report 95% confidence intervals,
using Stata MP (version 15). This approach is with standard errors robust to heteroskedastic-
akin to difference-in-differences but more flexi- ity and clustered by state, the level of most of the
ble, as it interacts the treatment variables with policy variation. Part B of the appendix provides
multiple indicators of time since implementa- the formal notation for the event-study model.20
tion, thereby tracing out the evolution of the Limitations There are several limitations to
treatment effects over time.21 our analysis. Official COVID-19 case counts are
For each of the four policies, we include seven known to understate the true prevalence of the
variables:whether itwas implemented 1?5, 6?10, disease, as they do not include asymptomatic
11?15, 16?20, or more than 20 days ago; and carriers, those who are not ill enough to seek
whether itwill be implemented5?9or 10 ormore medical care, and thosewho are unable to obtain
days later. Implementation on the current day a test due to supply constraints.1 Nonetheless,
through four days from now was, therefore, the confirmed case counts are crucial to the Trump
reference group. If a county never adopted the administration?s ?Opening Up America Again?
policy, each of these variables was set to 0 plan, which proposes either a ?downward trajec-
throughout the sample period. tory of documented caseswithin a 14-day period?
An event study model is particularly useful to or ?downward trajectory of positive tests as a
study the impact of social distancing policies percent of total tests within a 14-day period (flat
on COVID-19 cases for two reasons. First, after or increasing volumeof tests)? as criteria to loos-
accounting for the incubation period and time ening social distancingmeasures.24Moreover, to
betweenonsetof first symptomsandpositive test the extent that testing shortages led to only the
result, such policies likely only affect official sickest individuals receiving them, official case
cases after a considerable lag.22 Additionally, counts can loosely be interpreted as the preva-
the inclusion of variables reflecting future imple- lence of moderate-to-severe illnesses, a relevant
mentation allows for an analysis of pretreatment metric for policy purposes.
trends. Since it is not plausible for policies that A related caveat is that, ideally, we would like
have not yet been implemented to causally affect to be able to control more precisely for access to
current cases, finding such associations could testing. Available data only allowed us to control
suggest misspecification. For instance, one for number of tests performed at the state, rather
might expect counties with rapidly growing case than county, level. However, most of our policy
counts to be the most likely to enact these mea- variation is at the state level, so state-level testing
sures, leading to a reverse-causal relationship should go a long way towards alleviating bias.
between current cases and future policies that Additionally, number of tests performed is not
would be detected by our model. an ideal measure of the ease of obtaining a test
July 2020 39:7 Health Affairs 3
Downloaded from HealthAffairs.org on June 01, 2020.
Copyright Project HOPE?The People-to-People Health Foundation, Inc.
For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org.
COVID-19
because it also reflects the level of illness in the model.20 Relative to the reference category of
community. 0?4 days before implementation, SIPOs lead to
Also,wemight ideallywant to estimate a richer statistically significant (p < 0:01) reductions in
econometric model. It would be interesting to the COVID-19 case growth rate of 3.0 percentage
trace out the timing of impacts more exactly points after 6?10 days, 4.5 after 11?15 days, 5.9
and study the policies? interactions with each after 16?20 days, and 8.6 from day 21 onward.
other or county characteristics. Future work Because the model held constant the other types
should also examine the impacts of other social of policies, these estimates shouldbe interpreted
distancing policies such as closing public parks as the additional effect of SIPOs beyond shutting
and beaches, the requirement to wear masks in down schools, large gatherings, and entertain-
public, restrictions on visitors innursinghomes, ment-related businesses. This additional effect
state announcements of first cases or fatalities, may come from either the requirement/strong
and federal government actions such as prohib- advisement to shelter-in-place aside from ?es-
iting international travel.9However, it is difficult sential? activities or the accompanying closure
to include numerous correlated treatment vari- of any ?non-essential? businesses that remained
ables without reducing precision to the point open.We did not observe any statistically signifi-
where statistical inference is uninformative. cant ?placebo? effects of SIPOs in the periods
Finally, as is typical of observational data an- prior to implementation, giving credence to a
alyses, we cannot rule out all possible threats to causal interpretation of our main results. If any-
causal inference. Numerous possible confound- thing, the pre-trend appears to point upward,
ers could vary across time and space, including which would make our estimates in the post-
the other policies mentioned above, informal treatment period conservative.
encouragement by government officials to wear We found no evidence that bans on large social
masks or improve hygiene, changing business gatherings influenced the growth rate. The point
practices, and social norms regarding distanc- estimates for banning gatherings were statisti-
ing. That said, including Census-Division-by- cally insignificant (p > 0:56 in all cases). How-
day and county fixed effects in our model and ever, the 95% confidence intervals included re-
examining pretreatment trends helped us to ductions of up to 3?6 percentage points, so the
push in the direction of causality. lack of evidence of an effect should not be mis-
interpreted as clear evidence of no effect. Also,
the lack of a statistically significant reduction in
Study Results the post-treatment period could potentially be
Descriptive Information Confirmed COVID- due to an upward (though not statistically signif-
19 cases grew rapidly during the sample period, icant) pre-treatment trend. However, results
from just 30 on March 1 to 978,047 on April 27. from the aforementioned event study with sepa-
Part C of the appendix shows the number of rate variables for each day showed that the pre-
counties with any COVID-19 cases on each day.20 trend disappeared four days prior to implemen-
On March 1, the vast majority of counties had tation.
zero cases, and across all days, 49 percent of Supplemental exhibit 3 shows estimates for
unweighted county-by-day observations were the restaurant-and-entertainment-related busi-
zero. However, counties with zero cases tended nesses and school closings.20 Closing restaurant
to have low populations, so our population dining rooms/bars and/or entertainment cen-
weights limited the influence of these counties ters/gyms led to statistically significant reduc-
on the results. tions in the growth rate of COVID-19 cases in
Supplemental exhibit 1 illustrates the reach of all timeperiods after implementation (p < 0:05).
social distancing policies on the US population The estimated effect was 4.4 percentage points
over time.20 The SIPOwas generally the last poli- after 1?5 days, 4.7 after 6?10 days, 6.1 after 11?15
cy to be implemented, and adoption was uni- days, 5.6 after 16?20 days, and 5.2 after 21 or
formly lower than theotherpolicies.OnMarch 1, more. Prior to implementation, policies related
no jurisdiction had implemented all four mea- to businesses showed no effect on the growth
sures. By March 22, nearly 25 percent of the rate, again passing the ?placebo? test.
US population was covered by all the measures, In contrast, we found no evidence that school
growing to approximately 65 percent by closures influenced the growth rate. The point
March 29 and 95 percent by April 7, when the estimates were never close to statistically signifi-
last SIPO took effect. cant (p > 0:37 in all cases), but the 95% confi-
Impact Of Social Distancing Policies Sup- dence intervals meant that we could not rule out
plemental exhibit 2 illustrates the coefficients reductions of up to 4?5 percentage points.
(and confidence intervals) for SIPOs and bans Adding the coefficient estimates for each poli-
on large gatherings derived from the event-study cy gives the combined effect of implementing all
4 Health Affairs July 2020 39:7
Downloaded from HealthAffairs.org on June 01, 2020.
Copyright Project HOPE?The People-to-People Health Foundation, Inc.
For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org.
four social distancing policies. In days 1?5 after rate would have fallen to 11 percent. The actual
implementation, the bundle of restrictions re- growth rate, which reflects all implemented dis-
duced the growth rate of COVID-19 cases by tancingpolicies includingSIPOs, fell to 3percent
5.4 percentage points. In days 6?10 after imple- by that date.
mentation, the growth rate fell by 6.8 percentage Supplemental exhibit 5 compares the reported
points. This reduction grew to 6.8 percentage number of COVID-19 cases over time to the num-
points after 6?10 days, 8.2 percentage points ber of cases predicted by our event-study regres-
after 11?15, 9.1 after 16?20, and 12.0 after 21 sion under these same two counterfactual sce-
or more. As discussed previously, the estimate narios.20 Part E of the appendix describes the
for 21+ days should be viewed with caution, as it technical details of these simulations along with
did not utilize the same geographic balance of the required assumptions.20 The graph uses the
treatments as the estimates for the other time natural logarithm of nationwide cases (or pre-
intervals. A conservative interpretation of these dicted cases) for the y-axis scale, but with corre-
results would therefore be that the impact sponding numbers labeled on the y-axis instead
reached 9.1 percentage points after 16?20 days of logs.
andappeared to remainat least ashighafter that. In all three scenarios, cases increased roughly
Robustness Checks Part D of the appendix linearly on the log scale, as expected under
presents and discusses results from a number of exponential growth, until the last week of
robustness checks designed to address possible March?approximately two weeks after the first
concerns with our model.20 These checks begin restrictions and one week after the first SIPO.
with regressionswith just one variable per policy The actual curve then began to flatten substan-
to help rule out the null results for gathering tially, eventually leading to 978,047 cases by
bans and school closures being due to multicol- April 27. In contrast, the two counterfactual
linearity (appendix exhibit 4).20 We then evalu- curves only flattened slightly. By the end of the
ated robustness to using different functional sample period, the model predicts that cases
forms for the testing controls or omitting them would have been 10 times higher without SIPOs
(appendix exhibit 5).20 Next, appendix exhibit 6 (10,224,598) and 35 times higher (35,257,098)
varied the sample start date and the approach without any social distancing restrictions. Inter-
used to dealing with counties with no cases.20 estingly, the closures of restaurants/entertain-
Appendix exhibit 7 shows results from dropping ment facilities accounted a larger share of the
the uniquely affected states of NY,WA, and CA.20 reduction in cases than SIPOs, despite SIPOs
Appendix exhibit 8 displays results from a more having larger coefficient estimates. This is be-
fine-grained event-study model with separate cause restaurant/entertainment facilities were
variables for each day from 10 days before treat- implemented earlier and in more places than
ment to 20 days after.20 Finally, appendix exhib- SIPOs.
it 9 presents results from other checks related to
data and measurement issues as well as control-
ling for county-specific pre-treatment trends.20 Discussion
Thegeneral patternof resultswas robust to these Nuance is requiredwhen interpreting the results
different specifications. in supplemental exhibit 5.20 We view the simula-
Counterfactual Simulations Supplemental tion as providing an illustration of the power of
exhibit 420 uses the results from the baseline exponential growth and the effectiveness of so-
model to compare the observed growth rate of cial distancing restrictions at ?flattening the
COVID-19 cases to two counterfactuals: 1) none curve,? even when their impacts are not imme-
of the four social distancingmeasures ever being diately visible. As late as April 6, nearly a month
imposed and2)noSIPOeverbeing imposed.The after the earliest interventions, the number of
process for creating these counterfactuals is de- caseswould still havebeenunder 1,000,000even
scribed in the appendix.20 Themean exponential without any restrictions?just 2.4 times the ac-
growth rate without any interventions was tual number of cases. The explosion in cases
16.2 percent over the full time period. The ob- without social distancing measures happens lat-
served and both counterfactual growth rates er, and by time it is happening, the lagged effects
peaked on March 19, 2020 at 26?28 percent of these measures mean it is too late to stop it.
but started to diverge afterward, eight days after At the same time,weurge caution about taking
the earliest restriction. Without any social dis- the specific numbers of cases averted too literal-
tancing policies, the model predicts the case ly. Simulations that use estimated parameters to
growth rate would have stayed similarly high predict outside the range of observed policy var-
for another week before gradually falling to iation are inherently subject to a high level of
14 percent by April 27, 2020. Without SIPOs? uncertainty that is difficult to quantify. More-
but keeping the other restrictions?the growth over, had policymakers not taken action and
July 2020 39:7 Health Affairs 5
Downloaded from HealthAffairs.org on June 01, 2020.
Copyright Project HOPE?The People-to-People Health Foundation, Inc.
For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org.
COVID-19
COVID-19 had continued to spread throughout ply been replaced by informal gatherings. Alter-
April in themanner depicted by our simulations, natively, official prohibitions may have been
voluntary social distancing by individuals and largely redundant since the largest events (such
businesses would have likely increased as panic as college and professional sports) were already
over the risingdeath toll andhospital overcrowd- being cancelled due to CDC guidance or other
ing across the country mounted. In technical information.
terms, the Census-Division-by-day fixed effects Also note that school closures and large event
would have evolved differently than what we ob- bans occurred prior to the implementation of
served.Thiswouldhave likely offset at least some SIPOs, meaning substitute types of social gath-
of the additional predicted cases?though, be- erings were still allowed. Our results, therefore,
cause of the lag to impact, it is unclear howmuch should not be interpreted as a forecast about
of this offsetting could have occurred before the what would happen if schools were reopened
end of our sample period. or certain large gatherings were allowed while
Relatedly, testing shortages would likely have other aspects of SIPOs remained in place.
prevented official case counts from reaching the
numbers presented in our simulations. Howev-
er, this is largely a semantic distinction, as these Conclusion
infections would still be severe enough to war- We estimated the separate and combined impact
rant testing in the absence of a shortage. If any- of four widely adopted social distancing policies.
thing, not being confirmed as a COVID-19 case Both SIPOs and closures of restaurants/bars/
could lead to inadequate treatment. entertainment-related businesses substantially
As striking as our counterfactual estimates slowed the spread of COVID-19.We did not find
are, they still are not worst-case scenarios be- evidence that bans on large events and closures
cause they account for at least some voluntary of public schools also did, though the confidence
social distancing. Even without any government intervals cannot rule out moderately sized ef-
restrictions, supplemental exhibit 420 illustrated fects. Interestingly, two recent papers on the
a 14.3 percentage point drop from the peak effect of social distancing restrictions onmobili-
growth rate to the end of the sample period. ty found the same pattern as we did in terms of
The most plausible explanation is responses which restrictions mattered and which ones did
of individuals and businesses to information not, suggesting that null effects of gathering
about the severity of the pandemic and federal bans and school closures on case growth are at
guidelines. least plausible.7,8
While our results suggest both SIPO and Our contribution was to provide credible em-
non-SIPO measures can be effective at averting pirical evidence on whether US social distancing
COVID-19 cases, the lack of evidence of effects of measures worked as intended in flattening the
school closures or bans on large social gather- curve. Estimating other important benefits and
ings is noteworthy.We cannot rule out the pos- costs from social distancing, including the total
sibility that these null results are due to statisti- lives saved and economic harm, was beyond the
cal imprecision, but it is also possible that both scope of our study. Other work has attempted to
policies may displace social interaction rather estimate job losses, simulate effects on the over-
than reducing it. For example, school closures all economy and economic growth, or estimate
may have led families to continue social inter- distributional consequences from current and
actions outside of the school setting, such as at past pandemics.1,6,28?31
day care centers or parks. Google mobility data Nonetheless, we provide important informa-
through April 5, 2020 show increases of 10 per- tion about benefits of social distancing for
cent or more in visits to parks in 28 states.25 A policymakers to consider as they decide on strat-
new study finds that schools are only slightly egies for restarting economic activity. For in-
more dangerous than parks and playgrounds stance, our results argue against returning to
for COVID-19 transmission, supporting this ex- partial measures such as school closures and re-
planation.26 Alternatively, school closures pri- strictions on large gatherings, while removing
marily affect children and the vast majority of the restrictions that prevent the redirection of
children experience mild symptoms and there- social activity to other settings. At issue moving
fore may not be included in confirmed cases.27 forward iswhethercases averted simply turn into
While asymptomatic children can pass the virus cases delayed, and a premature return to light
to adults who become more severely ill, our re- measures would make this more likely. At the
sults imply that the extent to which this led to same time, our results are not informative about
confirmed cases did not change when schools the effectiveness of intermediatemeasures, such
were closed. as lifting a SIPO but requiringmasks in public or
Similarly, official group events may have sim- opening restaurants at reduced capacity. Further
6 Health Affairs July 2020 39:7
Downloaded from HealthAffairs.org on June 01, 2020.
Copyright Project HOPE?The People-to-People Health Foundation, Inc.
For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org.
research is needed as gradual, untested steps
toward reopening are takenacross the country.?
Charles Courtemanche has received States Department of Agriculture, and
funding in the past year from the the American Beverage Institute.
National Institutes of Health, United [Published online May 14, 2020.]
NOTES
1 Friedson AI, McNichols D, Sabia JJ, Apr 8 [cited 2020 May 11]. Available No. 26917). Available from: https://
Dave D. Did California?s Shelter in from: https://papers.ssrn.com/sol3/ www.nber.org/papers/w26917
Place OrderWork? Early Evidence on papers.cfm?abstract_id=3571421 14 Ahlander J, O?Connor P. Sweden?s
Coronavirus-Related Health Benefits 8 Andersen M. Early Evidence on So- liberal pandemic strategy questioned
[Internet]. Cambridge (MA): Na- cial Distancing in Response to as Stockhom death toll mounts.
tional Bureau of Economic Research; COVID-19 in the United States [In- Reuters [serial on the Internet].
2020 Apr [cited 2020 May 11]. ternet]. Greensboro (NC): University 2020 Apr 3 [cited 2020 May 7].
(NBER Working Paper No. 26992). of North Carolina Greensboro; 2020 Available from: https://www.reuters
Available from: https://www.nber Apr 6 [cited 2020 May 11]. Available .com/article/us-health-coronavirus-
.org/papers/w26992 from: https://papers.ssrn.com/sol3/ sweden/swedens-liberal-pandemic-
2 Bodas M, Peleg K. Self-Isolation papers.cfm?abstract_id=3569368 strategy-questioned-as-stockholm-
Compliance In The COVID-19 Era 9 Gupta S, Nguyen TD, Rojas FL, death-toll-mounts-idUSKBN21L23R
Influenced By Compensation: Find- Raman S, Lee B, Bento A, et al. 15 The State of Georgia. Providing
ings From A Recent Survey In Israel. Tracking Public and Private Re- guidance for reviving a healthy
Health Aff (Millwood). 2020 Apr 9. sponses to the COVID-19 Epidemic: Georgia in response to COVID-19
DOI: 10.1377/hlthaff.2020.00382. Evidence from State and Local Gov- [Internet]. Atlanta (GA): Governor?s
[Epub ahead of print]. ernment Actions [Internet]. Office; 2020 Apr 23 [cited 2020 May
3 Painter M, Qiu T. Political Beliefs Cambridge (MA): National Bureau of 11]. (Executive Order). Available for
affect Compliance with COVID-19 Economic Research; 2020 Apr [cited download from: https://gov.georgia
Social Distancing Orders [Internet]. 2020 May 11]. (NBER Working .gov/executive-action/executive-
St. Louis (MO): Saint Louis Univer- Paper No. 27027). Available from: orders/2020-executive-orders
sity; 2020 Apr 30 [cited 2020 May https://www.nber.org/papers/ 16 US Department of Agriculture, Eco-
11]. Available from: https://papers w27027 nomic Research Service. Population
.ssrn.com/sol3/papers.cfm? 10 Dave DM, Friedson AI, Matsuzawa estimates for the U.S., States, and
abstract_id=3569098 K, Sabia JJ.When Do Shelter-in-Place counties, 2010?18 [Internet].
4 Wright AL, Sonin K, Driscoll J, Orders Fight COVID-19 Best? Policy Washington (DC): USDA; [updated
Wilson J. Poverty and Economic Heterogeneity Across States and 2019; cited 2020 May 11]. Available
Dislocation Reduce Compliance with Adoption Time [Internet]. from: https://data.ers.usda.gov/
COVID-19 Shelter-in-Place Protocols Cambridge (MA): National Bureau of reports.aspx?ID=17827
[Internet]. Chicago (IL): University Economic Research; 2020 May [cit- 17 Johns Hopkins University and
of Chicago, Becker Friedman Insti- ed 2020 May 11]. (NBER Working Medicine, Coronavirus Resource
tute for Economics; 2020 Apr 29 Paper No. 27091). Available from: Center. COVID-19 Dashboard by the
[cited 2020May 11]. (Working Paper https://www.nber.org/papers/ Center for Systems Science and En-
No. 2020-40). Available from: w27091 gineering (CSSE) at Johns Hopkins
https://papers.ssrn.com/sol3/ 11 Siedner MJ, Harling G, Reynolds Z, University (JHU)[Internet]. Balti-
papers.cfm?abstract_id=3573637 Gilbert RF, Venkataramani A, Tsai more (MD): Johns Hopkins Univer-
5 Ferguson NM, Laydon D, Nedjati- AC. Social distancing to slow the U.S. sity; 2020 [cited 2020 May 11].
Gilani G, Imai N, Ainslie K, Baguelin COVID-19 epidemic: an interrupted Available from: https://coronavirus
M, et al. Report 9: Impact of non- time-series analysis [Internet]. .jhu.edu/map.html
pharmaceutical interventions (NPIs) Boston (MA): Massachusetts 18 Bursztyn L, Rao A, Roth C,
to reduce COVID19 mortality and General Hospital; 2020 [cited 2020 Yanagizawa-Drott D. Misinforma-
healthcare demand [Internet]. May 11]. Available for download tion During a Pandemic [Internet].
London: Imperial College London; from: https://www.researchgate Chicago (IL): University of Chicago,
2020 Mar 16 [cited 2020 May 11]. .net/publication/340518052_ Becker Friedman Institute for Eco-
Available at: https://www.imperial Social_distancing_to_slow_ nomics; 2020 Apr [cited 2020 May
.ac.uk/media/imperial-college/ the_US_COVID-19_epidemic_an_ 11]. (Working Paper No. 2020-44).
medicine/sph/ide/gida-fellowships/ interrupted_time-series_analysis Available from: https://bfi.uchicago
Imperial-College-COVID19-NPI- 12 Abaluck J, Chevalier JA, Christakis .edu/wp-content/uploads/
modelling-16-03-2020.pdf NA, Forman HP, Kaplan EH, Ko A, BFI_WP_202044.pdf [accessed
6 Thunstrom L, Newbold S, Finnoff D, et al. The Case for Universal Cloth 2020 7 May].
Ashworth M, Shogren JF. The Ben- Mask Adoption and Policies to In- 19 Killeen BD, Wu JY, Shah K,
efits and Costs of Using Social Dis- crease Supply of Medical Masks for Zapaishchykova A, Nikutta P,
tancing to Flatten the Curve for Health Workers [Internet]. New Tamhane A, et al. A County-level
COVID-19 [Internet]. Laramie (WY): Haven (CT): Yale University; 2020 Dataset for Informing the United
University of Wyoming; 2020Mar 27 Apr 1 [cited 2020 May 11]. Available States? Response to COVID-19 [In-
[cited 2020 May 11]. Available from: from: https://papers.ssrn.com/sol3/ ternet]. Baltimore (MD): Johns
https://papers.ssrn.com/sol3/ papers.cfm?abstract_id=3567438 Hopkins University; 2020 [cited
papers.cfm?abstract_id=3561934 13 Harris JE. The Coronavirus Epi- 2020 May 11]. Available from:
7 Abouk R, Heydari B. The Immediate demic Curve is Already Flattening in https://arxiv.org/pdf/2004.00756v1
Effect of COVID-19 Policies on Social New York City [Internet]. Cambridge .pdf
Distancing Behavior in the United (MA): National Bureau of Economic 20 To access the appendix, click on the
States [Internet]. Wayne (NJ): Research; 2020 Apr [cited 2020 Details tab of the article online.
William Paterson University; 2020 May 11]. (NBER Working Paper 21 Saloner B, Maclean JC. Specialty
July 2020 39:7 Health Affairs 7
Downloaded from HealthAffairs.org on June 01, 2020.
Copyright Project HOPE?The People-to-People Health Foundation, Inc.
For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org.
COVID-19
Substance Use Disorder Treatment .gstatic.com/covid19/mobility/ Deaths [Internet]. Stanford (CA):
Admissions Steadily Increased In 2020-04-05_US_Mobility_Report_ Stanford University; 2020 Apr 24
The Four Years After Medicaid Ex- en.pdf [cited 2020 May 11]. (Working Pa-
pansion. Health Aff (Millwood). 26 Benzell S, Collis A, Nicolaides C. per). Available from: https://web
2020;39(3):453?61. Rationing Social Contact During the .stanford.edu/~chadj/
22 Lauer SA, Grantz KH, Bi Q, Jones COVID-19 Pandemic: Transmission Consumption_v_Covid.pdf
FK, Zheng Q, Meredith HR, et al. Risk and Social Benefits of US Lo- 30 Greenstone M, Vishan N. Does So-
The Incubation Period of Coronavi- cations [Internet]. Cambridge (MA): cial Distancing Matter? [Internet].
rus Disease 2019 (COVID-19) From Massachusetts Institute of Technol- Chicago (IL): University of Chicago,
Publicly Reported Confirmed Cases: ogy; 2020 Apr 18 [cited 2020 May Becker Friedman Institute for Eco-
Estimation and Application. Ann 11]. Available from: https://papers nomics; 2020 Mar 30 [cited 2020
Intern Med. 2020 Mar 10. [Epub .ssrn.com/sol3/papers.cfm? May 11]. (Working Paper No. 2020-
ahead of print]. abstract_id=3579678 26). Available from: https://papers
23 US Census Bureau. Census Regions 27 Editorial. Pandemic School Closures: .ssrn.com/sol3/papers.cfm?
and Divisions of the United States Risks and Opportunities. The Lancet abstract_id=3561244
[Internet].Washington (DC): Census Child & Adolescent Health. 2020; 31 Correia S, Luck S, Verner E. Pan-
Bureau; [cited 2020 May 11]. Avail- 4(5):341. demics Depress the Economy, Public
able from: https://www2.census 28 Scherbina AD. Determining the Op- Health Interventions Do Not: Evi-
.gov/geo/pdfs/maps-data/maps/ timal Duration of the COVID-19 dence from the 1918 Flu [Internet].
reference/us_regdiv.pdf Suppression Policy: A Cost-Benefit Washington (DC): Board of Gover-
24 White House. Opening up America Analysis [Internet]. Boston (MA): nors of the Federal Reserve System;
again. Washington (DC): White Brandeis University; 2020 Mar 24 2020 Mar 26 [cited 2020 May 11].
House; 2020. [cited 2020 May 11]. Available from: Available from: https://papers
25 Google. COVID-19 Community Mo- https://papers.ssrn.com/sol3/ .ssrn.com/sol3/papers.cfm?
bility Report [Internet]. [Updated papers.cfm?abstract_id=3562053 abstract_id=3561560
2020 Apr 5; cited 2020 May 11]. 29 Hall RE, Jones CI, Klenow PJ. Trad-
Available from: https://www ing Off Consumption and COVID-19
8 Health Affairs July 2020 39:7
Downloaded from HealthAffairs.org on June 01, 2020.
Copyright Project HOPE?The People-to-People Health Foundation, Inc.
For personal use only. All rights reserved. Reuse permissions at HealthAffairs.org.