Determinants of Small Business Reopening Decisions (NBER)
Summary
NBER Working Paper No. 27362, Determinants of Small Business Reopening Decisions After COVID Restrictions Were Lifted, published by the National Bureau of Economic Research in June 2020 and revised September 2021. Using a survey of small business owners conducted by Alignable in early May of 2020, the paper reports that the plurality of firms reopened within days of the end of legal restrictions, while a minority delayed. The authors find that delays were driven by expectations of stricter regulation and pessimistic demand projections rather than public health concerns, and that a 10% decline in expected demand raises by 1.5 percentage points the likelihood of expecting to stay closed for at least one month. The paper also describes a grant experiment and data sources. It closes with an appendix table of average time to reopen by 3-digit NAICS code.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
Full text
NBER WORKING PAPER SERIES
DETERMINANTS OF SMALL BUSINESS REOPENING DECISIONS
AFTER COVID RESTRICTIONS WERE LIFTED
Dylan Balla-Elliott
Zoë B. Cullen
Edward L. Glaeser
Michael Luca
Christopher T. Stanton
Working Paper 27362
http://www.nber.org/papers/w27362
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
June 2020, Revised September 2021
We thank the founders of Alignable for sharing data. The views expressed herein are those of the
authors and do not necessarily reflect the views of the National Bureau of Economic Research.
At least one co-author has disclosed additional relationships of potential relevance for this
research. Further information is available online at http://www.nber.org/papers/w27362.ack
NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.
© 2020 by Dylan Balla-Elliott, Zoë B. Cullen, Edward L. Glaeser, Michael Luca, and
Christopher T. Stanton. All rights reserved. Short sections of text, not to exceed two paragraphs,
may be quoted without explicit permission provided that full credit, including © notice, is given
to the source.
Determinants of Small Business Reopening Decisions After COVID Restrictions Were Lifted
Dylan Balla-Elliott, Zoë B. Cullen, Edward L. Glaeser, Michael Luca, and Christopher T.
Stanton
NBER Working Paper No. 27362
June 2020, Revised September 2021
JEL No. D22,E32,I15,L23
ABSTRACT
The COVID-19 pandemic led to dramatic economic disruptions, including government-imposed
restrictions that required millions of American businesses to temporarily close. We present three
main facts about business decisions to reopen at the end of the lockdown, using a nation-wide
survey of thousands of small businesses. First, the plurality of firms reopened within days of the
end of legal restrictions, suggesting that the lockdowns were generally binding for businesses -
although a sizable minority delayed their reopening. Second, decisions to delay reopenings were
not driven by public health concerns. Instead, businesses in high-proximity sectors planned to
reopen more slowly because of expectations of stricter regulation rather than concerns about
public health. Third, pessimistic demand projections played the primary role in explaining delays
among firms that could legally reopen. Owners expected demand to be one-third lower than
before the crisis throughout the pandemic. Using experimentally induced shocks to perceived
demand, we find that a 10% decline in expected demand results in a 1.5 percentage point (8%)
increase in the likelihood that firms expected to remain closed for at least one month after being
legally able to open.
Dylan Balla-Elliott Michael Luca
University of Chicago Harvard Business School
dballaelliott@uchicago.edu Soldiers Field
Boston, MA 02163
Zoë B. Cullen and NBER
Rock Center 210 mluca@hbs.edu
Harvard Business School
60 N. Harvard Christopher T. Stanton
Boston, MA 02163 210 Rock Center
and NBER Harvard University
zcullen@hbs.edu Harvard Business School
Boston, MA 02163
Edward L. Glaeser and NBER
Department of Economics christopher.t.stanton@gmail.com
315A Littauer Center
Harvard University
Cambridge, MA 02138 and
NBER
eglaeser@harvard.edu
1. INTRODUCTION
The COVID-19 pandemic led to economic disruptions that have not been seen since the Great
Depression (Baker et al., 2020; Bartik et al., 2020; Forsythe et al., 2020). Government-imposed
restrictions or lockdowns, including regulations on which businesses could operate, forced millions
of businesses throughout the United States to temporarily close. At the time of the lockdowns,
there was some optimism that lifting regulations would result in a speedy recovery. For example,
then-President Trump tweeted that if policymakers were to “reopen our country,” then businesses
would rapidly come back online because “our people want to return to work.” At the same time,
important barriers to reopening may have existed, even absent restrictions. An important input for
understanding the efficacy and need for regulation is understanding how small business owners
responded both to pandemic-related risks and government-imposed operating restrictions.1
There are several factors that might have led a business to delay reopening after restrictions
were lifted. First, business owners might have had concerns about their own health, as reopening
may have exposed them or their employees to COVID-19. Second, the pandemic may have led to
supply chain disruptions and coordination challenges. Third, business owners might have expected
regulations to continue or to snap back into place after lapsing. Fourth, even before any govern-
ment interventions, many households had begun self-isolating to reduce the risk of transmission of
COVID (Couture et al., 2020; Glaeser et al., 2021; Gupta et al., 2020; Sears et al., 2020), leading
to demand reductions that predated the lockdowns. Businesses might have expected lower demand
to continue after restrictions were lifted, which in some cases might provide insufficient margin to
cover the fixed and variable costs of even limited operations.
In this paper, we investigate businesses’ reopening decisions, and factors that slowed the re-
opening process. To do this, we analyze responses from a survey of tens of thousands of small
business owners conducted by the small-business network Alignable in early May of 2020. An in-
1
The pandemic has led to some unusual patterns when compared to typical business cycles. Moscarini & Postel-
Vinay (2012) show that small businesses are less sensitive to standard negative aggregate productivity shocks. The
pandemic recession appears to differ from business cycle fluctuations they consider. We find small businesses are hurt
by anticipated demand-side factors, including lockdowns, in-person restrictions, and lower demand, due to contagion
fears.
2
formation provision experiment on customer demand projections embedded in the survey allows us
to recover causal estimates of the link between expected demand and intentions to reopen. To under-
stand other factors influencing business operations, we merge the survey-experiment results with
O-NET data on workplace conditions, crowd-sourced data on industry attributes, and county-level
data on COVID prevalence and related covariates. We also explore actual reopenings following
the surprise lifting of operating restrictions in two states. Wisconsin and Florida lifted restrictions
after a State Supreme Court order and an executive order, respectively.2 Through the summer and
fall, we conducted follow-up surveys to trace whether owners’ projections about reopening could
be corroborated.
Overall, we find that regulations were often a binding constraint for firms’ operating decisions.
Just under 60% of owners reported an intention to open within a few days of the lifting of legal
restrictions. Our follow-up surveys and difference-in-differences analyses show that lifting reg-
ulations resulted in owners reopening more quickly. Our results suggest that governments were
setting more stringent guidelines for reopening, relative to what many businesses would have se-
lected - which is consistent with the fact that the government is better positioned to make decisions
that account for the externalities of higher foot traffic.
A minority of businesses chose to delay reopening, even beyond the mandated closure. Eigh-
teen percent of firms reported intentions to delay reopening at least one month after the end of
restrictions on their operations, and this estimate is likely a lower bound. This rises to thirty per-
cent if we exclude firms that were open at the time of the survey. Why did these businesses expect
additional delay? Our second main finding is that expectations of prolonged low demand played an
important and causal role in explaining delayed reopenings. Overall, the demand projections were
grim. Heading into the summer of 2020, the average firm in our sample expected that demand for
its services would be 35.3 percent of pre-crisis levels the following fall.3 These demand projec-
tions correlate with political preferences: Republican vote share strongly and positively predicts
2
The executive order extended “full phase one” reopening to all counties. This reopening included the statewide
reopening of indoor dining, retail, gyms, and libraries and museums at 50% capacity.
3
Unfortunately, we do not have price data, unlike Jaravel & O’Connell (2020), so we focus on the share of customers
returning, relative to before the crisis.
3
higher levels of projected future demand. Other factors, such as health concerns and supply chain
concerns, played a more minor role in delaying reopening. For example, only 5% of firms that
were not fully open at the time of the survey cited supply concerns as a barrier that would prevent
reopening. Furthermore, the interaction between local COVID case loads with measures of physi-
cal proximity to coworkers or customers, owner age, or share of high-risk older customers do not
offer additional explanatory power for the timing of individual business reopenings.
We use an instrumental variables approach to establish a causal relationship between expecta-
tions of low demand and and plans to delay reopening. We shift demand projections using a survey
experiment that presented aggregated projections about similar businesses’ anticipated future de-
mand to a subset of owners. We expected that receiving this information would cause owners to
update their own beliefs, as observing information about overall demand may be hard to ascer-
tain for individual owners. Optimistic owners receiving the information rationally shifted beliefs
downward, while pessimistic ones had rosier forecasts after receiving the information treatment.
Comparing owners with similar initial beliefs, but with different information treatments, reveals
that delayed reopening hinged on customer demand. Over the longer term, consumer demand is
correlated with the firm’s reported expected probability of surviving until the end of the year.
Our evidence from early in the pandemic suggests owners were eager to return to work, despite
salient reports on health risks and an uncertain disease progression. This raises important questions:
did small business owners believe they could adapt practices to operate safely? Were they desperate
for operating revenue? Did stimulus packages encourage reopening before owners would have
otherwise chosen to open? We shed light on these questions in this paper. We asked owners whether
they would choose to open or close over two weeks if they received a grant. We randomized the
size of the grant and whether it included a requirement that the business remain closed. When
the grant was not conditional on closing, around three-quarters of business owners chose to open,
regardless of the grant size, even when the grant was close to 0 or as large as $50,000. Hence
we find no evidence that additional funds helped “tide over” businesses, allowing them to cover
fixed costs while remaining closed to weather the health risks of the pandemic. When the grant
4
was conditional on closing, half of small business owners would remain closed for an additional
two weeks in exchange for a modest sum of $2,500, but a quarter would reject $25,000 to reopen
immediately. This indicates that, for about a quarter of businesses, opening was perceived to be
extremely valuable. For most firms, the benefits of reopening for two weeks were limited but of
higher value than being closed. These results suggest that stimulus packages with modest incentives
to change operating status, such as the Payroll Protection Program’s incentives to continue active
employment, likely shifted the decisions of these firms on the margin.
Absent health concerns, a risk-neutral employer might weigh the expected profit of opening and
the carrying costs of closing temporarily versus shutting down. Conditional on remaining open in
some fashion, the firm would also consider the financial returns to mitigating health hazards. These
calculations would depend on firm expectations about customer demand and employee productivity,
given risks of contagion. However, owners making decisions may be just as likely as consumers
to fear contagion personally and may make decisions with these personal fears in mind. Moreover,
recent research on “behavioral firms” (DellaVigna & Gentzkow, 2019) highlights the ways in which
firms may also make systematic mistakes.
The literature on the 2020 pandemic seems to confirm that consumer demand in many industries
fell dramatically because of COVID-19, and many firms closed (Abu-Rayash & Dincer, 2020;
Alexander & Karger, 2020; Baker et al., 2020; Brinkman & Mangum, 2020; Chetty et al., 2020;
Cintia et al., 2020; Coibion, Gorodnichenko, & Weber, 2020; Couture et al., 2020; Dunn, Hood,
& Driessen, 2020; Glaeser, Gorback, & Redding, 2020; Goolsbee & Syverson, 2020; Huang et al.,
2020; Kevin Linka, Alain Goriely, & Ellen Kuhl, 2021; Sheridan et al., 2020; Soucy et al., 2020;
Velde, 2020). Yet there is little clarity about whether firms shuttered because of reduced demand
or other issues, including suppliers’ fears of contracting disease or bankruptcy. Bartik et al. (2020)
provided an early survey of small businesses that found that 45 percent of their sample was not in
operation around April 1, 2020. Closure rates were much higher for businesses that dealt in face-
to-face services, like the arts, than in information services, like finance. Similarly, Fairlie (2020)
found that the “number of active business owners in the United States plummeted by 3.3 million
5
or 22% over the crucial two-month window from February to April 2020.”
Chetty et al. (2020) also examine business closures, which they measure as a business with
zero credit card payments over multiple days, effectively combining temporary and permanent
closures. They estimate that lockdowns explain almost 15 percent of the variation in business
closures over time and space, which is modest but far more than the variation these regulations
explain for spending or employment. They do not try to distinguish whether the businesses closed
because of reduced demand or concerns for worker safety. Papanikolaou & Schmidt (2020) study
the default probabilities implied by the premium firms must pay in their borrowing. They find
both that employment dropped more during the pandemic in industries that cannot easily switch
to remote work, and that markets expected more defaults in those industries. They interpret this
as a labor supply shift, due to the pandemic, but it is difficult to differentiate between consumer
demand and labor supply using their measure. de Vaan et al. (2020) find that the closing decisions
of national brands also influenced the closing decisions of community establishments nearest the
establishments of the parent brand, suggesting local social learning.
Our paper contributes to this literature by assessing the importance of consumer demand, supply-
side factors, and the extent to which business owners internalize the risk of contagion when choos-
ing whether to reopen. Weighing these factors can help inform policy decisions about whether to
regulate businesses and how best to channel resources to them. For example, our results suggest
that stimulus packages that boost consumer spending can encourage the reopening of small busi-
nesses. The results also suggest that the voluntary behavior of small businesses is unlikely to help
mitigate health risks, absent sweeping regulation, as health concerns generally did not prevent busi-
nesses from planning to reopen. And the results further suggest that expectations about depressed
consumer demand caused some firms to delay reopening even after regulations had been lifted.
2. DATA SOURCES
The small business owner data used in our analysis was collected through surveys conducted by
Alignable Inc. Alignable is the largest network and community of small business owners in North
6
America. We combine the survey data with Alignable’s business profiles, O-NET details about
physical proximity of working conditions at the industry level, and data on geographic variation
of COVID-19 cases obtained from The New York Times. Together, these data help illuminate the
factors driving business decisions about whether to reopen.
In this section, we describe the Alignable survey, its representativeness compared to Census
data, and detail the other data sources used in the analysis.
2.1. About Alignable and the Alignable Small Business Survey
The Alignable platform has approximately 5 million registered small businesses across North Amer-
ica. Each week, Alignable distributes a survey link through email to its members. This link allows
the company to merge the individual responses of participants to data from their user profiles.
Our primary sample comes from one wave of Alignable surveys that focused on business re-
opening, with the link emailed to users on May 9, 2020. This survey received 35,069 total responses
to at least one question. 27,263 respondents completed all core questions that form the bulk of the
analysis. The core questions contained several modules. The first module collected information
about the current operational status of the business (fully open, partially open, temporarily closed,
permanently closed) and any potential dependencies with other businesses that may have affected
their decision to fully open or their ability to remain fully open.The second module asked about
expectations about the return of their customers. To obtain an estimate of conditions on the ground,
independent of the operational status of the business, owners were asked about the share of cus-
tomers that would return in the event that they were fully open on a specified date in the future.
That date was randomized over survey respondents, allowing us to trace out owners’ expectations
about the pandemic over varying horizons. To further separate the evolution of the pandemic from
individual circumstances and trace out dependencies between businesses, an additional question
asked about the expected reopening of other businesses. In the third module, respondents were
asked about when they expected legal restrictions impacting their business to be lifted, and when
they would be most likely to re-open fully if they were not fully open at the time of the survey. A
7
final question asked about the likelihood that the business would be operational come December
2020.4
In the middle of the survey, before questions regarding expected demand and expected reopen-
ing and survival, a subset of respondents were shown information about how prior survey respon-
dents had projected demand. The message read “based on your profile, location, and concerns, our
polls show that similar businesses anticipate [X%] of customers will return by [date]. The variable
X was calculated using data from the first 16,038 respondents. One-third of the respondents after
the first batch received this message.”5 The complete survey tool is available in the appendix. Ta-
ble 1 Panel A provides more detail about the data and the measures collected from the main May 9
survey.
We supplement this survey wave with questions from other survey waves. From earlier and later
waves, we can construct a time series of business operational status and demand using responses
from 190,600 unique business owners from March to September 2020. We merge in demographic
details about the owner’s age and industry collected by Alignable in later surveys. We also include
data on industry classification that comes from respondents entering their industry in a text box that
would auto-complete to the text of four- and six-digit NAICS industry descriptions. We also use a
question, delivered toward the end of May in an external survey conducted by Harvard Business
School, to assess how participants would evaluate trade-offs between cash and health considera-
tions. We presented users with a hypothetical grant, in amounts we randomized between $2,500
and $50,000. The grant could be one of two types: either the grant stipulated that the business
would have to remain closed for two weeks to receive the grant, or the grant did not have condi-
tions for receipt. We then asked users whether they would remain closed over the next two weeks
under their particular hypothetical grant condition.
4
A module tracking responses to CARES Act PPP status came prior to the final question about long-term business
operations prospects, causing drop off to 17,098 completed responses for this last question.
5
One-third of respondents saw a different message, but its mapping to a concept like demand is less clear.
8
2.2. Comparison of Survey Responses with U.S. Census Data
One challenge in conducting surveys of businesses is the potential for selection bias. This sample
is selected in two ways: (1) they are firms that have chosen to join Alignable, and (2) they are
Alignable firms that have chosen to take surveys. Bartik et al. (2020) provide a variety of diagnostic
checks for a survey of Alignable businesses conducted from late March to early April to assess its
representativeness and shed light on sample selection. The sample provides broad coverage across
the United States, industries, and business size (within small businesses). Roughly speaking, the
sample matches Census data reasonably well along the dimensions of industry and geography, but
the sample skews toward smaller businesses, relative to the full set of US small businesses. A cross-
validation against a phone survey suggests that these surveys reasonably estimate business closure,
though the random phone survey suggested the survey might overrepresent closed businesses.6 This
is consistent with expectations of Alignable executives, who believe that owners of permanently
closed businesses will be less likely to respond to surveys.
Validation exercises of the May 9 survey wave reach similar conclusions to those of Bartik et al.
(2020), indicating that the survey has nearly representative coverage by firm size and geography
(see Appendix Figure A1). Appendix Figure A2 displays how closures co-vary with the local
COVID caseload and unemployment rate.
2.3. Other Data Sources
Table 1 Panel B provides details about outside data sources that we merged with the Alignable data.
We supplement the survey data with detailed characteristics about the industries of businesses at the
four-digit NAICS level. We determine the extent that each industry can serve online customers and
the likely age distribution of those customers, by posting a description of each industry on Amazon’s
Mechanical Turk and asking questions related to the nature of the industry and its customers. The
6
As a test of selection into taking surveys, Bartik et al. (2020) reports the results of randomly calling 400 business
owners using the contact information collected by Alignable at registration. The current status of these 400 business
owners, open or closed, matches the ratio of open versus closed in a prior survey wave. This suggests that the survey
responses are unlikely to understate the degree of businesses being permanently closed, at least conditional on having
registered with Alignable.
9
first question asked is, “how easy or common would it be for this business to provide services or
goods online?” The second question is, “how likely is it for customers of this business to fall in each
age bracket (listed below)?” We offer answers that correspond with 0-10 percent, 10-25 percent,
25-75 percent, and greater than 75 percent. Five unique individual Mechanical Turk responses were
collected for each industry code and description. We average responses from these individuals at
the industry level. The table presents the raw responses, while later analysis uses Z-Scores for these
variables to ease interpretation.
We collect information at the occupational level about the proximity of employees with each
other and with customers the O-NET proximity variable “To what extent does this job require the
worker to perform job tasks in close physical proximity to other people?” The underlying encoding
of the proximity measure ranges from “I don’t work near other people (beyond 100 ft)” as the lowest
category to “Very close (near touching)” as the highest category. We follow Mongey & Weinberg
(2020) by merging the O-NET version 24 proximity variable to the Occupational Employment
Statistics (OES) data collected by the BLS. The OES data provides a mapping between occupation
codes and NAICS industries. We take the employment weighted average of proximity by four-digit
NAICS code.7 Data about coronavirus cases at the county level were collected by The New York
Times. While coverage is extensive, some counties were grouped together.8
Subsequent surveys help with validation exercises and confirm these early responses. Alignable
asked about operations status at a monthly frequency starting at the end of July. We use the July,
August, and September waves of these surveys to validate our early responses. These surveys
asked respondents about the share of customers returning in the previous month as well as the
current operating status of their business. Using a unique account ID, we can match respondents
in our main survey wave in May to the later surveys. While these surveys are designed as repeated
cross-sections, not as panels, in practice we observe more than 3,000 of the initial respondents in
7
Examples of high proximity industries are retail establishments, personal care (i.e., barber & beauty shops), and
restaurants. In contrast, low proximity industries are insurance agents, legal services, designers/architects, and credit
intermediaries.
8
For example, a single value for New York City is reported, comprising New York, Kings, Queens, Bronx and
Richmond Counties. The data are available at github.com/nytimes/covid-19-data
10
at least one of the subsequent surveys.
3. ANALYTICAL FRAMEWORK
To fix ideas about the drivers of reopening, we consider the following simplified framework. A
business owner’s decision to open is a function g(.) of consumer demand ct , supplier availability
st , owner or employee health concerns ht , and government regulations θt . A government lifts
mandatory business closures for firms when COVID prevalence, xt , is below a threshold. The
thresholds may depend on how operations contribute to contagion. For simplicity we consider
two types of business, high and low proximity, associated with two different regulatory thresholds,
τ high proximity and τ low proximity respectively. Let θt be an indicator for whether the business remains
below the threshold and hence is allowed to be open. Let yt be an indicator for whether a business
is actually open.
Each of ct ; st ; ht ; and θt is a function of COVID prevalence, xt , and other factors outside the
model. Our model predicts that certain characteristics of the business and owner will exacerbate
the reopening response to local COVID prevalence. For example, if health concerns drive reopen-
ing decisions, we predict that personal characteristics of the owner and business, namely owner
age, customer age, and the proximity of employees interacted with local COVID prevalence will
be highly correlated with reopening decisions. If customer demand channels are pivotal for the
reopening decision, the model predicts that the risk characteristics of customers, namely age and
in-person contact, will interact with local case prevalence to predict reopening. If supplier avail-
ability or downstream businesses clients enter the reopening calculus, the case prevalence local to
those businesses will interact with the proximity conditions in those businesses or essential business
status to predict the reopening decision of the businesses we study.
Hence we write expected demand for direct-to-consumer (B2C) firms as a function of the inter-
action of local case loads with customer age and the demand for business-to-business (B2B) firms as
a function of the status of downstream businesses, which may in turn depend on COVID case loads
and regulations in their local environment, c(xt , {customer risk factors}) or c(Ȳt (X̄t , T̄proximity
t )),
11
where Ȳt and X̄t and T̄proximity
t refer to the open-status of downstream businesses and their local
health and regulatory conditions respectively. The decision function is then
proximity
ytype,t = g(ctype (.)t ; s(xt ); h(xt ); θ(xt ; τ proximity )) (1)
where type indicates whether the firm is direct-to-consumer or business-to-business, and ctype (.)
is the corresponding consumer demand function for B2C firms, c(xt , {customer risk factors}) or
B2B firms, c(Ȳt (X̄t , T̄proximity
t )). To gather evidence about the importance of the health channel xt
in determining the choice to reopen, we make use of our theoretical prediction that the relevance
of the local COVID caseload increases with the health risk factors of the owner and employees. In
other words, if the owner’s direct health concerns drive the reopening decision, then the expected
time to reopen should rise with the age of the owner and rise faster with high levels of local exposure
to the coronavirus. If the owner is concerned about the health of customers or liability for their
health, then the expected time to reopen should rise with the share of customers in the high risk
demographic groups, and rise faster when local exposure to the coronavirus is high. We estimate
the equation
Yi =β1 (Owner-Age × COVID case load)+
β2 (Share Older Customers × COVID case load) + γXi + ϵi (2)
We estimate this linear function using ordinary least squares regressions. We then check the
sensitivity to using OLS on censored outcomes.9 We proxy for COVID-19 prevalence in the out-
side population with COVID-19 cases per capita in the county in which the business is located.
9
For open firms, the time to reopening is censored from below at zero. For the closed firms, the time to reopening
is censored above because the latest date for reopening they could report was September or later. To address censoring,
Appendix Table A3 presents results using a Tobit regression. The results are similar in sign, but the Tobit coefficients
when including all businesses are often larger in magnitude.
12
We jointly address exposure to co-workers and customers by using a measure of workers’ physi-
cal proximity to others, based on O-NET data. For most firms in our sample, the owner’s age is
available in Alignable’s administrative data. Customer age is measured through the MTurk survey
instrument detailed in Section 2.
To test whether firms fail to reopen because of problems further back in their supply chain, we
directly ask whether respondents have (or anticipate having) supply problems or problems with
downstream businesses and whether those problems have delayed (or will delay) their reopening.
To test the causal role of government regulation on reopening decisions, we can use sharp
changes in regulations that occurred in mid-May, between two survey rounds, to identify how busi-
nesses changed their behavior and beliefs when restrictions were suddenly lifted. In Wisconsin, a
State Supreme Court decision suddenly lifted restrictions on May 14 10 . In Florida, an executive
order issued by Governor DeSantis on May 14 (effective May 18) accelerated and standardized
reopening statewide. We use a difference-in-difference model to compare the changes between
survey rounds (running from May 9 to May 13 and then from May 14 through June 1) in Florida
and Wisconsin to the changes in the other states.11
We estimate the following equation, where Yi is the reopening decision of business owner i, or
the projected demand of business owner i.
Yi = δ (Reopened Statei × Post) + β1 Post + β2 Reopened Statei + γXi + ϵi (3)
Post is an indicator equal to one for responses after the shocks to regulations. Reopened Statei
is an indicator equal to one for responses in states that experience the surprise reopening.12 Xi
contains separate vectors of fixed effects for states and four-digit NAICS industries, and ϵi captures
10
The decision was announced late in the evening of May 13.
11
While the later survey remained open until the following survey was distributed, over 95% of responses were
collected by May 18.
12
We will estimate three versions of this model: in the first, Wisconsin and Florida are both treated. In the second
and third versions, we consider Wisconsin only and Florida only, respectively. In the specification examining only the
effect in Wisconsin, respondents from Florida are omitted from the regression (and vice versa) to avoid biasing the
estimates.
13
shocks outside the model that we assume are orthogonal to the surprise reopening. Our coefficient
of interest is δ, which identifies the effect of the surprise regulatory changes on outcomes.
To further help us identify the importance of the demand channel, we isolate factors that shift
perceptions of future customer demand and yet are orthogonal to COVID prevalence. For example,
perceptions of future demand can be influenced by the projections of similar business owners (de
Vaan et al., 2020), a dimension we vary experimentally. The survey asked all survey respondents
that were not permanently closed at the time of the survey to report their expectations regarding
demand. After collecting several thousand responses, we could calculate the average expected
demand in each region for various types of businesses.13
We then randomly assigned the remaining respondents to the control or to one of two treatment
arms. In the treatment arms, the survey revealed projected changes in demand or reopening plans
by similar respondents.14 Respondents were then asked about their own beliefs about demand and
finally about their predicted behavior around reopening. Respondents in the control group were
asked for the demand projections without being shown any information. For individuals whose
initial beliefs were below those of similar businesses in the industry by region conditioning set, the
revelation pushes beliefs upward. For those with more pessimistic beliefs, the revelation pushes
beliefs downward. Overall, beliefs for treated firms should be less diffuse after treatment. As a
result, we use the gap between the aggregated information displayed and the initial beliefs as our
instrument.
We can express the process by which the information treatment changes demand projections
with a Bayesian learning model. In a simple framework, the mean of the posterior belief is a
13
We created groups of business by pooling respondents of the same business type (serving business customers or
consumers) who gave similar answers to our survey questions about downstream and upstream business dependencies,
who operated in the same region of the United States, and who were asked about the same date in the future.
14
The precise wording of the messages were: “Based on your profile, location, and concerns, our polls show that
similar businesses anticipate [X1 ]% of customers will return by [date].” and “Before continuing, we want to share
some interesting information. Based on your profile, location, and concerns, our polls show that [X2 ] % of similar
businesses expect to be fully open by [future date].” Date in this case is the same date used for subsequent questions
about expected demand. These estimates are derived from a subset of the earliest responses to the survey. We used
these early responses to estimate these demand signals that were then randomly shown to an experimental subset of
the main body of survey respondents. For the sake of statistical power, we pool both treatment arms.
14
weighted average between the signal and the mean of the prior belief:
Dipost = α · Disignal + (1 − α) · Diprior (4)
where the parameter α is the learning rate. The parameter α ranges from 0 (respondents com-
pletely ignore the signal) to 1 (respondents fully update their beliefs to exactly match the signal),
and depends on the relative precision between the prior belief and the signal (Hoff, 2009).15
Initial beliefs are elicited through a related set of questions at the beginning of the survey before
the information is shared. Because these earlier questions allow us to infer demand expectations
(but do not ask about them directly), we combine the questions that precede the information treat-
ment in a linear model, estimated with ordinary least squares, to predict beliefs for the exact question
about return customers. The prediction model is estimated only using an early batch of respondents
who did not receive any information and who are excluded from the remaining analysis.
Using 16,038 early respondents who were not shown the information treatment, we estimate
the model
D̂iprior
training
= β̂Xitraining + ϵ (5)
We use this β̂ to estimate the implied prior beliefs D̂iprior = β̂Xi using the full matrix X for the
experimental sample. This yields a sample analog D̂iprior of the prior belief distribution in equation
4.
This prediction and the Bayesian learning model combine to provide the first stage of this in-
strumental variables model. Since we experimentally provide the information treatment to only
15
This form of updating characterizes a Bayesian model in which both the prior and posterior distributions are
normal with known variance. In our setting with an unknown variance and a variable (demand) that is bounded at
zero, an exponential likelihood model is more appropriate. To build intuition, we interpret the first-stage coefficients
in the simplified normal-normal setting; interpretation of the parameter α becomes more complicated if beliefs are not
normally distributed (e.g., follow an exponential distribution).
15
some of the sample, we can write the first stage of the instrumental variables specification as
( ) ( )
Dipost = γTi Disignal − D̂iprior + λ Disignal − D̂iprior + βXi + ϵi (6)
where Ti is an indicator that evaluates to one if the respondent received the information treatment
and Xi is a vector of controls, including the projected prior D̂prior . In this model, γ captures
the Bayesian learning parameter, after λ nets out spurious updating (i.e., mean reversion) among
respondents who never receive a signal. In the second stage, we can use the exogenous component
of the shift in the posterior belief distribution to identify the causal effect of changes in demand
projections on reopening plans. This second stage equation is then
Yi = ηDipost + βXi + ξi (7)
where Yi is our outcome of interest, (i.e., expected months until reopening, an indicator for a lag of
at least one month between expected reopening and the expected lifting of restrictions, and finally
the probability that the business expects to be open in December 2020) and Xi is the vector of
control variables. In the baseline specification, Xi contains controls for the projected prior D̂prior ,
the share of similar businesses that are open, current operating status, an indicator for whether the
business is essential, and the future date to which the demand projections correspond. Additionally,
we nonparametrically adjust for time trends in the composition of respondents within the survey
by including controls for the time of the response. In an alternative specification, we add industry
level controls for physical proximity, ease of doing business online, and the share of customers over
65, as well as county level controls for COVID cases per capita, population density, Republican
vote share in the 2016 election, and the age of the owner. We cluster standard errors at the business
type by region cell, which is the cell at which the information signal [X] is defined.
16
4. RESULTS
At the time of the survey (May 9, 2020), 32% of surveyed small businesses were fully open (offer-
ing the same products and services as before the pandemic), 34% were partially open (offering more
limited products and services than before the pandemic), 32% had closed temporarily (offering no
products and services for a temporary period), and the remaining 2% had closed permanently with
no plan to reopen.16 In Figure 1, we show how the operational status of businesses at the time of the
survey vary with respect to industry characteristics, location characteristics, and owner characteris-
tics that are each intended to capture contagion risk experienced by the employees, customers, and
owners.17 Most of these contagion-related characteristics are uncorrelated with operating status,
including the share of sales that can be carried out online, and share of elderly customers at high
risk of serious illness. While measurement error could mask a correlation, the modest relationship
between business operating status and health factors is replicated across a number of alternative
proxies, with two exceptions. The first is proximity status at the industry level, known to be highly
correlated with regulation. High-proximity industries account for a full two-thirds of temporarily
closed businesses but only 37% of fully open businesses. In Panel A of Appendix Figure A4, we
show that this gap persists through later survey rounds in the summer and early fall. We also see
a correlation with the age of the owner and permanent closures. Closed businesses are more than
25% more likely to have an owner over 65, potentially a category of owners who were considering
retirement when the pandemic began.
16
In the survey received by business owners, the question asking about business status specified that fully open and
partially open are respectively defined as ”offering the same” and ”offering more limited” products or services than
before COVID. We point out that an alternative definition of partially open could have been premised on employment.
However, Kurmann, Lale, & Ta (2020) show that businesses very quickly regained employment after reopening. In
our sample, fully open businesses were at 101% of their pre-pandemic employment and partially open businesses
were at 93%. Even if we exclude the top 1% of firms, then fully open businesses were at 97% of their pre-pandemic
employment and partially open businesses were at 92%.
17
Please refer to Section 2 for detailed definitions of each data source represented in this graph.
17
4.1. Are Regulations Binding?
To distinguish between the effects of restrictions per se and the effects of the underlying factors
that drive both restrictions and the reopening decisions, we focus on the roughly 66% of businesses
that are temporarily closed or partially open. We elicit beliefs regarding their reopening timeline,
future regulations (and their expiration), and future demand.
Figure 2 shows the joint distribution of firm expectations about when all lockdown regulations
will end (x-axis) and when they expect to fully reopen their businesses (y-axis). The share of firms
along the diagonal gives us the share of business owners who said that they would reopen fully at
the moment that they were legally allowed to do so. The entries above the diagonal represent those
business owners who expected to wait. 18 The majority expect to reopen as soon as they can do so.
However, there is also a large minority of owners that anticipates a gap between the expected end
of the lockdowns and the expected time of reopening. 19
Respondents in high and low proximity industries faced different regulatory hurdles at the time
of the survey. Panel A of Figure 3 shows the cumulative share of firms in a given industry that ex-
pect to open fully on or before a given date. About 70 percent of firms in low proximity industries
said they were legally allowed to open by the time of the survey (May 9th) in contrast to only 43 per-
cent of high proximity firms. Respondents expected regulatory barriers to partially converge over
the summer. Roughly 90% of firms in low proximity industries expected restrictions preventing
them from reopening to expire by July, compared to roughly 82% of high proximity firms.
For the purpose of our empirical analysis, we underscore that legal restrictions at the time of
the survey and expectations about the evolution of these restrictions are correlated with—but not
perfectly determined by—local COVID conditions. In our framework, different regions choose
18
Somewhat surprisingly, there are also firms that expect to be fully open before the restrictions on fully opening
end. We believe that this reflects the gray area around the words “restrictions” and “fully.” For example, a state order
that mandated social distancing in retail establishments can be interpreted as a limitation on the ability to fully open.
Yet the same business owner who expected that limitation to persist through July might choose to think of himself as
being fully open at the time of the survey or at some other point before July.
19
For all survey respondents that listed August as their expected date of deregulation, at least 23 percent expected
to take some time before reopening. Fifteen percent of firms that expected the restrictions to end by early June also
expected to remain at least partially closed until July. Thirty-one percent of business owners who expected lockdowns
to end in late June expected that they would remain closed until July.
18
different thresholds τt for the imposition of restrictions. Among other explanatory variables, local
politics (proxied by share of Republican votes in the 2016 presidential election) explains substantial
variation in threshold choices, holding fixed local pandemic conditions. In Appendix Figure A3, we
explore heterogeneity across states in the expected length of the lockdowns and show that, despite
large differences, the average gap between when firms expect lockdowns to end and when they
expect to reopen is between one and two weeks.
Focusing on the surprising changes in regulations in Florida and Wisconsin, the timing of which
is arguably orthogonal to local conditions, we corroborate our other findings about regulation’s
impact on the timing of business reopenings. In Table 2, we present evidence that the abrupt lifting
of restrictions increased the share of open businesses in these states by an additional 5.6 percentage
points (15%). We also see a statistically significant but economically small increase in customer
demand projections, conditional on reopening, by 0.4 percentage points (<1%), relative to the
national trend. In this way, we see that that the surprise regulatory reopening caused a meaningful
subset of firms in the affected states to reopen but without a corresponding increase in demand
projections. This suggests that the regulations bind per se, not that business owners think that
regulations directly reduced demand or communicated information about the safety of patronizing
these businesses to consumers. 20
To explore the trade-offs of reopening versus remaining closed for an additional two weeks,
we asked owners what they would choose to do under hypothetical scenarios. They were either
asked about a scenario where they received an unconditional cash grant, or they received a scenario
where the cash grant was conditional on remaining closed for two weeks. In both scenarios, we
randomized the size of the grant. We show the results graphically in Figure 5. Around three-quarters
of businesses chose to open with any sized grant when there was no condition to accepting the grant.
When the grant was conditional on closing, half of small businesses would remain closed for an
20
This is consistent with the literature on consumer behavior that finds a strong relationship between spending and
health considerations, with regulations playing a smaller role (Chetty et al., 2020; Dunn, Hood, & Driessen, 2020;
Goolsbee & Syverson, 2020; Velde, 2020). As mentioned in Section 1, it is now well documented that the fall in
mobility and consumer activity preceded formal restrictions (Abu-Rayash & Dincer, 2020; Huang et al., 2020; Soucy
et al., 2020).
19
additional two weeks in exchange for a modest sum of $2,500, but a quarter of firms would reject
$25,000 to reopen immediately. This indicates that for about a quarter of businesses, opening as
soon as possible was extremely valuable. For most firms, the benefits of reopening for two weeks
were limited but still of value. Finally, we found no evidence that additional funds helped “tide
over” businesses, allowing them to cover fixed costs while closed to weather the health risks of
the pandemic. Small businesses were no more likely to remain closed when offered unconditional
grants of any size: that is, the gradient of reopening with respect to the size of the grant offered is
flat.
4.2. Will Health Fears Deter Reopening?
We now examine the correlation between expected time to reopening and health related variables.
It is challenging to fully capture health concerns on the part of owners. We attempt to do this by
proxying for health risks using such variables as the level of COVID-19 cases, employee proximity,
and owner and customer age. While these proxies are all imperfect, they help provide insight into
the role of health risk.
Table 4 provides our core results. We estimate equation 2, looking at firm expectations about
reopening, future restrictions, and reopening, conditional upon restrictions being lifted.
In regressions (1) and (5), we explore the expected time, in months, to fully reopen. Regressions
(2) and (6) focus on expectations about how many months it would take for restrictions on fully
reopening to be lifted. Regressions (3) and (7) estimate the impact of health-related variables on
expected time to reopening, controlling for the expected number of months until the full lifting of
restrictions.
Regressions (1) through (4) include our entire sample of firms. Regressions (5) through (8) in-
clude only those firms that were not currently open. Both samples have benefits and disadvantages.
Using the entire sample for a table that is focused on barriers to reopening includes many zeros,
as those firms had already either reopened or never been closed. But using the closed subsample
is also problematic, because the sample of firms that were closed looks quite different in low and
20
high COVID counties.
In the first column of Table 4, we look at the overall correlates of expected time to reopening.
The specification includes firms that were already open and does not control for expectations about
the lifting of current restrictions. The first row shows that businesses expected to be closed longer
in counties where the number of COVID cases was higher. Using population-weighted statistics,
the difference between the 90th and 10th percentile of log deaths per capita is 2.85, implying an
opening delay of about seven days between hard-hit and less affected counties.
The next two rows show the impact of worker proximity alone and then the interaction between
worker proximity and COVID-19 prevalence in the county. Employee proximity is a significant
predictor of delayed reopening. A one standard deviation increase in this variable is associated
with a 0.24 month, or 7.5 day, delay in reopening. Perhaps more surprisingly, there is no inter-
action between COVID-19 prevalence and employee proximity. We hypothesized that employee
proximity would be more problematic in high COVID environments, but there is little evidence
that this interaction entered into firms’ expectations about reopening. Panel B of Figure 3 presents
the cumulative distributions of total time to reopening by industry. Patterns resemble the patterns
regarding time to restrictions being lifted in Panel A, with low proximity industries opening sooner.
The fourth and fifth rows look at owner age and the interaction with COVID prevalence. We
expected that reopening would be less attractive to older owners, who face greater mortality risk
from COVID, and that this effect would be larger in high COVID environments. But older owners
did not seem to expect to delay reopening, and there is no significant interaction between age and
the prevalence of the pandemic.
The sixth and seventh rows look at customer age and interactions with COVID prevalence. We
expected to find that firms with older customers would have been more likely to delay their opening,
either because of reduced demand from skittish customers or out of concern for customers or legal
liability. The coefficient goes in the opposite direction, where owners of firms that served older
customers expected that they were more likely to open sooner. One possible explanation for this
fact is that firms serving older customers specialize in products, including health services, that are
21
more likely to face robust demand. We also do not find a positive interaction between customer
age and the COVID rate in the county.
The eighth row shows that essential businesses expect that they will open 0.3 months (or nine
days) sooner than non-essential businesses. The ninth row shows the ease of operating online. This
variable does predict an earlier reopening, but the effect is relatively small.
The last two rows show the impact of our two other county level variables: density and the
Republican vote share in 2016. Density is negatively associated with time to reopening, either
because of health-related concerns or because of regulation. Republican vote share is even more
strongly negatively related to time to reopening.
The second column attempts to separate expectations about regulation alone from other firm
beliefs about their own decisions. The outcome variable in this column is the number of months
until all restrictions on business for this firm are lifted. Somewhat remarkably, almost all of the
coefficients are quite close to the coefficients estimated in the first column. For example, a 100
percent increase in the number of COVID cases per capita is associated with a 0.08 month increase
in the amount of time until all restrictions are lifted. The similarity of slopes with respect to health
concerns and other factors suggests that a constant offset between lifting restrictions and reopening
fits the data quite well. For example, a one standard deviation increase in physical proximity is
associated with a 0.3 month increase in the expected time until restrictions are lifted. The coefficient
is larger but similar to Column 1.
One modest difference between the two columns is that owner age is negatively associated with
the expected time until restrictions are lifted. That effect withstands county fixed effects, which
is shown in Appendix Table A4, so it does not reflect any spatial correlation between owner age
and local regulatory regimes. Older owners may have been in industries that were less subject
to local regulation, or they may have just been more optimistic. Overall, the second regression
shows that our proxies for health concerns, when they matter for delays at all, seem to matter just
as much for prognostications about the end of regulation. Consequently, health fears may figure
little in deterring firms’ reopening patterns. To test this hypothesis, the third column examines
22
expectations about reopening, controlling for the expected time until restrictions are expected to be
lifted. The coefficients in this column can be interpreted as telling us whether particular variables
predict delays after reopening becomes legally feasible.
If firms intended to delay reopening because of health fears for either their workers or customers,
then we would expect many of these coefficients to be significant both statistically and in magnitude.
Yet we find that almost none of them are. Both the COVID-case and physical-proximity coefficients
retain statistical significance, but they are much smaller in size. The COVID-case coefficient drops
by about 75 percent between regressions (1) and (3). The coefficient on employee proximity drops
by 85 percent.
Figure 4 shows the gap in post-lockdown reopening between high and low proximity industries.
There is no visible difference in time to reopening after lockdowns end. As we have already seen,
this fact does not imply that there is no delay after the restrictions end. There is a delay, but the
average delay seems to be essentially independent of the duration of the restrictions and relates only
loosely to the health-related factors that we have explored. Instead, regulations appear to explain
most of the variation in reopening times.
Policymakers and public health officials should take note of a key distinction between the role
of regulations in binding consumer and business behavior. The difference between the reopening
plans of high and low proximity businesses was driven by differences in the expected duration of
restrictions. The reopening plans of high and low proximity businesses are essentially identical
once we adjust for these differences in when owners expected to be able to reopen. This contrasts
with how consumers responded; travel, for example, fell before restrictions were implemented.
Further, local case prevalence can explain travel reductions, even after controlling for regulations
(Brinkman & Mangum, 2020).
In this way, individual contagion concerns complemented government restrictions to limit travel
to locations with the highest caseloads. We don’t find evidence of this same kind of complemen-
tarity among business owners. That is, owners of high-proximity businesses did not analogously
plan to delay reopening beyond the legal requirements. Policymakers should note that this suggests
23
legal restrictions were the mechanism by which business owners expected health concerns to affect
their reopening plans.
Regression (4) of Table 4 considers an indicator for a reopening time greater than one month
from the lifting of restrictions. At the mean, 17.7% of the sample reports their planned date of
fully reopening will occur more than four weeks after the date they believe restrictions will end.
This estimate is likely a lower bound because we cannot calculate this lag for firms that believe
restrictions will end after August. Over 80 percent of firms anticipate reopening within a month
of being able to do so, but a significant share anticipate drawn out delays before fully reopening.
There are only two significant variables, essential business and Republican vote share, suggesting
that much of the variation in long delays is unrelated to health concerns.
Regression (5)-(8) repeat these regressions looking only at those firms that were closed or par-
tially open at the time of the survey. These firms are a selected sample, and the selection depends
on COVID cases at the county level. A larger share of businesses were not fully open in counties
with high levels of COVID. Panel A of Appendix Figure A2 shows the relationship between the
share of businesses that were open and the level of COVID at the county level across counties with
more than 110 businesses in our sample. Over 40 percent of firms were fully open in the counties
with low COVID rates. Less than 20 percent of firms were open in the counties near New York City
that had the highest COVID rates. This selection may explain why the relationship between the
level of COVID cases and predicted time to reopening is weaker among firms that were then closed.
In the high COVID counties, most firms were closed, and many of these firms had attributes that
would facilitate reopening. In the low COVID counties, the firms well suited for being open were
already open, and consequently only the most vulnerable firms are closed. This selection problem
inhibits interpretation of all of the county-level variables in this later sample.
Only a few variables are significant in regression (5). Owners of firms with older customers
expected they would open sooner. Those in essential industries expected they would open sooner.
And those in counties with a higher Republican vote share also expected to open sooner.
The sixth regression again examines beliefs about when regulations would end. Those same co-
24
efficients again predict expectations about deregulation. Owners of essential businesses expected
regulations to end more quickly. Owners with older customers and those in more Republican coun-
ties also expected regulations would end sooner.
In the seventh regression, we look at the correlates of expected time to reopening, conditional
on expectations about legal restrictions. The patterns in this regression are broadly similar to those
before, except physical proximity and the measure of local pandemic severity become insignificant
for explaining the lag among these businesses. These patterns continue to hold in column 8, where
28% of the businesses that were not fully open anticipated having delays in reopening greater than
one month. The primary difference between this and other columns is that the coefficient on local
COVID caseload is much smaller than in any other outcome, underscoring that the set of businesses
in this regression are selected based on differences in county characteristics.
This table and the related figures tell a clear story that health concerns matter greatly for regula-
tion but much less so for firms’ behavior post-regulation. Firms with older customers expected to
reopen sooner rather than later. Greater COVID-19 prevalence predicts expected regulatory delay
but not economically significant differences in firm choices, absent regulation. We interpret this as
suggesting that firms opening behavior might suggest there are health concerns, but digging deeper
suggests these patterns arise because of regulations.
Another piece of evidence that supports this view is shown in Figure 5. We gave respondents a
hypothetical question about whether they would be willing to remain closed if they received either
an unconditional grant or a grant conditional upon remaining closed. We randomly varied the size
of the hypothetical grant. If owners wanted to remain at home because of health fears, then we
would expect the unconditional grant to have a large impact that increased with the size of the
grant, as larger grants would allow owners to consume or pay their bills without the need to access
cash flows generated from their businesses. Reopening decisions were invariant to the size of the
cash grant, which we interpret to suggest that owners minimally traded off liquidity concerns with
worries about well-being.21
21
These findings contrast with other work that shows commuting often slowed dramatically before lockdown reg-
ulations were implemented, suggesting that some firms stopped in-person work before being forced to do so. On
25
This suggests there is substantial residual variation in reopening times that is not captured by
average health risk, conditions on the ground (cases, density), industry characteristics (proximity,
essential), or attitudes (GOP vote share). While regulation explains a substantial portion of the
reopening variability, much remains. We explore two additional hypotheses in the next section: co-
ordination with other businesses in the ecosystem and reductions in (or uncertainty about) demand.
4.3. Reopening and Coordination Between Customers and Suppliers
Figure 6 illustrates the complementary nature of businesses throughout the United States. The top
panel asks those business owners who were currently open “Although you are currently open, if
these other businesses closed, would it affect your ability to remain open? (Select the category that
matters most.)” Thirty-six percent of owners of open businesses said their ability to remain open
would be impacted if their customers closed. A business’ survival naturally depends on demand for
its services or products. Suppliers mattered less than customers among this group, but both were
important. A smaller share also cited the importance of businesses that refer them customers. If we
add together the businesses that refer and the business customers, we find that almost 50 percent of
firms emphasized downstream linkages. That share is almost double the 25 percent of firms that
highlighted upstream linkages.
This difference between upstream and downstream connections is also shown in the bottom
panel of Figure 6. This panel shows the responses to a question that was asked only of firms that
were temporarily closed or partially open: “are you waiting on other businesses to open before
fully opening yourself?” Somewhat surprisingly, more than half of our small business owners said
no. Most of currently closed businesses did not require any coordination with other businesses.
Sixty-five percent of respondents to this question were in consumer-facing businesses, helping to
explain these results. For the business-to-business respondents presented with this question, it is
likely that their business customers were already open at the time of the survey.
the reopening question, our analysis would point to more firms reopening quickly, but our analysis may be putting
more weight on small firms that had lower capacity for telecommuting or were less exposed to potential health-related
lawsuits.
26
Nonetheless, almost half of business owners did note that they were waiting on other businesses.
The largest category in this group was firms waiting on business customers. Together, more than
20 percent of respondents said that they were waiting for either customers and businesses that refer
customers to them. This represents more than 40 percent of the dependency in this sample.
Another 20 percent said that they were waiting on businesses that were similar to theirs to open.
While we might usually think that the reopening of competing businesses would depress demand
for a particular enterprise, the respondents seemed to take the opening of competitors as a signal
that demand had returned. There might also have been some advantage to waiting and learning
from the reopenings of peers (de Vaan et al., 2020).
Only 5 percent of respondents cited the need to wait until suppliers had reopened. This share
does not mean that suppliers are unimportant. The top panel confirmed that, if supply relationships
ended, then this could stifle a business. Instead, this means that currently closed firms were not
worried as much about supply, presumably because upstream firms were more likely to be open or
because global supply chains allowed them to source from somewhere else. If upstream suppliers
produced goods in lower density factories, then it was likely easier for them to stay open.
These results confirm the importance of linkages for reopening but also suggest that slightly
more than half of closed firms in May 2020 could reopen without any other firm reopening as well.
The results suggest that downstream linkages seem likely to be most important. For that reason, we
now turn to the firm’s forecasts about future demand and the impact of future demand on projected
reopening behavior.
4.4. Forecasting Post-Crisis Demand
We start with the firms’ forecasts about future demand. The survey asked owners to predict what
share of their pre-COVID demand would return. The future date was one of six randomized dates
ranging from early May to September 2020. The exact wording was: “If you are fully open in
[date], what share of your customers do you expect at that time, compared to before the crisis?
27
Please provide your best guess.” Response options were top-coded at “greater than 90 percent.” 22
On average, across all industries, demand was expected to return to 65% of its pre-COVID
level by September. Appendix Table A7 reports both the share of firms that expected their demand
to fully return (90 percent or more of their pre-crisis levels) and reports the mean level of demand
predicted, again relative to pre-crisis levels.23
Face-to-face sectors, including educational services, retail trade and restaurants and accommo-
dation, all expected large decreases in demand through September. For example, accommodation
and food service providers expected their demand to be 58 percent of its pre-crisis level in Septem-
ber. Similarly, arts, entertainment, and recreation only expected demand to be at 55 percent of
pre-crisis levels by September.
In contrast, finance and insurance appears to be the sector with the smallest reductions in de-
mand - with financial firms expecting demand to return to 70 percent of pre-crisis levels by Septem-
ber. More broadly, industries that deliver information-intensive products tended to be more opti-
mistic about future demand.
Before we examine whether these drops in projected demand can explain slow rates of planned
reopening, we turn to a more systematic exploration of the correlates of predicted drops in demand.
Table 5 shows predictors of demand expectations for all businesses (column 1) and businesses
that were not fully open (column 2). The regressions pool results for projected demand across future
months and include a control for the reference month that was contained in the survey question. To
separate the impact of regulations from other factors, we control for the months until reopening
restrictions were lifted. In both columns, the length of delay until the lifting of restrictions is
associated with lower levels of expected demand. One more month of restrictions is associated with
17.4 percent lower projected demand in the entire sample and a 13 percent reduction in projected
demand in the sample that was currently not fully open.
One interpretation of the correlation between the expected length of restrictions and the reduc-
22
Our estimates may miss some reallocation of demand because of top-coding of survey responses (Barrero, Bloom,
& Davis, 2020). Table A7 allows an assessment by examining the share of responses indicating demand would exceed
90% of its pre-pandemic level.
23
Appendix Table A8 shows a more granular industry breakdown.
28
tion in projected demand could be that firms anticipated that consumers would switch to alternative
suppliers and products if the delay lasted longer. In this case, the lost demand might be recouped
across different sectors of the economy, even though a specific firm had lost customers. An alter-
native interpretation is that restrictions were correlated with reduced demand because both reflect
omitted factors, such as aspects of the health crisis not captured by our COVID case measure.
With the exception of employee proximity, most of our health related variables are not corre-
lated with projected demand. The level of COVID cases itself is unrelated to the expected drop
in demand. Owner age is uncorrelated with projected future demand, while customer age is pos-
itively correlated. Presumably, this reflects the tendency of older customers to have more stable
consumption patterns and to purchase services, like healthcare, that they are likely to need going
forward.
A notable exception is businesses where employee proximity is higher. A standard deviation
increase in proximity reduces demand forecasts by roughly 8 percent across specifications. There
is also a negative interaction effect with COVID cases, and magnitudes are larger for businesses
that were not fully open. Comparing these magnitudes for demand reductions in high proximity
businesses to these businesses’ reopening plans suggests that, despite the potential for demand to
decline, owners intended to reopen high proximity businesses to serve a smaller customer base.
These workplaces appear able to operate at a smaller scale than their pre-pandemic levels, possibly
because the opportunity cost of operating (a service provider’s outside option) deteriorated.
Two other industry-specific variables also predict demand. Projected demand is 11 or 12 percent
higher for essential businesses than for non-essential ones. If demand were not top-coded at “greater
than 90 percent”, we might have detected an even larger boost in projected demand for essential
businesses.
There is also a greater drop in projected demand for businesses than can be performed online.
One interpretation is that the business owners in our sample expected that they would lose their cus-
tomers to online competitors. An alternative view is that ease of online delivery captures relatively
non-essential services.
29
Two place-based variables predict expected future demand. Future demand is generally higher
in denser areas, possibly because these markets will facilitate finding a new group of customers.
Future demand relates strongly to the share of Republican voters in 2016.
4.5. The Impact of Demand on Reopening
We now turn to the impact that projected demand has on future reopening intentions. To identify the
causal effect of demand projections on reopening plans (generally and conditional on restrictions
lifting) and long-term business viability, we use experimental variation in information provision
about future demand.
We turn to the estimation of equations 6 and 7, our IV approach, using experimental variation in
information provision about future demand. Panel B of Table 6 displays the first-stage regression.
The instrument, which is the interaction between receiving information in the survey and the differ-
ence in the logarithm of the signal and the constructed prior belief, has a strong positive impact on
predicted demand. A 10 percent larger gap between the signal and prior, leads just under a 2 percent
increase in the owner’s projected demand. This shows the posterior beliefs move in the direction
of the signal. Throughout this table, we present results with a standard set of controls germane to
the instrumental variable specification in particular. We also add columns with an additional set of
controls from the more expansive OLS specification in Table 4. Results are stable across these two
alternative specifications.
Columns 3-10 show reduced form estimates, where the various outcomes (lags to reopening,
lags to reopening with restriction date fixed effects, indicators for long lags, and indicators for long-
run prospects) are regressed directly upon the instrument. These results are again stable across
specifications. The reduced form coefficients show the importance of the instrument, presumably
through the demand channel, on these outcomes.
Panel A presents the two-stage least squares estimates of the causal effect of changes in pro-
jected demand. In column 1, we estimate that a 10 percent increase in projected demand decreases
the time to reopen by 0.088 months or roughly 2.7 days. This point estimate is stable when we
30
include additional controls for a range of industry (proximity, ease of conducting business online,
etc.) and geography (COVID cases, population density, GOP vote share, etc.). However, when
we include fixed effects for the projected date that restrictions will be lifted, the coefficient falls to
0.53, meaning that a 10 percent increase in projected demand will reduce planned time to open by
about 1.6 days.
The estimates in columns 1-4 reflect changes in the average planned time to reopen caused by
shifts in demand projections. However, these means necessarily obscure differences across various
margins. Columns 5-6 examine lags of greater than one month. Here a 10 percent increase in
demand reduces the probability of a long delay in reopening by about 1.6 percentage points or an 8
percent reduction relative to the mean. This highlights the long tail of reopening times and suggests
that pessimistic owners are influenced by changes in their demand projections.
In column (7), we look at the probability of being operational by the end of 2020 as our depen-
dent variable. A 10 log point increase in the share of customers expected to return increases the
expected probability of survival by 3.1 percentage points. In other words, a 20 percent increase
in projected demand increases expected survival probability by six percentage points on average.
Given that the mean failure rate is 21 percent, a drop from 21 to 15 percent is economically highly
significant.
While we are interested in the impact of expectations of demand on reopening and survival
decisions, providing information on demand can also indirectly affect beliefs about regulation. To
account for this, we control for expectations of regulation, which we collected after the information
intervention as well. Columns 3 and 4 show that expectations of demand seem to have a large and
direct impact on reopening decisions, even after controlling for expectations about regulation.
4.6. The Impact of Demand and Other Variables on Survival
One of the most important questions about COVID-related lockdowns is whether a temporary pe-
riod of firm closure leads to permanent elimination of thousands or millions of American businesses.
Consequently, we now examine whether any of our variables predict survival until December 2020.
31
We have already estimated the impact of projected demand on expected survival in the last two
columns of Table 6, but we have not linked this survival rate with any of our other variables. In
both exercises, projected demand relates positively with long-run projected survival rates, often
substantially so. Although tracing out actual survival ex-post is notoriously difficult, these pro-
jected survival rates have been shown to correlate with follow-up phone audits done by Bartik
et al. (2021).
In Table 7, we build in the correlation between our core set of additional variables and the ex-
pected probability of survival until December. The first two rows look at the impact of projected
demand and months until the end of restrictions are anticipated. Projected demand positively pre-
dicts survival expectations, but the estimated coefficient is smaller than in all of the two-stage least
squares estimations.
The most striking and important fact is that the length of expected restrictions is strongly neg-
atively associated with the anticipated probability of survival. As the expected restriction duration
increases by one month, the expected probability of survival drops by 2.6 percentage points. This
fact does not mean that restrictions are wrong, but it does suggest that the economic cost of longer
lockdowns, especially as experienced by small entrepreneurs, is likely to be large.24
Three other variables are significant in every specification. Essential businesses were between
1.3 and 2.1 percentage points more likely to survive in expectation. This gap could reflect the
advantage of being able to continue in business throughout the crisis, or it could reflect more stable
demand for essential businesses. Firms with higher worker proximity were less likely to report
optimism about survival. A one standard deviation increase in worker proximity is associated with
at least a 1.7 percentage point decrease in the expected probability of survival. This may reflect
the expected difficulty of operating in a high contact work environment. Finally, businesses with
older customers have higher expectations of survival, possibly because this customer base was
more stable. None of the other variables have reliable correlations with the projected probability
24
Past work, since at least Hamilton (2000), suggests that many small businesses are likely to be fragile even in
good times. Related work studies how business owners respond to shocks over their careers (Catherine, 2019; Dillon
& Stanton, 2017; Hincapié, 2020).
32
of survival.
4.7. Reliability of Demand and Reopening Projections
Subsequent surveys help with validation exercises and confirm these early responses. In the subse-
quent Alignable series, cross-sections of small business owners were surveyed at the end of July,
August, and September. These surveys asked about the share of customers returning in the previous
month as well as the current operating status of their business. Using a unique account ID, we can
match respondents in our main survey wave in May to the later surveys. While these surveys are
designed as repeated cross-sections, not as panels, in practice we observe more than 3,000 of the
initial respondents in these later surveys.
In Panel A of Table A5, we compare demand projections to the retrospective realized demand
when resurveyed. Respondents with the most optimistic demand projections—those projecting
more than 90% of their pre-COVID customers returning by the date in question—report relative
demand of 93% in July, 87% in August, and 89% in September. In these waves, roughly 85% to
90% of these respondents report being fully open.
In contrast, respondents with the lowest demand projections—those anticipating less than 10%
of their pre-COVID demand—report 44%, 50%, and 37% of their customers returning in July, Au-
gust, and September, respectively. While these figures are significantly higher than 10 percent,
only roughly 30%-40% of these businesses were open when resurveyed. Since we only observe
retrospective demand for businesses that were open in the preceding month, we don’t see the coun-
terfactual demand level for the majority of these businesses that are still closed later in the summer.
In this way, these demand estimates come from the best performing third of these businesses. Even
conditional on being open when surveyed, demand in this group is roughly half of the realized
demand for respondents projecting over 90% of their customers returning.
In Panel B, we can compare the projected reopening date to the share of businesses that are open
in each successive survey round. At each survey round throughout the summer, respondents who
projected reopening by early May were two to three times as likely to have been open than their
33
counterparts who projected remaining closed through September or later. While there is a strong
gradient, small business owners do appear to be optimistic in their reopening estimates. When we
resurveyed respondents in July, 80% of respondents who projected reopening in early May were
open, but only 25% of those who projected reopening in September were open. However, this
means that, even in the most optimistic group, 20% of business owners who projected being open
in May remained closed in July; by September, as case loads began to rise, 28% of respondents
who projected reopening in early May were closed. By September, only 35% of businesses who
projected reopening in August were open.
The reopening projections have strong explanatory power; the reopening regressions displayed
in Panel B are linear probability models that regress an indicator for being open on a vector of
indicators that correspond to each possible reopening date; the R2 statistics range from 0.71 to
0.73. While these projections correlate well with realized reopenings, the projections appear to
systematically underestimate delays in reopening.
In Appendix Table A6, we show that differential attrition as a function of the variables of in-
terest is minimal. The industry distribution is broadly similar in the baseline and re-sampled wave.
There is a statistically significant but economically small shift away from finance & insurance and
professional services ( two percentage points in each category) and toward retail (+ 2pp) and other
services (+ 1.5pp). The geographic mix remains broadly stable.
We do see a slight shift in the composition of the small business owners with respect to the
business status in May when we compare the subset who replied to the later survey to the full
survey, but these shifts are modest. The share of businesses that were fully open in May is roughly
three percentage points higher in the subset that can be matched to a later round. However, these
changes are qualitatively small: businesses fully open in May were 31.6% of the baseline sample
and 34.8% of the validation sample; partially open businesses shift in the opposite direction from
34% of the main sample to 32% of the validation sample.
The composition with respect to the share of customers has similarly small shifts. Perhaps
counterintuitively, business owners projecting fewer than 10% of their customers returning make
34
up a slightly larger share of the validation sample than the full sample (13.7% relative to 10% in the
baseline), and there is a similar shift in the opposite direction for businesses projecting 50%-75%
relative demand (22.3% relative to 25.1% in the baseline). These differences in weights are too
modest to threaten the qualitative results of this exercise.
Additionally, we validate that the Alignable measures on business operations are correlated
with administrative data on labor market performance. Panel B of Appendix Figure A2 plots the
share of business that were currently open against the county level unemployment rate, indicating
that the impact is not being felt equally nationwide. The striking correlations suggest the Alignable
measures accurately portray economic activity at a granular level. Of course, as mentioned above,
part of the county differences are driven by differences in regulations.
A final important driver of behavior and expectations is the partisan environment. People liv-
ing in strongly Democratic or strongly Republican areas of the country made decisions based on
different information about the underlying health risks posed by the virus (Bursztyn et al., 2020).
We find that county-level Republican vote share in the 2016 election is very strongly correlated
with all of the outcomes of interest. In Figure 7, we show that Republican vote share is associated
with expectations of earlier reopening, removal of restrictions, and shorter gaps between reopen-
ing and when it is legal to do so. It is striking that this result is robust to controls for population,
population density, COVID prevalence, state and granular industry (NAICS four digit). By consid-
ering the time to reopen conditional on it being legal to do so, the estimates of the delay in Panel C
even adjust for differences in the regulatory environment. This raises important questions that are
beyond the scope of this paper. How much this effect reflects attitudinal differences of business
owners in these counties? How much is driven by differences in customer demand caused by dif-
ferent attitudes among local consumers? And how much is driven by misinformation about health
risks, rather than different levels of risk tolerance? Understanding how partisan and ideological
commitments interact with public health interventions is of vital importance for policymakers.
35
5. CONCLUSION
The Alignable Survey of Small Business Owners provides a snapshot of small business behavior
and expectations during the COVID-19 crisis. Firms gradually reopened, but those in some places
reopened faster than those in others.
Although restrictions influenced the reopening decision, many businesses expected to delay re-
opening when restrictions were lifted. The average business in our sample expected to be closed
two weeks longer than the restrictions lasted, although some businesses expected to be closed for
months after they were legally allowed to reopen. When considering future regulations, policy-
makers will likely seek to understand both average behavior and sources of heterogeneity. Were
owners responding to health concerns for themselves, changes in demand, or other factors? Which
of these arise due to regulations themselves, and which of these would remain absent restrictions
on behavior?
The delay in reopening did not appear to relate to health concerns, at least for the small busi-
nesses in the survey. The lag between the predicted end of restrictions on operations and the pre-
dicted time for reopening is not correlated with any of our measures of health risk. Neither older
customers nor older owners predicted longer delays after the end of restrictions. And while COVID
case prevalence predicted the presence of restrictions on operations into the future, COVID cases
per capita did not predict delays in opening after restrictions on operations were lifted. These facts
suggest that small firms’ reopenings were driven more by their economic needs to survive than by
their worries about public health.
Several other findings underscore the importance of demand projections and interdependencies
among businesses for owners’ reopening decisions. We use an information provision experiment to
show that the reopening decision depends on expectations about demand. If downstream businesses
don’t open, then this will ripple through the network of firms.25 Adding to the headwinds businesses
face, this crisis was—and continues to be—both a health crisis and an economic crisis. Businesses
expected demand for their services to be greatly depressed for many months, likely justifying some
25
See Akbarpour et al. (2020) for a discussion of other aspects of networks related to reopening policy.
36
of the government aid to businesses, which aimed to allow them to weather lower projected demand
while health risks to consumers lingered.
Entrepreneurs can play an important role in post disaster situations. For example, Chamlee-
Wright & Storr (2010, 2014) document the role of social and commercial entrepreneurs in resilience
to and recovery from Hurricane Katrina. Storr, Haeffele-Balch, & Grube (2015) tell a wider tale
of how entrepreneurs provide a protective ecosystem that enabled the recovery of New York after
Hurricane Sandy and New Orleans after Katrina. Our findings examine the current ongoing COVID
crisis and illuminate the barriers small businesses faced when deciding about reopening.
37
References
Abu-Rayash, A., & Dincer, I. (2020). Analysis of mobility trends during the COVID-19 coron-
avirus pandemic: Exploring the impacts on global aviation and travel in selected cities. Energy
Research and Social Science, 68.
Akbarpour, M., Cook, C., Marzuoli, A., Mongey, S., Nagaraj, A., Saccarola, M., Tebaldi, P. …
Yang, H. (2020). Socioeconomic Network Heterogeneity and Pandemic Policy Response. NBER
Working Paper No. 27374. Cambridge, MA: National Bureau of Economic Research. Retrieved
December 31, 2020, from www.nber.org/papers/w27374.
Alexander, D., & Karger, E. (2020). Do stay-at-home orders cause people to stay at home? Ef-
fects of stay-at-home orders on consumer behavior. Mimeo. Retrieved March 8, 2021, from
www.chicagofed.org/publications/working-papers/2020/2020-12.
Baker, S., Bloom, N., Davis, S., Kost, K., Sammon, M., & Viratyosin, T. (2020). The Unprece-
dented Stock Market Impact of COVID-19. NBER Working Paper No. 26945. Cambridge, MA:
National Bureau of Economic Research. Retrieved December 31, 2020, from www.nber.org/papers/w26945.
Barrero, J. M., Bloom, N., & Davis, S. (2020). COVID-19 Is Also a Reallocation Shock. Technical
report, National Bureau of Economic Research, Cambridge, MA. NBER Working Paper No.
27137. Cambridge, MA: National Bureau of Economic Research. Retrieved December 31, 2020,
from www.nber.org/papers/w27137.
Bartik, A. W., Bertrand, M., Cullen, Z., Glaeser, E. L., Luca, M., & Stanton, C. (2020). The
impact of COVID-19 on small business outcomes and expectations. Proceedings of the National
Academy of Sciences of the United States of America, 117, 17656–17666.
Bartik, A. W., Cullen, Z., Glaeser, E. L., Luca, M., & Stanton, C. (2021). The Targeting and
Impact of Paycheck Protection Program Loans to Small Businesses. NBER Working Paper No.
27623. Cambridge, MA: National Bureau of Economic Research. Retrieved July 1, 2021, from
www.nber.org/papers/w27623.
Brinkman, J., & Mangum, K. (2020). The Geography of Travel Behavior in the Early Phase of the
Covid-19 Pandemic. Mimeo. Retrieved March 30, 2021, from doi.org/10.21799/frbp.wp.2020.38
Bursztyn, L., Rao, A., Roth, C., & Yanagizawa-Drott, D. (2020). Misinformation During a Pan-
demic. NBER Working Paper No. 27417. Cambridge, MA: National Bureau of Economic
Research. Retrieved July 1, 2021, from www.nber.org/papers/w27417.
Catherine, S. (2019). Keeping Options Open: What Motivates Entrepreneurs? Mimeo. Retrieved
December 31, 2020, from ssrn.com/abstract=3274879.
Chamlee-Wright, E., & Storr, V. H. (2010). The role of social entrepreneurship in post-Katrina
community recovery. International Journal of Innovation and Regional Development, 2, 149-
164.
38
Chamlee-Wright, E., & Storr, V. H. (2014). Commercial relationships and spaces after disaster.
Society, 51, 656-664.
Chetty, R., Friedman, J., Hendren, N., Stepner, M., & Team, T. O. I. (2020). The Economic Impacts
of COVID-19: Evidence from a New Public Database Built Using Private Sector Data. NBER
Working Paper No. 27431. Cambridge, MA: National Bureau of Economic Research. Retrieved
March 8, 2021, from www.nber.org/papers/w27431.
Cintia, P., Pappalardo, L., Rinzivillo, S., Fadda, D., Boschi, T., Giannotti, F., ... & Pedreschi, D.
(2020). The relationship between human mobility and viral transmissibility during the COVID-
19 epidemics in Italy. Retrieved March 8, 2021 from arxiv.org/abs/2006.03141.
Coibion, O., Gorodnichenko, Y., & Weber, M. (2020). Labor Markets During the COVID-19 Crisis:
A Preliminary View. NBER Working Paper No. 27017. Cambridge, MA: National Bureau of
Economic Research. Retrieved March 8, 2021, from www.nber.org/papers/w27017.
Couture, V., Dingel, J. I., Green, A., Williams, K. R., & Handbury, J. (2020). Measuring movement
and social contact with smartphone data: A real-time application to COVID-19. NBER Working
Paper No. 27560. Cambridge, MA: National Bureau of Economic Research. Retrieved March
8, 2021, from www.nber.org/papers/w27560.
de Vaan, M., Mumtaz, S., Nagaraj, A., & Srivastava, S. B. (2020). Social Influence in the COVID19
Pandemic: Community Establishments’ Closure Decisions Follow Those of Nearby Chain Estab-
lishments. Mimeo. Retrieved December 31, 2020, from www.hbs.edu/faculty/Pages/item.aspx?num=59464.
DellaVigna, S., & Gentzkow, M. (2019). Uniform Pricing in US Retail Chains. NBER Working
Paper No. 23996. Cambridge, MA: National Bureau of Economic Research. Retrieved March
8, 2021, from www.nber.org/papers/w23996.
Dillon, E., & Stanton, C. (2017). Self-Employment Dynamics and the Returns to Entrepreneurship.
NBER Working Paper No. 23168. Cambridge, MA: National Bureau of Economic Research.
Retrieved March 8, 2021, from www.nber.org/papers/w23168.
Dunn, A., Hood, K., Driessen, (2020). Measuring the Effects of the COVID-19 Pandemic on
Consumer Spending Using Card Transaction Data. Mimeo. Retrieved December 31, 2020, from
www.bea.gov/system/files/papers/BEA-WP2020-5_0.pdf.
Fairlie, R. (2020). The Impact of COVID-19 on Small Business Owners: The First Three Months af-
ter Social-Distancing Restrictions. NBER Working Paper No. 27462. Cambridge, MA: National
Bureau of Economic Research. Retrieved March 8, 2021, from www.nber.org/papers/w27462.
Forsythe, E., Kahn, L., Lange, F., & Wiczer, D. (2020). Labor Demand in the time of COVID-
19: Evidence from vacancy postings and UI claims. NBER Working Paper No. 27061. Cam-
bridge, MA: National Bureau of Economic Research. Retrieved December 31, 2020, from
www.nber.org/papers/w27061.
Glaeser, E. L., Gorback, C., & Redding, S. J. (2020). How Much Does Covid-19 Increase with
Mobility? Evidence from New York and Four Other U.S. Cities. NBER Working Paper No.
39
27519. Cambridge, MA: National Bureau of Economic Research. Retrieved November 12, 2020,
from www.nber.org/papers/w27519.
Glaeser, E. L., Jin, G. Z., Leyden, B. T., & Luca, M. (2021). Learning from Deregulation: The
Asymmetric Impact of Lockdown and Reopening on Risky Behavior During COVID�19. Jour-
nal of Regional Science.
Goolsbee, A., & Syverson, C. (2020). Fear, Lockdown, and Diversion: Comparing Drivers of Pan-
demic Economic Decline 2020. NBER Working Paper No. 27432. Cambridge, MA: National
Bureau of Economic Research. Retrieved December 12, 2020, from www.nber.org/papers/w27432.
Gupta, S., Nguyen, T. D., Rojas, F. L., Raman, S., Lee, B., Bento, A., Simon, K. I., & Wing, C.
(2020). Tracking Public and Private Responses to the COVID-19 Epidemic: Evidence from State
and Local Government Actions. NBER Working Paper No. 27027. Cambridge, MA: National
Bureau of Economic Research. Retrieved December 12, 2020, from www.nber.org/papers/w27027.
Hamilton, B. H. (2000). Does entrepreneurship pay? An empirical analysis of the returns to self-
employment. Journal of Political Economy, 108, 604–631. Hincapié, A. (2020). Entrepreneur-
ship Over the Life Cycle: Where are the Young Entrepreneurs? International Economic Review,
61, 617–681.
Hoff, P. D. (2009). A first course in Bayesian statistical methods. New York: Springer.
Huang, X., Li, Z., Jiang, Y., Li, X., & Porter, D. (2020). Twitter reveals human mobility dynamics
during the COVID-19 pandemic. PLoS ONE, 15.
Jaravel, X., & O’Connell, M. (2020). Real-time price indices: Inflation spike and falling product
variety during the Great Lockdown. Journal of Public Economics, 191, 104270.
Kurmann, A., Lale, E., & Ta, L. (2020). The Impact of COVID-19 on U.S. Employment and
Hours: Real-Time Estimates With Homebase Data. Mimeo. Retrieved March 17, 2021 from
www.lebow.drexel.edu/sites/default/files/1588687497-hbdraft0504.pdf.
Linka, K., Goriely, A. & Kuhl, E. (2021). Global and local mobility as a barometer for COVID-19
dynamics. Biomechanics and Modeling in Mechanobiology, 20, 651–669.
Mongey, S., & Weinberg, A. (2020). Characteristics of Workers in Low Work-From-Home and
High Personal-Proximity Occupations. Mimeo. Retrieved April 12, 2020 from bfi.uchicago.edu/working-
paper/characteristics-of-workers-in-low-work-from-home-and-high-personal-proximity-occupations/.
Moscarini, G., & Postel-Vinay, F. (2012). The contribution of large and small employers to job
creation in times of high and low unemployment. American Economic Review, 102, 2509–2539.
Papanikolaou, D., & Schmidt, L. D. (2020). Working Remotely and the Supply-side Impact of
Covid-19. NBER Working Paper No. 27330. Cambridge, MA: National Bureau of Economic
Research. Retrieved December 31, 2020, from www.nber.org/papers/w27330.
Sears, J., Villas-Boas, J. M., Villas-Boas, S., & Villas-Boas, V. (2020). Are We #StayingHome to
Flatten the Curve? Mimeo. Retrieved December 31, 2020, from www.medrxiv.org/content/10.1101/2020.05.23.
40
Sheridan, A., Andersen, A. L., Hansen, E. T., & Johannesen, N. (2020). Social distancing laws
cause only small losses of economic activity during the COVID-19 pandemic in Scandinavia.
Proceedings of the National Academy of Sciences of the United States of America, 117, 20468–
20473.
Soucy, J. P. R., Sturrock, S. L., Berry, I., Westwood, D. J., Daneman, N., MacFadden, D. R., &
Brown, K. A. (2020). Estimating effects of physical distancing on the COVID-19 pandemic us-
ing an urban mobility index. Mimeo. Retrieved March 8, 2021 from www.medrxiv.org/content/10.1101/2020.04
Storr, V. H., Haeffele-Balch, S., & Grube, L. E. (2016). Community revival in the wake of disaster:
Lessons in local entrepreneurship. Springer.
Velde, F. R. (2020). What Happened to the US Economy During the 1918 Influenza Pandemic? A
View Through High-Frequency Data. Mimeo. Retrieved December 31, 2020 from ssrn.com/abstract=3587634.
41
Tables and Figures
.8 Share High Proximity
Above Median Population Density
Share Online Sales/Services
Share Customers Over 65
Share Owners Over 65 0.65
.6 GOP Vote Share, County Level
0.58
0.50
0.46
.4 0.43 0.42
0.41
0.37
0.33
0.25 0.26 0.26 0.26
0.23 0.23 0.24
0.25
0.23 0.24 0.25 0.23 0.25 0.23
.2
0.21
0
Fully Open Partially Open Temporarily Closed Permanently Closed
Figure 1. Characteristics of Businesses by Operational Status as of May 9, 2020.
This figure plots characteristics of different businesses based on their industry characteristics, location char-
acteristics, or owner characteristics. Bars represent means and data are grouped by the operational status of
the business as reported in the May 9, 2020 survey. Please refer to Section 2 of the corresponding paper for
detailed definitions of each data source represented in this graph.
42
at 16 4 5 8 12 23 60
er
rl
ro
be 6 2 3 6 12 41 7
When will you reopen?
em t
us
pt ug
Se A
ly 11 5 8 17 45 13 9
Ju
e 13 10 16 43 13 9 9
Ju n
te
La
Ju 18 20 54 17 10 9 8
ne
rly
Ea
ay 17 52 11 8 6 3 6
M
te
La
ay 19 6 3 2 2 2 2
M
rly
Ea
ay ay June un e July A
t er
M eM eJ ugus or
at
rly La rly La l
Ea t Ea t be r
m
Se pt e
When Will Restrictions Be Lifted?
The percent of respondents in each cell is displayed, normalized within columns. Due to rounding, columns may not sum to 100.
Figure 2. Patterns in Regulation and Reopening at the Individual Business Level.
This figure displays when each business owner expects easing of legal restrictions around “fully reopening”
(x-axis) and the expected date when they will “fully reopen” (y-axis). The x-axis is derived from the question
“If there are legal restrictions on fully reopening your business, when do you expect them to be lifted?”.
Response possibilities ranged from “There are no legal restrictions.” to “September or later”. The y-axis
is derived from the question ”When will your business be fully open? Please provide your best guess.”
Responses possibilities ranged from “Early May” to “September or later”. Businesses that were fully open
were not asked the question and excluded from this figure. Numbers in each cell are the percent of responses
within each column.
43
Panel A: Share of Businesses Expecting to be Legally Able to Reopen
100%
Low Proximity
75% High Proximity
50%
25%
0%
May June July st
Augu
Panel B: Share of Businesses Expecting to be Fully Open
100%
Low Proximity
75%
High Proximity
50%
25%
0%
May June July August
Figure 3. Average Share of Businesses Projected to Be Fully Open in Each Industry by Date.
Panel A plots the average share of businesses fully open or projected to be fully open at future dates. Each line
represents a 4-digit NAICS code and is constructed using the cumulative distribution of individual responses
to the question “When will your business be fully open? Please provide your best guess.” Panel B plots the
average share of businesses legally able or expected to be legally able to reopen open at future dates. Fully
open businesses are included in both panels and are coded as open and legally able to open in the first period.
High proximity businesses, in yellow, are those above the median according to the proximity score. Green
indicates low proximity businesses.
44
100%
High Proximity
Share of Businesses Fully Open
75%
Low Proximity
50%
25%
0%
0 5 10 15 20
Weeks since Restrictions Lifted
Figure 4. Average Share of Businesses Reopening in Each Industry, Represented as Elapsed Weeks
after Restrictions are Lifted.
This figure plots the lag time in reopening between when respondents plan to reopen and when they are
legally allowed to do so. This is calculated as the difference between respondents’ projected reopening date
and their perceived date by which legal restrictions on operations will be lifted. Businesses that are fully
open are included at 0.
45
1.0
Conditional Grant
Unconditional Grant
0.8
Probability of Being Closed
0.6
0.4
0.2
0.0
0 10,000 20,000 30,000 40,000 50,000
Grant Size
Figure 5. Estimates of How Cash On Hand and Conditional Cash On Hand Change the Decision
to Remain Closed.
This figure plots how answers to a question about willingness to stay closed over the next 2 weeks changes as
a function of different hypothetical amounts of cash on hand. This is captured by “grant size” on the x-axis,
which comes from two parallel questions. Half of respondents (Unconditional Grant) were asked “Suppose
we could extend you a cash grant of [Grant Size]. Would you choose to open over the next two weeks?”
The other half of respondents (Conditional Grant) were asked “Suppose we could extend you a cash grant of
[Grant Size] but only on the condition that you remained closed for the next two weeks. Would you choose
to open over the next two weeks instead of taking the cash grant?” The sample for this figure comes from the
first wave of a panel survey of Alignable users conducted through Harvard Business School between May
20, 2020 and May 28, 2020 (N=780).
46
Panel A: Businesses that Are Fully Open in May 9 Survey
If these other businesses closed, would it affect your ability to remain open?
ers 36
ss C ustom
B usine
lose 27
rs O pen/C
t if Othe
van
Irrele
liers
25
Supp
rs 12
Cus tome
at Refer
ess th
Busin
rs] 1
Cus tome
m p e te for
o
ess [C
ar Busin 0 10 20 30 40
Simil
Panel B: Businesses that Are Partially Open or Temporarily Closed in May 9 Survey
Are you waiting on other businesses to open before fully opening yourself?
e 52
pen/Clos
if O t hers O
vant
Irrele ers 17
ustom
ess C
Busin
] 11
sto mers
or Cu
m pete f
s s [Co
ar B usine and]
9
Simil al Dem
[Sign
s iness
ar Bu
Simil mers
6
usto
tR efer C
e ss tha
Busin
liers
5
Supp
0 10 20 30 40 50
Percent of Respondents
Figure 6. Reopening Decisions as a Function of Other Businesses’ Actions.
This figure displays patterns of business dependency. Partially open or temporarily closed were asked “Are
you waiting on other businesses to open before fully opening yourself? (Select the category that matters
most.).” Fully open businesses were asked “Although you are currently open, if these other businesses closed,
would it affect your ability to remain open? (Select the category that matters most.)”.
47
Panel A: Months until Reopening Panel B: Months until Restrictions Lifted
1.7 1.3
1.6
Months Until Restrictions are Lifted
1.2
Months Until Reopening
1.5
1.4 1.1
1.3
1
1.2
1.1 .9
.2 .3 .4 .5 .6 .2 .3 .4 .5 .6
2016 County-Level GOP Vote Share 2016 County-Level GOP Vote Share
Panel C: Delay in Reopening Once Legal Panel D: Share Opening with No Lag
0.5 0.78
P(Open as Soon as Restrictions are Lifted)
0.76
0.4
Months Until Reopening
0.74
0.3 0.72
0.70
0.2
0.68
0.1 0.66
.2 .3 .4 .5 .6 .2 .3 .4 .5 .6
2016 County-Level GOP Vote Share 2016 County-Level GOP Vote Share
Figure 7. Effect of 2016 GOP Vote Share on Projected Time to Reopen and Time until Restrictions
Lifted.
The x-axis in every panel is the county-level GOP vote share in the 2016 Presidential election. Panel A plots
the projected months until the business reopens. Panel B plots the projected months until restrictions are
lifted. Panel C replicates Panel A, but nets out fixed effects for projected months until restrictions are lifted.
Panel D plots the share of respondents who selected the same period for projected reopening date and the
projected date by which restrictions will be lifted. All plots contain state and 4-digit NAICS fixed effects,
and control for population, population density and COVID cases (all control variables have been transformed
by the natural logarithm).
48
Table 1. Summary Statistics on Data from Survey and Additional Sources
Panel A: Data from Alignable Survey
Mean Std. Dev. 25th P’tile 75th P’tile min/max Obs.
Mo. until Reopen 1.33 1.49 0.00 1.75 0.00–4.50 29,305
Mo. until No Restrictions 1.06 1.42 0.00 1.75 0.00–4.50 28,763
Lag ≥ 4wks 0.18 0.38 0.00 0.00 0.00–1.00 28,538
Share Returning Customers 54.01 29.42 37.50 82.50 5.00–95.00 27,571
N. Employees (Jan, 2020) 10.63 32.77 1.00 7.00 0.00–500.00 20,505
Fully Open in May 9 Survey 0.32 0.46 0.00 1.00 0.00–1.00 33,356
Partially Open in May 9 Survey 0.34 0.47 0.00 1.00 0.00–1.00 33,356
Temporarily Closed in May 9 Survey 0.32 0.47 0.00 1.00 0.00–1.00 33,356
Permanently Closed in May 9 Survey 0.02 0.15 0.00 0.00 0.00–1.00 33,356
P(Open in December) 0.78 0.20 0.61 0.94 0.13–0.94 17,105
Panel B: Data from Additional Sources
Mean Std. Dev. 25th P’tile 75th P’tile min/max Obs.
COVID Cases per 1k 4.45 5.83 1.12 5.33 0.00–71.52 32,426
Emp. Physical Proximity 3.48 0.44 3.08 3.83 2.16–4.42 19,162
Likelihood Customers Over 65 24.26 13.91 12.50 30.50 5.00–87.50 22,856
Ease Operating Online 24.48 14.99 10.00 37.00 5.00–65.00 22,856
Essential Business (DE & MN) 0.53 0.50 0.00 1.00 0.00–1.00 22,883
GOP Vote Share (County) 0.44 0.16 0.33 0.55 0.04–0.90 33,117
Share Output → Intermed. Input 0.53 0.35 0.15 0.91 0.00–1.00 18,213
Share Business Buyers in Essential Ind. 0.55 0.30 0.33 0.73 0.00–1.00 20,995
Notes: Panel A presents summary statistics for survey responses. “Mo. until Reopen” and “Mo. until No
Restrictions” are the perceived months until the business will be fully open, and the perceived months until it is legal
to fully open, respectively. These figures are relative to the survey date of May 9. Responses were topcoded at
“September or Later”, which we top code at 4.5 months from early May. “N. Employees (Jan, 2020)” is the self
reported number of employees, including the respondent, in January 2020. The four indicator variables regarding
current status as of the May 9 survey correspond to the four options of the first question asked to respondents. For
this reason, these variables have the most observations. “P(Open in December)” is the numeric probability that a
businesses remains open in December, 2020. We code these probabilities from a multiple choice question shown to
respondents. This is the last question in the survey, which accounts for the fact that this variable has the fewest
responses.The text provides more detail about survey completion rates.
Panel B presents summary statistics for data taken from outside sources. “COVID Cases per cap.” is the county-level
number of COVID cases per capita. “Emp. Physical Proximity” is the the weighted average of a 5 point occupational
proximity scale over the industry-level (4-digit NAICS) distribution of occupations. “Likelihood Customers Over 65”
and “Ease Operating Online” are derived from MTurk answers at the 4-digit NAICS level. (See appendix for the
MTurk data collection tool.) “Essential Business (DE & MN)” is an indicator variable that indicates if a businesses
was considered essential in the guidelines made available in Deleware and Minnesota. “GOP Vote Share (County)” is
the share of votes for the Republican Presidential candidate in 2016. “Share Output → Intermed. Input” is derived
from the BEA 2012 Use table and is the share of total 3-digit industry output that used as intermediate inputs. “Share
Business Buyers in Essential Ind.” is derived from the same BEA series, as well as the “Essential Business (DE &
MN)” measure. This is the share of output that is used as an input by industries we identify as essential divided by the
total output that is used as intermediate inputs.
49
Table 2. Difference-in-Difference: Reopening and projected customers returning.
Percentage: Business Fully Open Customers Returning
After May 14 14.79∗∗∗ 14.83∗∗∗ 14.80∗∗∗ 6.756∗∗∗ 6.513∗∗∗ 6.768∗∗∗
(0.814) (0.821) (0.808) (1.019) (1.038) (1.030)
× Pooled WI & FL 5.552∗∗∗ 0.482∗∗
(0.441) (0.235)
× Just WI 3.114∗∗ 0.104
(1.138) (0.204)
× Just FL 6.092∗∗∗ 0.557∗∗
(0.137) (0.240)
R2 0.11 0.11 0.11 0.17 0.17 0.17
N 24,248 22,209 23,842 24,248 22,209 23,842
Note: In this table, we present difference in difference estimates of the effect of unexpected changes in the
regulatory environment on the share of businesses that are currently open and on demand projections in the future. In
column 1, we present difference in difference results with Wisconsin and Florida pooled together to form the
treatment group. In columns 2 and 3 we estimate the results separately for Wisconsin and Florida, respectively. To
avoid biasing these DiD estimates towards zero, Florida is excluded from the regression in 2 and Wisconsin is
excluded from column 3 to avoid contaminating the control group. We present analogous estimates of the effect of
the regulatory change on demand projections in columns 4 to 6.
We supplement the main survey data (collected from May 9 to May 13) with data from a later survey (collected from
May 14 to June 1); since the policy changes were announced on May 13 (WI) and 14 (FL), the main survey is the
pre-period, and the follow-up survey is the post-period. We reweight the post-period to match the pre-period at the
4-digit NAICS by county level, cells that do not include at least one observation in both surveys are dropped. Note
that while the second survey remained open until the following survey was distributed, over 95% of responses were
collected by May 18. Since the first survey was distributed starting on May 9th, the majority of responses are
collected over a nine day period in early May. See equation 3 and the discussion for more information.
50
Table 3. Factors contributing to differences in operational status
(1) (2) (3) (4)
Fully Open Partially Open Temp. Closed Perm. Closed
Emp. Physical Proximity -0.1050∗∗∗ -0.0156∗∗ 0.1170∗∗∗ 0.0036∗∗
(0.0046) (0.0059) (0.0048) (0.0016)
Owner Age 0.0009∗ -0.0009∗ -0.0004 0.0004∗∗
(0.0005) (0.0005) (0.0005) (0.0002)
Customers Over 65 -0.0079∗∗ 0.0128∗∗∗ -0.0029 -0.0020
(0.0032) (0.0047) (0.0031) (0.0012)
Essential Business 0.1188∗∗∗ 0.0433∗∗∗ -0.1581∗∗∗ -0.0040∗∗
(0.0067) (0.0093) (0.0093) (0.0019)
Ease Operating Online -0.0157∗∗∗ -0.0087∗ 0.0237∗∗∗ 0.0007
(0.0042) (0.0050) (0.0036) (0.0011)
ln(COVID cases per cap.) -0.0318∗∗∗ 0.0063 0.0256∗∗ -0.0001
(0.0090) (0.0040) (0.0100) (0.0008)
ln(Pop. Density) 0.0154∗∗ -0.0052 -0.0100∗ -0.0002
(0.0058) (0.0033) (0.0056) (0.0007)
GOP Vote Share (County) 0.3244∗∗∗ -0.0842∗∗∗ -0.2311∗∗∗ -0.0091∗
(0.0508) (0.0311) (0.0544) (0.0052)
DV Mean 0.316 0.340 0.321 0.022
DV SD 0.465 0.474 0.467 0.147
Residual SD 0.451 0.472 0.450 0.147
R2 0.0593 0.0060 0.0702 0.0024
N 33,236 33,236 33,236 33,236
Note: These columns correspond to answers to the question Are you currently open?. These options are collectively
exhaustive and mutually exclusive. Employee Physical Proximity, Customers Over 65, and Ease Operating Online
are converted to z-scores. Standard errors are in parentheses and clustered at the state level. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
51
Table 4. OLS: Contribution of various factors to the small business reopen decision
All Businesses Excluding Fully Open Businesses
(1) (2) (3) (4) (5) (6) (7) (8)
Reopen Restrictions Reopen Lag ≥ 4wk Reopen Restrictions Reopen Lag ≥ 4wk
ln(COVID cases per cap.) 0.0816∗∗∗ 0.0799∗∗∗ 0.0231∗∗∗ 0.0059∗ 0.0217∗∗ 0.0427∗∗ 0.0113 0.0001
(0.0207) (0.0258) (0.0080) (0.0033) (0.0106) (0.0197) (0.0101) (0.0044)
Emp. Physical Proximity 0.2433∗∗∗ 0.3011∗∗∗ 0.0365 0.0124 -0.0016 0.1109 -0.0353 -0.0078
(0.0540) (0.0688) (0.0273) (0.0088) (0.0492) (0.0740) (0.0340) (0.0096)
× ln(COVID cases p.c.) 0.0012 0.0005 0.0021 0.0005 0.0037 -0.0060 0.0039 0.0019
(0.0092) (0.0116) (0.0050) (0.0016) (0.0081) (0.0129) (0.0064) (0.0018)
Owner Age -0.0009 -0.0052∗∗∗ 0.0028 0.0006 0.0004 -0.0060∗∗∗ 0.0020 0.0003
(0.0024) (0.0015) (0.0022) (0.0006) (0.0033) (0.0018) (0.0029) (0.0008)
× ln(COVID cases p.c.) 0.0001 -0.0001 0.0002 0.0000 0.0000 -0.0002 0.0001 -0.0000
(0.0002) (0.0003) (0.0002) (0.0001) (0.0002) (0.0003) (0.0002) (0.0001)
Customers Over 65 -0.1098∗∗∗ -0.0804∗ -0.0547∗ -0.0099 -0.0800∗∗ -0.0437 -0.0789∗∗∗ -0.0171
(0.0318) (0.0432) (0.0290) (0.0116) (0.0368) (0.0625) (0.0276) (0.0156)
× ln(COVID cases p.c.) -0.0116∗ -0.0049 -0.0082 -0.0024 -0.0027 0.0066 -0.0078 -0.0026
(0.0058) (0.0073) (0.0051) (0.0019) (0.0063) (0.0106) (0.0048) (0.0027)
Essential Business -0.2995∗∗∗ -0.3244∗∗∗ -0.0727∗∗∗ -0.0268∗∗∗ -0.1204∗∗∗ -0.2264∗∗∗ -0.0480∗∗ -0.0186∗∗∗
(0.0257) (0.0278) (0.0161) (0.0044) (0.0291) (0.0293) (0.0220) (0.0055)
Ease Operating Online 0.0452∗∗∗ 0.0273∗∗∗ 0.0250∗∗ 0.0049 0.0215 0.0057 0.0256∗∗ 0.0033
(0.0129) (0.0098) (0.0099) (0.0034) (0.0138) (0.0126) (0.0107) (0.0036)
ln(Pop. Density) -0.0438∗∗∗ -0.0483∗∗∗ -0.0099 -0.0025 -0.0168 -0.0358∗∗ -0.0061 -0.0005
(0.0146) (0.0152) (0.0070) (0.0023) (0.0122) (0.0147) (0.0091) (0.0029)
GOP Vote Share (County) -1.3075∗∗∗ -1.0939∗∗∗ -0.5701∗∗∗ -0.1694∗∗∗ -0.9870∗∗∗ -0.8572∗∗∗ -0.5966∗∗∗ -0.1556∗∗∗
(0.1109) (0.1671) (0.0652) (0.0243) (0.0930) (0.1627) (0.0963) (0.0364)
Restriction Expectation FE No No Yes Yes No No Yes Yes
DV Mean 1.309 1.053 1.309 0.177 2.076 1.670 2.076 0.281
DV SD 1.489 1.409 1.489 0.382 1.386 1.456 1.386 0.449
Residual SD 1.452 1.369 1.119 0.375 1.373 1.439 1.162 0.379
R2 0.0479 0.0561 0.4352 0.0328 0.0192 0.0226 0.2979 0.2887
N 28,449 28,449 28,449 28,449 17,932 17,932 17,932 17,932
Note: Reopen is the expected months to reopen. Restriction is the estimated months until restrictions are lifted. Lag
≥ 4wk is a indicator variable that evaluates to 1 if the firm’s estimated reopening date is at least one month/four
weeks after the estimated date restrictions are lifted. Businesses that were permanently closed at the time of the
survey are excluded from these regressions; businesses that were fully open at the time of the survey are excluded
from columns 5 − 8. Employee Physical Proximity, Customers Over 65, and Ease Operating Online are converted to
z-scores. Restriction expectation fixed effects are included in the indicated models and are a vector of fixed effects
corresponding to the date at which respondents believe that restrictions will be lifted. Standard errors in parentheses,
clustered at county level. Note that the survey questions in which reopening and restriction beliefs are elicited is
mid-way through the survey, thus in some columns we are able to have more observations than we have complete
survey responses. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
52
Table 5. Contribution of Various Factors to Projected ln(Share Returning Customers)
(1) (2)
All Businesses Excluding Fully Open Businesses
Mo. until No Restrictions -0.1740∗∗∗ -0.1298∗∗∗
(0.0048) (0.0046)
ln(COVID cases per cap.) -0.0053 -0.0008
(0.0088) (0.0109)
Emp. Physical Proximity -0.0757∗∗∗ -0.0792∗∗∗
(0.0252) (0.0237)
× ln(COVID cases p.c.) -0.0110∗∗ -0.0167∗∗∗
(0.0044) (0.0044)
Owner Age -0.0005 -0.0011
(0.0017) (0.0017)
× ln(COVID cases p.c.) -0.0002 -0.0002
(0.0002) (0.0002)
Customers Over 65 0.0661∗∗∗ 0.0615∗∗
(0.0209) (0.0273)
× ln(COVID cases p.c.) 0.0074∗∗ 0.0047
(0.0035) (0.0044)
Essential Business 0.1233∗∗∗ 0.1115∗∗∗
(0.0130) (0.0185)
ln(Pop. Density) 0.0133 0.0159
(0.0080) (0.0109)
Ease Operating Online -0.0376∗∗∗ -0.0404∗∗∗
(0.0069) (0.0093)
GOP Vote Share (County) 0.4275∗∗∗ 0.4768∗∗∗
(0.0554) (0.0713)
DV Mean 3.727 3.558
DV SD 0.876 0.921
Residual SD 0.798 0.844
R2 0.1695 0.1609
N 27,185 18,157
Note: The outcome in all columns is the logarithm of projected demand, measured as the answer to the question “If
you are fully open by randomized date, what share of your customers do you expect at that time, compared to before
the crisis? Please provide your best guess”. Employee Physical Proximity, Customers Over 65, and Ease Operating
Online are converted to z-scores. Standard errors in parentheses, clustered at county level. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
53
Table 6. Instrumenting expected customer demand
Reopen Lag ≥ 4 weeks Open Dec.
(1) (2) (3) (4) (5) (6) (7) (8)
ln(Share Customers Returning) -0.876∗∗∗ -0.854∗∗∗ -0.528∗∗ -0.531∗∗ -0.163∗ -0.150∗ 0.310∗∗∗ 0.325∗∗∗
(0.187) (0.193) (0.216) (0.210) (0.083) (0.082) (0.041) (0.043)
Restriction Expectation FE No No Yes Yes Yes Yes Yes Yes
Additional Controls No Yes No Yes No Yes No Yes
Kleibergen-Paap F stat 8.4 9.6 8.4 9.5 8.4 9.5 8.7 9.7
Mean Dep. Var. 1.38 1.38 0.19 1.37 0.19 0.19 0.79 0.79
Std. Dev. Dep. Var. 1.49 1.49 0.39 1.49 0.39 0.39 0.18 0.18
R2 0.70 0.70 0.79 0.79 0.46 0.47 0.91 0.90
Observations 16,357 16,357 16,265 16,265 16,265 16,265 13,395 13,395
First Stage Reopen Lag ≥ 4 weeks Open Dec.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ ∗∗∗ ∗ ∗ ∗∗∗
ln(Signal) - ln(Prior) × Shown Info 0.189 0.187 -0.239 -0.236 -0.144 -0.146 -0.042 -0.039 0.080 0.083∗∗∗
(0.044) (0.041) (0.047) (0.043) (0.054) (0.048) (0.022) (0.021) (0.006) (0.005)
Restriction Expectation FE No No No No Yes Yes Yes Yes Yes Yes
Additional Controls No Yes No Yes No Yes No Yes No Yes
Mean Dep. Var. 3.73 3.73 1.38 1.38 1.37 1.37 0.19 0.19 0.79 0.79
Std. Dev. Dep. Var. 0.87 0.87 1.49 1.49 1.49 1.49 0.39 0.39 0.18 0.18
R2 0.34 0.34 0.52 0.52 0.63 0.63 0.39 0.39 0.10 0.11
Observations 16,357 16,357 16,357 16,357 16,265 16,265 16,265 16,265 13,395 13,395
Note: In Panel A, the dependent variable in Col. (1,2,3,4) Reopen is the expected months to reopen. The model in
Col. (3,4) includes a fixed effect for the date restrictions are lifted. In Col. (5,6) Lag ≥ 4wk is a indicator variable
that evaluates to 1 if the firm’s estimated reopening date is at least one month/four weeks after the estimated date
restrictions are lifted. In Col. (7,8) the dependent variable Open Dec. is the self-reported probability of being
operational by December 31st, 2020. Controls across all regressions include the prior and the gap between the signal
and the prior (log units), date fixed effects, the current status of business, and an indicator for classification as an
essential business. In Panel B, the dependent variable in Col. (1) is the log expected demand, the response to the
question “If you are fully open by randomized date, what share of your customers do you expect at that time,
compared to before the crisis? Please provide your best guess.” The instrument for expected demand is an
information instrument shown to a random subset of participants before we elicit demand expectations. See equations
6 and 7 and the related discussion for more information about the econometric specification. Columns with additional
controls contain the additional controls from the main OLS specification in Table 5, namely the natural logarithm of
(1+COVID cases per capita), physical proximity, owner age, likelihood of having customers over 65, ease of
conducting business online, the natural logarithm of population density, and the county-level share of the vote that
went to the Republican candidate in the 2016 presidential election. Restriction expectation fixed effects are included
in the indicated models and are a vector of fixed effects corresponding to the date at which respondents believe that
restrictions will be lifted. Standard errors are clustered at the region × business type level, which is the level at which
the information treatment is assigned. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
54
Table 7. Factors Contributing to the Probability of Being Open in December, 2020
(1) (2) (3)
Mo. until No Restrictions -0.0257∗∗∗ -0.0164∗∗∗
(0.0011) (0.0011)
ln(Share Returning Customers) 0.0637∗∗∗ 0.0567∗∗∗
(0.0023) (0.0024)
ln(COVID cases per cap.) 0.0013 -0.0002 0.0010
(0.0015) (0.0015) (0.0015)
Emp. Physical Proximity -0.0209∗∗∗ -0.0210∗∗∗ -0.0167∗∗∗
(0.0047) (0.0064) (0.0054)
× ln(COVID cases p.c.) -0.0016∗∗ -0.0009 -0.0010
(0.0008) (0.0010) (0.0009)
Owner Age 0.0001 0.0002 0.0001
(0.0004) (0.0004) (0.0004)
× ln(COVID cases p.c.) -0.0001∗ -0.0001 -0.0001
(0.0000) (0.0000) (0.0000)
Customers Over 65 0.0140∗∗∗ 0.0103∗∗ 0.0098∗∗
(0.0045) (0.0045) (0.0044)
× ln(COVID cases p.c.) 0.0007 0.0002 0.0002
(0.0008) (0.0008) (0.0008)
Essential Business 0.0211∗∗∗ 0.0171∗∗∗ 0.0126∗∗∗
(0.0029) (0.0024) (0.0027)
ln(Pop. Density) -0.0003 -0.0001 -0.0006
(0.0016) (0.0013) (0.0014)
Ease Operating Online -0.0010 0.0008 0.0011
(0.0013) (0.0013) (0.0013)
GOP Vote Share (County) 0.0162 0.0042 -0.0078
(0.0116) (0.0106) (0.0098)
DV Mean 0.792 0.792 0.792
DV SD 0.185 0.185 0.185
Residual SD 0.180 0.175 0.174
R2 0.0593 0.1085 0.1224
N 16,747 16,747 16,747
Note: The outcome in all columns is the answer to the question “What is the likelihood of your business remaining
operational by Dec. 31, 2020? Please provide your best guess.”. Businesses that were permanently closed at the time
of the survey are excluded from these regressions. Employee Physical Proximity, Customers Over 65, and Ease
Operating Online are converted to z-scores. Standard errors in parentheses, clustered at county level. ∗ p < 0.1, ∗∗
p < 0.05, ∗∗∗ p < 0.01
55
A. Appendix: Additional Figures and Tables
Panel A. Firm Size in the Survey and Census
Census
.6
Survey
Share of Total Firms
.4
.2
0
<5 5-9 19 99 -49
9
10- 20-
100
Panel B. Firm Locations in the Census and Survey
.15
Census Share
Survey Share
Share of Total Firms
.1
.05
0
rni
a ork ida xas noi
s nia sey Oh
io
org
ia ina
o wY Flo
r Te Illi lva Jer rol
alif e n s y w Ge hC
a
C N
Pen Ne o r t
N
Figure A1. Firm Size and Location in the Survey and Census.
Note: This figure plots the share of firms in each employment category and state for the 2017 Census of US
Businesses and the survey respondents for May 9, 2020. The sample size is 22,492 responses from May 9th survey
wave with non-missing employment data and 34,941 responses with non-missing state data.
56
Panel A: Local COVID Prevalence
0.5 Share of Small Business Open (May 9th)
Sarasota County, FL East North Central
East South Central
Mid Atlantic
0.4 Mountain
Maricopa County, AZ Clark County, NV New England
Pacific
South Atlantic
Orange County, CA
0.3 West North Central
West South Central
0.2 Middlesex County, MA
Westchester County, NY
Philadelphia County, PA
0.1
0.0
-7 -6 -5 -4 -3
ln(COVID Cases per Capita)
Panel B: Local Unemployment
0.5 Share of Small Business Open (May 9th)
Sarasota County, FL
0.4 Maricopa County, AZ Clark County, NV
Orange County, CA
0.3
East North Central
0.2 Middlesex County, MA East South Central
Philadelphia County, PA Mid Atlantic
Westchester County, NY Mountain
New England
0.1 Pacific
South Atlantic
West North Central
West South Central
0.0
.05 .1 .15 .2 .25 .3 .35
April Unemploment Rate
Figure A2. Share of Small Businesses Open by COVID Cases and Unemployment.
Note: This figure plots the share of firms that are fully open as of the May 9th survey wave against COVID cases per
capita, and the unemployment rate, at the county level. Counties with fewer than 100 observations not plotted; rings
representing counties with more responses are drawn larger than those representing counties with few responses.
57
Months until Restrictions are Lifted
1.52 − 1.98
1.36 − 1.52
1.11 − 1.36
0.96 − 1.11
0.83 − 0.96
0.49 − 0.83
Figure A3. Perceived Months Until Restrictions Lifted, by State.
Note: This map presents the average perceived months until restrictions lifted, by state.
58
Panel A: Shares of Businesses that are Temporarily Closed by Above and Below Median
Proximity
0.50
High Proximity Industries
Low Proximity Industries
Share of Businesses Temporarily Closed
0.40
0.30
0.20
0.10
0.00
April 1 May 9 June 1 July 1 July 25 August 29 September 26
Panel B: Shares of Businesses by Essential and Non-Essential Industry Classifications
Non-Essential Businesses
Essential Businesses
0.50
Share of Businesses Temporarily Closed
0.40
0.30
0.20
0.10
0.00
April 1 May 9 June 1 July 1 July 25 August 29 September 26
Figure A4. Shares of Businesses that are Temporarily Closed.
Note: This figure plots the share of firms that are temporarily closed across waves of Alignable’s data collection, split
by whether the business is in an above or below median proximity industry. Proximity is defined by the O-NET
Physical Proximity measure “To what extent does this job require the worker to perform job tasks in close physical
proximity to other people?” We merge the proximity measure to the OES data based on occupation and then take an
employment-weighted average by industry. We thank Simon Mongey and Alex Weinberg for publicly sharing this
measure. Industries are classified as essential if they are on the list of essential NAICS codes in both Delaware and
Minnesota, two states that have done this classification based on NAICS industries.
59
Table A1. Census industry versus survey industry breakdown.
Industry Census Percentage Survey Percentage
Agriculture, Forestry, Fishing and Hunting 0.4 1.1
Mining, Quarrying, and Oil and Gas Extraction 0.3 0.3
Utilities 0.1 0.3
Construction 11.7 7.6
Manufacturing 4.1 6.0
Wholesale and Retail Trade 15.7 13.1
Transportation and Warehousing 3.1 1.2
Information 1.3 2.1
Finance and Insurance 4.0 6.8
Real Estate and Rental and Leasing 5.2 8.8
Professional, Scientific, and Technical Services 13.5 14.6
Management of Companies and Enterprises 0.3 0.0
Administrative and Support and Waste Remediation Svcs 5.8 3.8
Educational Services 1.5 3.3
Health Care and Social Assistance 10.9 8.8
Arts, Entertainment, and Recreation 2.2 6.9
Accommodation and Food Services 9.0 5.2
Other Services (except Public Administration) 11.6 9.2
Note: This table reports results of Census and Survey shares by industry for firms with fewer than 500 employees.
Survey response shares are conditional on being able to classify industries, with unavailable or “Other” industry
classifications omitted from the denominator. We combine wholesale and retail trade.
60
Table A2. Expected demand by industry (NAICS 2-digit).
Industry May June July September N
Share Expecting >90% Mean >90% Mean >90% Mean >90% Mean
Arts, Entertainment, & Recreation 0.07 0.37 0.07 0.43 0.07 0.44 0.11 0.55 1851
Educational Services 0.08 0.47 0.07 0.47 0.11 0.53 0.13 0.60 832
Accommodation & Food Services 0.02 0.42 0.05 0.48 0.06 0.50 0.14 0.58 1452
Retail Trade 0.10 0.51 0.13 0.55 0.15 0.57 0.17 0.64 3057
Admin. and Waste Services 0.14 0.47 0.13 0.48 0.18 0.51 0.18 0.60 1085
Real Estate & Leasing 0.14 0.54 0.16 0.57 0.20 0.60 0.20 0.61 2162
Information 0.15 0.53 0.19 0.57 0.22 0.59 0.20 0.65 586
Manufacturing 0.12 0.56 0.14 0.60 0.13 0.59 0.21 0.67 1744
Health Care & Social Assistance 0.07 0.55 0.13 0.58 0.15 0.61 0.22 0.70 2331
Professional & Technical Services 0.18 0.54 0.18 0.58 0.19 0.62 0.24 0.67 3985
Other Services, Except Public Admin. 0.12 0.54 0.15 0.58 0.16 0.60 0.24 0.68 2322
Construction 0.17 0.59 0.18 0.60 0.19 0.63 0.25 0.69 1998
Finance and Insurance 0.19 0.66 0.28 0.66 0.26 0.65 0.27 0.70 1629
Note: This table reports answers to a question about the expected share of customers returning by a certain randomly
chosen date in the future. Each cell reports a share of customers relative to pre-COVID customers conditional on
being able to classify industries. Columns are the share of respondents who report having greater than 90% of
pre-COVID customers (the highest category) and then mean share of pre-COVID customers using the mid-point of
categorical answers.
61
Table A3. Tobit: Contribution of Various Factors to the Small Business Reopen Decision
All Businesses Excluding Fully Open Businesses
(1) (2) (3) (4) (5) (6) (7) (8)
Reopen Restrictions Reopen Lag ≥ 4wk Reopen Restrictions Reopen Lag ≥ 4wk
ln(COVID cases per cap.) 0.1772∗∗∗ 0.1665∗∗∗ 0.0533∗∗∗ 0.0059∗ 0.0249∗ 0.0449∗∗ 0.0149 0.0001
(0.0438) (0.0488) (0.0155) (0.0033) (0.0128) (0.0214) (0.0122) (0.0044)
Emp. Physical Proximity 0.4343∗∗∗ 0.5125∗∗∗ 0.0488 0.0124 -0.0126 0.1382 -0.0598 -0.0078
(0.0894) (0.0990) (0.0462) (0.0088) (0.0615) (0.0874) (0.0412) (0.0096)
× ln(COVID cases p.c.) -0.0161 -0.0142 -0.0060 0.0005 0.0032 -0.0041 0.0020 0.0019
(0.0145) (0.0166) (0.0082) (0.0016) (0.0099) (0.0152) (0.0076) (0.0018)
Owner Age -0.0020 -0.0077∗∗∗ 0.0052 0.0006 0.0008 -0.0068∗∗∗ 0.0026 0.0003
(0.0041) (0.0030) (0.0036) (0.0006) (0.0039) (0.0022) (0.0034) (0.0008)
× ln(COVID cases p.c.) 0.0002 -0.0001 0.0003 0.0000 0.0000 -0.0002 0.0000 -0.0000
(0.0004) (0.0004) (0.0003) (0.0001) (0.0003) (0.0004) (0.0003) (0.0001)
Customers Over 65 -0.1934∗∗∗ -0.1346∗∗ -0.0780 -0.0099 -0.1100∗∗ -0.0605 -0.1022∗∗∗ -0.0171
(0.0569) (0.0598) (0.0527) (0.0116) (0.0435) (0.0708) (0.0329) (0.0156)
× ln(COVID cases p.c.) -0.0227∗∗ -0.0115 -0.0143 -0.0024 -0.0051 0.0059 -0.0104∗ -0.0026
(0.0106) (0.0106) (0.0093) (0.0019) (0.0074) (0.0120) (0.0057) (0.0027)
Essential Business -0.6071∗∗∗ -0.6139∗∗∗ -0.1462∗∗∗ -0.0268∗∗∗ -0.1476∗∗∗ -0.2588∗∗∗ -0.0616∗∗ -0.0186∗∗∗
(0.0433) (0.0449) (0.0270) (0.0044) (0.0353) (0.0355) (0.0264) (0.0055)
Ease Operating Online 0.0965∗∗∗ 0.0660∗∗∗ 0.0497∗∗∗ 0.0049 0.0272∗ 0.0054 0.0333∗∗∗ 0.0033
(0.0264) (0.0203) (0.0190) (0.0034) (0.0165) (0.0151) (0.0124) (0.0036)
ln(Pop. Density) -0.0908∗∗∗ -0.0945∗∗∗ -0.0211 -0.0025 -0.0193 -0.0409∗∗ -0.0071 -0.0005
(0.0309) (0.0301) (0.0129) (0.0023) (0.0145) (0.0169) (0.0107) (0.0029)
GOP Vote Share (County) -2.4685∗∗∗ -2.0884∗∗∗ -1.0520∗∗∗ -0.1694∗∗∗ -1.1728∗∗∗ -0.9962∗∗∗ -0.7041∗∗∗ -0.1556∗∗∗
(0.2686) (0.3181) (0.1132) (0.0243) (0.1091) (0.1835) (0.1142) (0.0363)
Restriction Expectation FE No No Yes Yes No No Yes Yes
DV Mean 1.309 1.053 1.309 0.177 2.076 1.670 2.076 0.281
DV SD 1.489 1.409 1.489 0.382 1.386 1.456 1.386 0.449
Psuedo R2 0.0211 0.0251 0.1762 0.0413 0.0091 0.0097 0.1024 0.2783
N 28,449 28,449 28,449 28,449 17,932 17,932 17,932 17,932
Note: Reopen is the expected months to reopen. Restriction is the estimated months until restrictions are lifted. Lag
≥ 4wk is a indicator variable that evaluates to 1 if the firm’s estimated reopening date is at least one month/four
weeks after the estimated date restrictions are lifted. Businesses that were permanently closed at the time of the
survey are excluded from these regressions; businesses that were fully open at the time of the survey are excluded
from columns 5 − 8. Employee Physical Proximity, Customers Over 65, and Ease Operating Online are converted to
z-scores. Restriction expectation fixed effects are included in the indicated models and are a vector of fixed effects
corresponding to the date at which respondents believe that restrictions will be lifted. Standard errors in parentheses,
clustered at county level. Note that the survey questions in which reopening and restriction beliefs are elicited is
mid-way through the survey, thus in some columns we are able to have more observations than we have complete
survey responses. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
62
Table A4. Contribution of Various Factors to the Small Business Reopen Decision, County Fixed
Effects
All Businesses Excluding Fully Open Businesses
(1) (2) (3) (4) (5) (6) (7) (8)
Reopen Restrictions Reopen Lag ≥ 4wk Reopen Restrictions Reopen Lag ≥ 4wk
Emp. Physical Proximity 0.325∗∗∗ 0.416∗∗∗ 0.0373 -0.00218 0.0551 0.225∗∗ -0.0157 -0.0244
(0.0623) (0.0657) (0.0475) (0.0187) (0.0831) (0.104) (0.0697) (0.0214)
× ln(COVID cases p.c.) 0.0144 0.0189 0.00270 -0.00162 0.0120 0.0123 0.00636 -0.000739
(0.0110) (0.0114) (0.00838) (0.00313) (0.0135) (0.0181) (0.0123) (0.00378)
Owner Age -0.000389 -0.00716∗∗∗ 0.00464 0.00111 0.00189 -0.00823∗∗∗ 0.00433 0.000808
(0.00301) (0.00202) (0.00301) (0.000905) (0.00455) (0.00224) (0.00411) (0.00107)
× ln(COVID cases p.c.) 0.000220 -0.000306 0.000451 0.000116 0.000307 -0.000348 0.000398 0.0000769
(0.000328) (0.000415) (0.000364) (0.000119) (0.000400) (0.000493) (0.000441) (0.000135)
Customers Over 65 -0.169∗∗∗ -0.175∗∗∗ -0.0420 -0.00548 -0.112∗∗ -0.130 -0.0921∗ -0.0263
(0.0436) (0.0559) (0.0379) (0.0192) (0.0491) (0.0779) (0.0462) (0.0240)
× ln(COVID cases p.c.) -0.0209∗∗∗ -0.0196∗∗ -0.00639 -0.00172 -0.00779 -0.00707 -0.0102 -0.00423
(0.00759) (0.00912) (0.00629) (0.00312) (0.00862) (0.0129) (0.00807) (0.00414)
Essential Business -0.302∗∗∗ -0.329∗∗∗ -0.0703∗∗∗ -0.0263∗∗∗ -0.121∗∗∗ -0.235∗∗∗ -0.0471∗ -0.0183∗∗∗
(0.0286) (0.0302) (0.0169) (0.00482) (0.0330) (0.0311) (0.0239) (0.00611)
Ease Operating Online 0.0516∗∗∗ 0.0379∗∗∗ 0.0234∗∗ 0.00406 0.0288∗ 0.0201 0.0268∗∗ 0.00176
(0.0137) (0.00978) (0.0107) (0.00345) (0.0151) (0.0141) (0.0120) (0.00361)
Restriction Expectation FE No No Yes Yes No No Yes Yes
County FE Yes Yes Yes Yes Yes Yes Yes Yes
DV Mean 1.302 1.047 1.302 0.177 2.076 1.671 2.076 0.282
DV SD 1.486 1.406 1.486 0.382 1.386 1.455 1.386 0.450
Residual SD 1.414 1.323 1.092 0.366 1.331 1.387 1.127 0.366
R2 .0946 .115 .46 .0806 .0782 .091 .339 .337
N 27,266 27,266 27,266 27,266 16,890 16,890 16,890 16,890
Note: Reopen is the expected months to reopen. Restriction is the estimated months until restrictions are lifted. Lag
≥ 4wk is a indicator variable that evaluates to 1 if the firm’s estimated reopening date is at least one month/four
weeks after the estimated date restrictions are lifted. Businesses that were permanently closed at the time of the
survey are excluded from these regressions; businesses that were fully open at the time of the survey are excluded
from columns 5 − 8. Employee Physical Proximity, Customers Over 65, and Ease Operating Online are converted to
z-scores. Restriction expectation fixed effects are included in the indicated models and are a vector of fixed effects
corresponding to the date at which respondents believe that restrictions will be lifted. Standard errors in parentheses,
clustered at county level. Note that the survey questions in which reopening and restriction beliefs are elicited is
mid-way through the survey, thus in some columns we are able to have more observations than we have complete
survey responses. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
63
Table A5. Customers returning and business reopening: projected vs realized.
Panel A: Customers Returning Real Time: Customer Demand Report, Percent of Pre-Crisis Level
July August September
Demand Open Demand Open Demand Open
Projected Customers Returning if Fully-Open
< 10% 43.75 31.13 50.17 37.82 37.14 35.00
(3.326) (2.356) (5.033) (4.159) (6.225) (5.026)
10% − 25% 50.21 40.89 53.65 46.02 59.95 56.47
(2.798) (2.341) (4.682) (4.268) (4.754) (4.876)
25% − 50% 54.83 53.46 54.33 55.75 47.34 54.79
(1.998) (1.933) (3.428) (3.439) (3.683) (3.720)
50% − 75% 66.91 67.02 67.17 67.58 62.52 74.31
(1.639) (1.792) (3.044) (3.363) (3.184) (3.746)
75% − 90% 78.96 75.00 73.71 78.87 67.03 84.27
(1.824) (2.130) (3.190) (3.807) (3.803) (4.765)
> 90% 93.37 84.69 87.14 90.35 89.25 83.91
(1.867) (2.279) (3.327) (4.249) (3.855) (4.819)
R2 0.83 0.65 0.81 0.68 0.80 0.69
N 1,454 2,850 532 844 411 631
Panel B: Business Reopening Real Time: Percent Fully and Partially Open in the Last Month
Projected Customers Returning if Fully-Open July August September
Fully Partially Full. Part. Full. Part.
Projected Reopening Date
NA: Open at Time of Survey 90.78 98.05 92.38 98.68 89.18 96.10
(1.338) (0.964) (2.372) (1.733) (2.761) (1.904)
Early May 79.59 97.96 69.23 80.77 72 88
(4.223) (3.041) (8.085) (5.906) (8.391) (5.787)
Late May 64.83 94.07 66.67 97.10 75 96.88
(2.721) (1.960) (4.963) (3.625) (5.245) (3.617)
Early June 53.74 91.44 62.93 89.66 60.47 90.70
(2.162) (1.557) (3.828) (2.796) (4.524) (3.120)
Late June 42.86 86.07 49.37 82.28 55.56 93.65
(2.498) (1.799) (4.638) (3.388) (5.286) (3.645)
July 38.73 83.10 34.52 83.33 42.31 80.77
(2.481) (1.786) (4.498) (3.286) (5.818) (4.012)
August 29.56 76.73 23.81 71.43 35.48 77.42
(3.315) (2.387) (6.361) (4.647) (7.536) (5.197)
September or Later 25.06 70.20 30.16 75.40 24.05 75.95
(1.986) (1.430) (3.673) (2.683) (4.720) (3.255)
R2 0.71 0.90 0.73 0.90 0.73 0.91
N 2,850 2,850 844 844 631 631
Notes: In this table, we use subsequent Alignable surveys sent out at the end of July, August, and September to validate
demand expectations (Panel A) and reopening plans (Panel B). In Panel A, we plot the mean share of customers returning relative
to the pre-crisis level (Demand) and the share of businesses that are fully open (Open). We estimate these means conditional on the
respondent’s projected share of customers returning by a given date in the initial May survey. The realized demand is only available
for firms that report being fully open at the time of the later survey; therefore for each survey round we have more observations
for the business status (Open) than for the realized demand measure (Demand). Finally, in pair of columns, we restrict to the set
of respondents that appear in both the initial May survey, and the subsequent survey. In Panel B, we report the share of businesses
that are open at the time of the subsequent survey. We show both the share that are fully open (Fully) and the share that are either
fully open or partially open (Partially). Each row reports these means conditional on the respondent’s projected reopening date at
the time of the initial survey.
64
Table A6. Attrition Analysis
Full Survey Sample Re-Surveyed
(May, 2020) July August September
Number of Respondents 27,340 2,893 859 641
Share of Respondents Share p-value Share p-value Share p-value
Five Largest Industries
Professional and Technical Services 0.160 0.167 0.305 0.127 0.024 0.137 0.178
Retail Trade 0.129 0.128 0.874 0.155 0.060 0.152 0.141
Finance and Insurance 0.084 0.054 < 0.001 0.066 0.094 0.071 0.287
Health Care and Social Assistance 0.081 0.083 0.818 0.056 0.020 0.071 0.402
Real Estate and Rental and Leasing 0.074 0.053 < 0.001 0.049 0.018 0.044 0.014
Five Largest States
California 0.122 0.129 0.248 0.116 0.583 0.139 0.168
New York 0.071 0.077 0.190 0.078 0.429 0.067 0.684
Florida 0.066 0.072 0.202 0.070 0.614 0.072 0.527
Pennsylvania 0.050 0.042 0.031 0.041 0.217 0.038 0.147
Texas 0.050 0.051 0.807 0.047 0.683 0.041 0.289
Business Status
Fully Open 0.322 0.337 0.062 0.352 0.060 0.360 0.036
Partially Open 0.332 0.299 < 0.001 0.313 0.222 0.324 0.665
Temporarily Closed 0.319 0.348 < 0.001 0.318 0.939 0.300 0.285
Permanently Closed 0.026 0.015 < 0.001 0.017 0.094 0.016 0.082
(Predicted) Share of Customers Returning
< 10% 0.099 0.133 < 0.001 0.141 < 0.001 0.127 0.018
10% − 25% 0.122 0.135 0.027 0.134 0.279 0.135 0.319
25% − 50% 0.211 0.198 0.060 0.206 0.718 0.231 0.208
50% − 75% 0.252 0.230 0.005 0.216 0.014 0.228 0.170
75% − 90% 0.168 0.163 0.431 0.168 0.987 0.141 0.067
> 90% 0.148 0.142 0.330 0.135 0.274 0.138 0.458
Notes: In this table, we test for differential attrition across the subsequent survey waves. In the first column, we
present the share of respondents with a given attribute (industry, state) or response ( business status, predicted share
of customers returning) in the initial May survey. We report the share of businesses with each attribute or response in
the subsample of the original survey that can be matched to the later survey round. We also report the p-value of the
difference between the share of respondents in the subsequent round and in the initial May round. The last two
sections, business status and share of customers returning, are collectively exhaustive, though may not sum to one
due to rounding.
65
Table A7. Time to Reopen by Industry (2-digit NAICS), excluding Fully Open in May 9 Survey
Reopen Lag Lag ≥ 4 weeks
Management of Companies and Enterprises 0.750 0.500 0.000
Accommodation and Food Services 2.235 0.556 0.204
Real Estate and Rental and Leasing 1.928 0.670 0.275
Arts, Entertainment, and Recreation 2.654 0.673 0.241
Other Services, Except Public Administration 1.832 0.684 0.219
Construction 1.750 0.692 0.277
Retail Trade 1.755 0.706 0.228
Manufacturing 2.006 0.813 0.306
Public Administration 2.527 0.840 0.342
Health Care and Social Assistance 1.985 0.868 0.316
Educational Services 2.473 0.879 0.269
Agriculture, Forestry, Fishing and Hunting 2.034 0.912 0.311
Wholesale Trade 1.872 0.929 0.332
Utilities 1.938 0.969 0.348
Professional and Technical Services 2.158 0.992 0.355
Administrative and Waste Services 2.388 1.012 0.337
Information 2.698 1.049 0.383
Finance and Insurance 1.786 1.086 0.278
Transportation and Warehousing 2.371 1.113 0.416
Mining, Quarrying, and Oil and Gas Extraction 1.884 1.304 0.308
Note: This table presents average time to reopen by 2-digit NAICS code for the subset of firms that are not fully open
at the time of the survey. Reopen is the predicted time to reopen in months, Lag is the difference between the
predicted time to reopen and the predicted end of restrictions, and Lag ≥ 4 weeks is an indicator that evaluates to one
when the predicted reopening time is more than four weeks after the predicted end of restrictions.
66
Table A8. Time to Reopen by Industry (3-digit NAICS), excluding Fully Open in May 9 Survey
Reopen Lag Lag ≥ 4 weeks
Accommodation and Food Services: Accommodation 2.129 0.700 0.300
Accommodation and Food Services: Food Services and Drinking Places 2.270 0.509 0.172
Administrative and Waste Services: Administrative and Support Services 2.400 1.009 0.339
Agriculture, Forestry, Fishing and Hunting: Animal Production and Aquaculture 2.154 1.023 0.314
Arts, Entertainment, and Recreation: Amusement, Gambling, and Recreation Industries 2.023 0.461 0.173
Arts, Entertainment, and Recreation: Museums, Historical Sites, and Similar Institutions 2.238 0.844 0.214
Arts, Entertainment, and Recreation: Performing Arts, Spectator Sports, and Related Industries 3.183 0.815 0.299
Construction: Construction of Buildings 1.784 0.678 0.265
Construction: Specialty Trade Contractors 1.705 0.687 0.297
Educational Services: Educational Services 2.480 0.846 0.270
Finance and Insurance: Credit Intermediation and Related Activities 1.771 1.187 0.246
Finance and Insurance: Funds, Trusts, and Other Financial Vehicles 1.964 1.455 0.268
Finance and Insurance: Insurance Carriers and Related Activities 1.810 0.866 0.331
Finance and Insurance: Securities, Commodity Contracts, and Other Financial Investments and Related Activities 1.610 0.794 0.310
Health Care and Social Assistance: Ambulatory Health Care Services 1.920 0.887 0.325
Health Care and Social Assistance: Social Assistance 2.233 0.692 0.244
Information: Motion Picture and Sound Recording Industries 3.093 0.844 0.327
Information: Publishing Industries (except Internet) 2.455 1.455 0.519
Manufacturing: Beverage and Tobacco Product Manufacturing 2.075 0.488 0.163
Manufacturing: Food Manufacturing 2.353 1.176 0.344
Manufacturing: Machinery Manufacturing 1.823 0.746 0.328
Manufacturing: Miscellaneous Manufacturing 2.051 0.720 0.265
Manufacturing: Printing and Related Support Activities 1.833 0.654 0.330
Other Services, Except Public Administration: Personal and Laundry Services 1.696 0.532 0.171
Other Services, Except Public Administration: Religious, Grantmaking, Civic, Professional, and Similar Organizations 2.340 1.010 0.340
Other Services, Except Public Administration: Repair and Maintenance 1.546 0.819 0.252
Professional and Technical Services: Professional, Scientific, and Technical Services 2.159 0.994 0.356
Public Administration: Administration of Human Resource Programs 2.882 0.779 0.309
Real Estate and Rental and Leasing: Real Estate 1.916 0.648 0.278
Real Estate and Rental and Leasing: Rental and Leasing Services 2.041 0.912 0.250
Retail Trade: Clothing and Clothing Accessories Stores 1.576 0.507 0.136
Retail Trade: Food and Beverage Stores 1.911 0.998 0.270
Retail Trade: Furniture and Home Furnishings Stores 1.471 0.695 0.224
Retail Trade: Health and Personal Care Stores 1.776 0.595 0.242
Retail Trade: Miscellaneous Store Retailers 1.782 0.628 0.225
Retail Trade: Motor Vehicle and Parts Dealers 1.378 0.448 0.120
Retail Trade: Sporting Goods, Hobby, Musical Instrument, and Book Stores 2.014 0.797 0.269
Transportation and Warehousing: Transit and Ground Passenger Transportation 2.647 1.016 0.381
Wholesale Trade: Merchant Wholesalers, Durable Goods 1.767 0.822 0.276
Note: This table presents average time to reopen by 3-digit NAICS code for the subset of firms that are not fully open
at the time of the survey. Reopen is the predicted time to reopen in months, Lag is the difference between the
predicted time to reopen and the predicted end of restrictions, and Lag ≥ 4 weeks is an indicator that evaluates to one
when the predicted reopening time is more than four weeks after the predicted end of restrictions.
67
File and source
- File
- business-reopening-decisions-and-demand-forecasts-during-the-covid-19-pandemic.pdf
- Size
- 1,184,139 bytes
- SHA-256
- 26b2eae3bc16bff66aa41477132761cbf6c0f7aa2bf2526c7d002b17bdc2cd09
- Original
- www.nber.org