Buffington et al., Small Business Pulse Survey Estimates by Owner Characteristics and Rural/Urban Designation (2021-09)

Summary

A U.S. Census Bureau Center for Economic Studies working paper, CES 21-24, dated September 2021, Small Business Pulse Survey Estimates by Owner Characteristics and Rural/Urban Designation, by Catherine Buffington, Daniel Chapman, Emin Dinlersoz, Lucia Foster, James Hunt and Shawn Klimek. The abstract says that, at policymakers' request for more context on the survey's measures of COVID-19's impact on small businesses, the authors developed estimates by owner characteristics and by rural and urban location. They use geographic coding on the Business Register to estimate the pandemic's effects by urban and rural designation. A technical appendix explains margins of error and 90% confidence intervals used to test whether rural and urban estimates differ.

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

Text recognized from a scanned document; may contain errors. Pages 40, 44, 46, 48, 49, 52–55, 63, 64, 69–71, 74, 75, 77, 79–82, 85–90, 93–98, 101, 102, 104–106, 109–111, 113, 114, 117, 120, 122, 123, 128, 129, and 137 of 141 are not shown because the text could not be recognized reliably.

Small Business Pulse Survey Estimates by Owner Characteristics and
Rural/Urban Designation

by

Catherine Buffington
U.S. Census Bureau

Daniel Chapman
U.S. Census Bureau and University of Maryland

Emin Dinlersoz
U.S. Census Bureau

Lucia Foster
U.S. Census Bureau

James Hunt
U.S. Census Bureau

Shawn Klimek
U.S. Census Bureau

CES 21-24 September 2021

The research program of the Center for Economic Studies (CES) produces a wide range of
economic analyses to improve the statistical programs of the U.S. Census Bureau. Many of
these analyses take the form of CES research papers. The papers have not undergone the
review accorded Census Bureau publications and no endorsement should be inferred. Any
opinions and conclusions expressed herein are those of the author(s) and do not necessarily
represent the views of the U.S. Census Bureau. All results have been reviewed to ensure that no
confidential information is disclosed. Republication in whole or part must be cleared with the
authors.

To obtain information about the series, see www.census.gov/ces or contact Christopher Goetz,
Editor, Discussion Papers, U.S. Census Bureau, Center for Economic Studies 5KO38E, 4600 Silver
Hill Road, Washington, DC 20233, CES.Working.Papers@census.gov. To subscribe to the series,
please click here.
Abstract

In response to requests from policymakers for additional context for Small Business Pulse
Survey (SBPS) measures of the impact of COVID-19 on small businesses, we researched
developing estimates by owner characteristics and rural/urban locations. Leveraging geographic
coding on the Business Register, we create estimates of the effect of the pandemic on small
businesses by urban and rural designations. A more challenging exercise entails linking micro-
level data from the SBPS with ownership data from the Annual Business Survey (ABS) to create
estimates of the effect of the pandemic on small businesses by owner race, sex, ethnicity, and
veteran status. Given important differences in survey design and concerns about nonresponse
bias, we face significant challenges in producing estimates for owner demographics. We discuss
our attempts to meet these challenges and provide discussion about caution that must be used
in interpreting the results. The estimates produced for this paper are available for download.
Reflecting the Census Bureau’s commitment to scientific inquiry and transparency, the micro
data from the SBPS will be available to qualified researchers on approved projects in the
Federal Statistical Research Data Center network.

* Any opinions and conclusions expressed herein are those of the authors and do not represent the views of the
U.S. Census Bureau. All results have been reviewed to ensure that no confidential information is disclosed. The
Census Bureau has reviewed this data product for unauthorized disclosure of confidential information and has
approved the disclosure avoidance practices applied to this release (Approval ID: CBDRB-FY21-026, CBDRB-FY21-
073, CBDRB-FY21-291, and CBDRB-FY21-ESMD006-013). The development, testing, fielding, and dissemination of
the SBPS represents the work of about 60 Census Bureau staff. We thank David Brown, Carol Caldwell, John
Eltinge, John Haltiwanger, Nick Orsini, Michael Ratcliffe, and Jenny Thompson for helpful comments. We thank
Luke Streng for his assistance in preparing the research estimates for release.
1. Introduction

Early in the COVID-19 pandemic, the Census Bureau introduced new data products
designed to provide timely information about its impact on the U.S. economy and population
(Buffington et al. 2021b). One of these new data products is the Small Business Pulse Survey
(SBPS) which publishes weekly, timely information on metrics including revenue and
employment changes and business sentiment and expectations for small, single-location
employer businesses.” The ability of the SBPS to provide weekly insights on the differential
impacts of the pandemic on small businesses by sector and geography garnered much interest
among policymakers and researchers. These stakeholders, which included staff from both the
Executive and Legislative branches, were also interested in learning more about differential
impacts on businesses based on the demographics of business owners and whether the business is
located in urban or rural areas. This paper describes our attempt to respond to these stakeholder
requests in a timely fashion by linking the underlying micro-level SBPS data to other Census
Bureau data products to provide information on business owners and on urban versus rural
locations. We discuss the many challenges in producing this information, provide cautions in
interpreting the results, and discuss future research that could improve upon our approach.

To put the SBPS target population in context, it is helpful to consider the magnitudes of
businesses contained in the Census Bureau’s Business Register (BR). According to the Census
Bureau’s website, the BR “covers more than 160,000 multi-establishment companies,
representing 1.8 million affiliated establishments, 5 million single establishment companies, and
nearly 21 million nonemployer businesses.”? The SBPS target population excludes multi-
establishment companies and nonemployer businesses. The exclusion of nonemployer businesses
is especially relevant when we turn our attention to owner characteristics. Women-owned and
Black-owned businesses are more concentrated in nonemployer than employer businesses. In
2017, Black-owned businesses comprised 2.2% of employer businesses but 11.7% of

nonemployer businesses, and women-owned businesses comprised 19.8% of employer

2 We use small business and establishment here interchangeably as the target population of the SBPS is constrained
to single location businesses.
> https://www.census.gov/econ/overview/mu0600.html.
businesses but 41.7% of nonemployer businesses.* Thus, we need to keep in mind that the
“typical” business of these demographic groups is outside the target population of our study.

The SBPS is conducted in nine-week phases and, as of this writing, there have been six
phases.° This paper focuses on the first three phases of the survey because the bulk of the
empirical analyses were carried out while phase 4 was still underway (future work could repeat
the analysis in this paper for subsequent phases). The survey generally has about 20 questions
with core content appearing in all phases and rotating content as needs develop. The core content
provides information on business sentiment, business expectations, supply chain disruptions,
operations (changes in revenue, employment, and hours), and receipt of federal financial
assistance. Rotating content includes questions such as about changes in telework for employees
or changes in processes and delivery modes.

We highlight a few findings from Buffington et al. (2021a) for Phases 1-3 of the SBPS to
provide context and motivation for our empirical exercises.° Starting with temporal variation,
they find that business sentiment, business expectations, and operational changes reveal large
negative impacts on small businesses over all three phases but with some improvement in Phase
1, stalling in Phase 2, and worsening in Phase 3. They also find that shares of businesses
experiencing revenue changes are much larger than shares of businesses experiencing
employment changes suggesting that there may be some labor hoarding. In addition, shares of
businesses adjusting hours is larger than that adjusting employment (but smaller than that
adjusting revenue).

While all sectors of the economy are negatively impacted, Buffington et al. (2021a)
highlight that some sectors were more negatively impacted than others as measured over
multiple metrics. They find that these differences are persistent over time and that dispersion
increased over time. They focus on two sectors, Accommodation and Food Services and Finance
and Insurance, to illustrate characteristics of sectoral variation. Relative to small businesses in
Finance and Insurance, a higher percentage of small businesses in Accommodation and Food
Services face a large negative impact from the pandemic. For example, by the end of the third

phase (early January 2021), 64.9% of businesses in Accommodation and Food Services report a

4 Source: Annual Business Survey 2018 and Nonemployer Demographic Statistics; see Tables 2 and 4.

> The timing of the phases is as follows: (1) April-June 2020; (2) August-October 2020; (3) November 2020-January
2021; (4) February-April 2021; (5) May-July 2021 (6) August-October 2021.

® We focus on the first three phases of the SBPS in this paper.

3
large negative effect as compared to 13.7% of business in Finance and Insurance. Similarly, in
this last week of Phase 3, 56.1% of businesses in Accommodation and Food Services saw a
decrease in revenue as compared to 22.1% in Finance and Insurance.

In terms of the geographic impact, Buffington et al. (2021a) find that state-level variation
in overall impact is not as persistent as sectoral-level variation nor as large. They conclude that
it may be that state level is not always the most appropriate level for this type of analysis since
the presence of a large aggregate shock may overwhelm local shocks.’ Moreover, local shocks
may be specific to a county or metropolitan statistical area (MSA) and thus may not show up in
the state data. When there are state-level shocks, we see state-level variation in SBPS measures
that correlate with these state-level shocks (a good example is the power outages in Texas in
early 2021).°

As the results from SBPS were published, several stakeholders reached out to the Census
Bureau requesting supplemental information on the demographic characteristics of business
owners and rural versus urban locations. Recognizing the importance of providing timely data
during a pandemic, we began research into how this information could be incorporated with
SBPS results. Our goals are to leverage Census Bureau expertise to ensure that appropriate care
is taken with respect to methodology, produce summary data by linking SBPS micro-level data
to other Census Bureau micro-level data, and release the resulting product with detailed
documentation on the limitations of these estimates. More generally, this research supports
Census Bureau efforts to leverage existing data holdings to produce new products without
imposing additional respondent burden.

This paper provides a detailed examination of the challenges inherent in this exercise and
our attempt to provide timely information with transparency and robust discussion of caveats to
enable informed use of the supplemental information. Much of the paper provides detailed
information on the challenges in linking SBPS micro-level data with other Census Bureau assets

in order to provide this information. Given these challenges, we note the importance of future

’ This is not to say that variation in state policies related to the pandemic did not play a role in the dispersion
observed across states. However, if the sectors that are hardest hit by the pandemic (e.g. Accommodation and Food
Services) account for similar shares of economic activity across states, the sectoral variation could be more
prominent than the variation across states.

8 As one such example, while the national average for small businesses with temporary closures was less than 3%,
the average for Texas rose from 2.3% in early January to 15.1% by the end of February before falling back down to
1.2% by mid-March. Note that this example occurred during phase 4 of the SBPS which is not considered in the
remainder of this paper.
research into issues concerning sample selection for one of our main datasets. We show some
results from this linking and also conduct some empirical exercises using micro-level data that
provide greater context to the new data product by allowing us to control for characteristics
through regression analysis.

The paper proceeds as follows. Section 2 provides a short discussion of related
measurement efforts. Sections 3-5 discuss each of the three datasets that we use and how we link
them together. Section 6 presents results from the research public data product. Section 7
presents results from empirical exercises using micro-level data. Section 8 provides concluding

remarks including discussion of potential future directions.

2. Related Measurement Efforts

We highlight some related efforts to measure the impact of the COVID-19 pandemic on
businesses by business owner demographics and location. Starting with surveys of small
businesses, McKinsey conducted a “COVID-19 US Small and Midsize Business Financial Pulse
Survey” in early May 2020. With about 1,000 respondents, they found a higher percentage of
minority-owned businesses had difficulties accessing credit (42% as compared to 29% for all
respondents) and were “extremely” or “very concerned” about the financial viability of their
business (58% as compared to 47% for all respondents). A US Chamber of Commerce survey of
500 small businesses conducted with MetLife in July 2020 found minority-owned businesses had
a harder time securing loans for aid: 13% of minority-owned businesses reported troubles
compared to 8% of non-minority-owned businesses. Two-thirds of minority-owned businesses
are concerned about having to permanently close, whereas 57% of non-minority-owned
businesses are. In another report based on the same survey, the proportion of male-owned
businesses reporting the overall health of their business as good declined from 67% to 62%
before and after the coronavirus pandemic hit, while for women-owned businesses that

proportion declined more steeply from 60% to 47%’. Alekseev et al. (2020) utilize a sample of

2 See MetLife & U.S. Chamber of Commerce Special Report on Race and Inequality on Main Street | U.S. Chamber
of Commerce and Special Report on Women-Owned Small Businesses During COVID-19 | U.S. Chamber of
Commerce
66,000 respondents from an online Facebook survey sent between April 20 — 28, 2020. In the
sample, 47.4% of majority-female firms decreased the size of their workforce, while 40.6% of
majority-male firms did so. Respondents from majority-female firms were more likely to also
report difficulties with tasks associated with their household, such as taking care of other

household members.

Fairlie (2020) compares the February and April 2020 data releases from the Current
Population Survey (CPS) to assess the impact of COVID-19 on small business operation. The
number of active small business owners declined from 15 million in February to 11.7 million
people in April, a 22 percent decline. Looking deeper into this decline, the number of black
business owners fell by 41 percent, the number of Latino business owners declined by 32
percent, and the number of Asian business owners declined by 26 percent, all greater than
average declines. Mills and Battisto (2020) combined data from the 2019 Small Business Credit
Survey (SBCS), the 2012 Census Survey of Business Owners, and SBA Paycheck Protection
Program Loan Level Data to assess the role of geography in COVID impacts; they found that
Black-owned businesses are more likely to be concentrated in COVID-19 hotspots and less

likely to be covered by Paycheck Protection Program (PPP) loans.

Liu and Parilla (2020), in a report from the Brookings Institution, further detail
underlying financial issues that have faced minority-owned businesses, finding that in majority-
Black neighborhoods, 80% of loans to nonemployer small businesses were issued by financial
technology or online lending companies. Liu and Parilla (2020) also find that issues exist in
getting PPP loan coverage out to Black neighborhoods, noting that small businesses in majority-
Black neighborhoods in Washington DC waited on average 37 days for PPP assistance, 10 days
more than small businesses in majority-white neighborhoods. Fairlie and Fossen (2021) use
administrative data for PPP and Economic Injury Disaster Loan (EIDL) and find mixed results
with respect to the relationship between minority-owned businesses and receipt of COVID-19
financial aid. Their analysis shows that PPP funds flowed to majority-minority areas of the U.S.
at a slower than average rate, but that EIDL funds and receipt were more strongly positively
associated with minority communities. The 2021 Report on Employer Firms, published by the
Federal Reserve System as part of its Small Business Credit Survey, finds differential impacts by

race on business outlook, with 79% and 77% of Asian and Black-owned firms, respectively,
describing their financial outlook as “fair” or “poor”, compared to 57% of all survey participants.
Black- and Asian-owned businesses also differ in their primary challenge facing their businesses,
with Asian-owned firms most likely to report “weak demand for products and services” and

Black-owned firms most likely to report “credit availability” as a challenge.

These papers have used different sampling frames and methods to assess the economic
impacts of the COVID-19 pandemic. In this paper we attempt to complement these efforts by
combining existing Census Bureau data products with the Small Business Pulse Survey in order
to provide more detail about urban and rural small businesses and small businesses by
demographics of the owners (race, ethnicity, sex, and veteran status). By providing detailed
information about the data and our methodology, we allow users to determine the fitness for use

of our estimates for their needs.

3. Small Business Pulse Survey Data

The SBPS target population is all single-location businesses with 1-499 employees and
$1,000 or more in revenue; all active businesses that reported an email address in the 2017
Economic Census comprise the full SBPS collection set. About a million businesses meet this
requirement. The sample is divided equally across the nine weeks of each phase of the survey, so
that about 100,000 businesses are asked to participate in the survey each week.'° The response
rate is about 25% leading to about 25,000 businesses per week which gives us a total of about
621,000 responses across the three nine-week survey phases. Respondents are eligible to
participate in each phase, so the total number of responses includes some small businesses that
appear in our response set up to three times. Overall, we have about 350,000 unique businesses
in our sample and 621,000 total responses. Breaking this down by number of appearances:
165,000 businesses responded once, 102,000 responded twice, and 84,000 responded three times.

The survey is designed to be representative'! and the results are weighted’ to be representative

'0 Subsamples are not reused within a phase but are reused across phases so some businesses may respond in
multiple phases.

™ Roughly 1.7 million single-establishment employer businesses were in scope to receive the 2017 Economic
Census, and of those approximately 1.1 million had between 1-499 employees and reported a valid email address.
These comprise the SBPS target population and are representative of over 6 million businesses.

2 Though the SBPS is not a sample survey, weights are used to reflect the full population. These weights are
adjusted weekly to correct for sample non-response.
at the national and state by sector levels. Note that published subsector and MSA estimates as
well as these new estimates by owner characteristics and location described in this paper are not
used in the creation of weights or their adjustment and therefore these estimates should be

viewed with appropriate caution.

With a response rate of 25%, we are naturally concerned about non-response bias when
interpreting results from the SBPS. Moreover, we are also concerned with survival bias as the
survey extends over time (and especially as it extends over time during a pandemic-induced
recession). Buffington et al. (2021a) discuss many of these issues, but to these we add concerns
that the non-response bias may be correlated with some of the characteristics that we study in
this paper.

The SBPS starts by asking the respondent to confirm Employer Identification Number
(EIN); if the pre-printed EIN is incorrect, the respondent is asked to provide the correct EIN. As
will be evident when we discuss linking the SBPS to other data sources, the EIN is critical to our
ability to leverage existing Census Bureau information on the business. We now provide some
detail on the five questions comprising the core content: business sentiment; changes in revenue,
employment, and hours; and business expectations.

Business sentiment is collected through the question: “Overall, how has this business
been affected by the Coronavirus pandemic?” with checkbox responses: large negative, negative,
no effect, positive, and large positive. The question allows the business to provide a holistic
assessment of the impact on their business, without constraining the response to a specific time
frame since the start of the pandemic. We also focus on three questions which share the same
basic format for three concepts: revenue, employment, and employee hours. For example, the
revenue question: “In the last week, did this business have a change in operating
revenues/sales/receipts, not including any financial assistance or loans?” with checkbox
responses: “Yes, increased,” “Yes, decreased,” and “No.” These questions refer to a specific time
frame (prior week) for the response. The SBPS also asks respondents for their expectations for
their business: “In your opinion, how much time do you think will pass before this business
returns to its normal level of operations relative to one year ago?” The responses include little/no

effect, already back to normal, to varying durations (1 month or less up to more than 6 months),
to never and closed.'? The response is not constrained to a specific time frame, but the
comparison is always relative to the previous year, a moving window which allows expectations

to evolve over time.

With access to the micro data, it is possible to learn more about the relationships between
these five concepts. We would expect there to be some strong positive relationships between the
overall impact, changes in revenue, and expectations. We find that overall impact and revenue
changes are positively correlated '* (0.262). The fact that the correlation is not higher partially
reflects the difference in timing of the two question (recall overall impact has an open-ended
time frame, while change in revenue concerns the last week). It also suggests that businesses are
considering other factors in assessing the overall impact on the business (for example, whether
the business can remain open and retain employees). We also find that overall impact and
expectations are positively correlated (0.523). This correlation is higher than for revenue,
suggesting that these additional factors that are used to assess overall impact are also important
in developing expectations. Turning to the relationship between employment and hours, perhaps
not surprisingly, the changes in employment and hours are positively correlated (0.499). As
noted earlier, Buffington et al. (2021) find that the share of businesses changing hours generally
exceeds the share of businesses changing employment (and both are concentrated on negative

changes).

It is less clear a priori what the relationship between changes in revenue and changes in
employment (and hours) would be. For example, employment and hours could be decreasing
(even in the face of rising revenue) due to increasing labor productivity (possibly by relying
more heavily on technology). On the other hand, we might see revenue decreasing and
employment and hours holding steady, possibly reflecting labor hoarding. Given the publicly
available results, we know that most changes are for revenue declines and that shares in
businesses with declines in hours follow this pattern (as do changes in employment to a lesser
extent). Thus, it is not surprising that the correlation between revenue changes and changes in

employment and hours are positive and that the correlation is higher for hours (0.379 versus

'3 While the question appears in some form in all phases, changes were made to the question and responses over
time. In the fourth week of Phase 1, the question adds “relative to one year ago” and “usual” is replaced by
“normal.” At the start of Phase 2, “This business has returned to its normal level of operations” and “This business
has permanently closed” are added as responses (see Buffington et al. 2021a for more discussion).

'4 Correlations are weighted using the SBPS tabulation weights.

9
0.254). Finally, the smallest correlation in this set is between employment and expectations
(0.125), suggesting that there is a wedge of some sort between changes in employment and
expectations for when their business will get back to normal. Unfortunately, we do not have a

“normal” time period to compare to these results. '°

We dig a little deeper into the responses for changes in revenue and changes in
employment to understand a bit more about this relationship. We do a simple decomposition of
the changes in revenue and employment (Up, Down, No Change) where we consider a 3x3
matrix of outcomes for these two variables. The largest share of responses is in the cell on the
diagonal of ““No Change” for both revenue and employment. This cell accounts for more than
one-third of all observations. The cell with the second largest share is the off-diagonal element
where revenue is “Down” but employment is ““No Change.” This cell accounts for slightly more
than one-third of all observations. The importance of this off-diagonal cell provides some
evidence of labor hoarding. In fact, small businesses with declining revenue were more than
three times more likely not to change employment than to decrease employment. In the
regression results below, we examine the extent to which this possible labor hoarding is

correlated with receipt of PPP loans.

4. Firm Characteristics: Urban/Rural Designation

The Census Bureau maintains a database of businesses in the United States called the
Business Register which it continually updates with data from Economic Censuses, surveys, and
administrative data. Our Geography Division provides geographic codes (‘geocodes’) for
establishment addresses in the Business Register. Both the physical address of the establishment
and the mailing address of the establishment are geocoded to include a rural/urban flag. Because
mailing addresses do not necessarily reflect the location where business is conducted, the SBPS
tabulations using urban and rural designations are based on the physical address of the business.

Furthermore, establishment geocoding is more successful for physical addresses.

The Census Bureau defines urban areas based on the results of the most recent Decennial

Census. Urban areas are “urbanized areas of 50,000 or more population and urban clusters of at

'S Analysis of another Census Bureau collection suggests that business expectations for manufacturing plants are
correlated with actual outcomes in normal times (see Davis et al. (2020)).

10
least 2,500 and less than 50,000 population.’’!° Areas that lie outside of urban areas are
considered rural; Figure 1 presents a map of urbanized areas and urban clusters. Note that these
estimates are consistent with the Census Bureau’s 2010 urban-rural designations, but that there is
no consistent urban-rural designation across federal agencies.'”'* Thus, our measure is consistent
with standard Census Bureau usage, but we draw the reader’s attention to the caveat that the
designation is now based on older data and may not reflect changes resulting from continued

urban development over the course of the decade

Once rural/urban flags are assigned to SBPS responses, estimates and their standard
errors are calculated using the SBPS survey estimates methodology.'? Weekly estimates are
constructed for each week of the first three phases of the SBPS. For each question asked on the
SBPS, estimates are calculated as the percent of businesses responding to each answer such that
for any given question responses sum to 100%. SBPS tabulation weights are used in the
construction of the estimates. Standard errors are calculated using the delete-a-group jack knife
estimation. About 80% of small businesses in the target population of the SPBS have physical
locations that are geocoded urban; about 10% are rural and about 10% are not classified.”° The
distribution of responses, both unweighted and weighted, is very similar to the distribution of the
businesses in the target population of the SBPS. (Sample weights are used for the target

population, and tabulation weights are used for the responses.)

Table 1 presents the distribution of responses to the SBPS by sector overall as well as by
urban and rural designations using the physical location of the business. Distributions are
constructed using survey tabulation weights for each phase, then averaged across the three
phases. The distribution of urban businesses by sector is very similar to the national distribution,

as most businesses (80%) in the SBPS are classified as urban. Notable differences are that there

'6 See https://www.federalregister.gov/documents/201 1/08/24/2011-21647/urban-area-criteria-for-the-2010-census
'” For information on Census Bureau urban-rural designation see Ratcliffe et al (2016).

'8 For example, the Economic Research Service, Department of Agriculture uses a metro/nonmetro county
designation as well as additionally refined measures of rurality; recent work by Bucholz, Molfino, and Kolko (2020)
apply machine learning to survey data to predict the likelihood of the average household in a census tract describing
their location as rural, urban, or suburban.

'9 See the SBPS Methodology at https://portal.census. gov/pulse/data/#methodology.

20 These can be compared with the results from the Small Business Credit Survey (SBCS) for reference year 2019
but with the caveat that the SBCS uses the 2016 Annual Survey of Entrepreneurs to reweight their results by
demographics (see page 31 of their report). Thus, with the caveat that the SBCS uses Census Bureau data to re-
weight their results, this can be compared to the SBCS which finds 84% of small businesses are urban and 16% are
tural in 2019.

11
are fewer urban respondents in Construction (23) and more in Professional, Scientific, and

Technical Services (54) and Health Care and Social Assistance (62).

Rural differences are much larger: the share of rural businesses in Construction (23), at
23.2%, is nearly double the national share of 12.9% and is the largest sector for rural businesses.
As a share, there are fewer rural businesses in Finance and Insurance (52); Professional,

Scientific, and Technical Services (54); and Health Care and Social Assistance (62).

Over phases 1-3, small businesses in construction fared relatively better than small
businesses in health care and social services in terms of overall impact and revenue. Thus, we
might expect that these sectoral differences would lead to relatively better outcomes for small
businesses in rural areas, but going against this tendency is the larger share of urban small
businesses in finance and insurance and professional, scientific, and technical services, which

also had relatively better outcomes.

5. Owner Characteristics: Sex, Race, Ethnicity, Veteran Status

The Annual Business Survey (ABS), conducted by the Census Bureau in partnership with
the National Center for Science and Engineering Statistics within the National Science
Foundation, produces data on employer businesses by sex, race, ethnicity and veteran status of
owners across all non-farm sectors of the economy. The ABS was first conducted for reference
period 2017, however researchers have used its precursors to study owner characteristics and
business outcomes. For example, Brown et al. (2019) use information on race, ethnicity, and
gender from the Survey of Business Owners when examining high-growth businesses. Fairlie
and Robb examine differences by race (2007) and gender (2009) using the Characteristics of

Business Owners.

The main race categories used by the ABS for publication are American Indian or
Alaska Native (AIAN), Asian, Black or African American, Native Hawaiian or Other Pacific

Islander (NHOPI) and White.*!” Not all businesses are classified by ownership characteristics

21 The ABS collects more detailed race data but publishes using only these five categories in non-Economic Census
years. See, e.g., the 2020 ABS questionnaire for additional details.

>? The race and ethnicity categories used by the ABS and this paper are based on the Office of Management and
Budget Statistical Policy Directive 15.

12
in the ABS; this includes publicly-held businesses as well as those not classifiable because a
majority ownership share does not exist. Differences between the totals produced in the ABS and
the detailed breakout by owner characteristics represent these not classified businesses.
Businesses may be classified in more than one race category if the owner(s) report more than one

race.

Because businesses may have more than one owner, the share of the business owned by
persons that are the same sex, race, or ethnicity across owners are used to calculate ownership by
these characteristics; data are collected for up to four owners. Ownership of a business by a
given characteristic requires ownership of more than 50% by owners reporting that
characteristic.”> Note that race and ethnicity are not mutually exclusive, and owners may be more
than one race in addition to (non-)Hispanic. For example, if a business has four owners with
equal ownership shares and three owners are Black and Hispanic, the business is classified as
both Black owned as well as Hispanic owned, as the share of the business owned by Black
owners and Hispanic owners (75%) exceeds 50%. If a business has one female owner and one
male owner with equal ownership shares, the business is classified as equally owned by sex, and
similarly if a business is owned equally by Hispanic and non-Hispanic owners the business is
classified equally owned by ethnicity. ABS data are published by each of these statuses -
majority owned and equally owned - for sex, ethnicity, veteran status, and minority/non-minority

ownership. ABS data are published only for majority owned by race.

Overall, the 2018 ABS surveyed over 800,000 businesses and the 2019 ABS surveyed
over 300,000 businesses. The published results for businesses with less than 500 employees from
the 2018 and 2019 ABS are shown in Table 2; note that is not directly comparable to the SBPS
because of the differences in scope between the ABS and the SBPS (the published ABS data
include multi-unit employer businesses). We have reproduced Table 2 using the ABS data for
only businesses in-scope for the SBPS and find similar results to column 2; however, given the
small number of multi-unit businesses in some of the owner characteristics categories we cannot

publish the table.

3 See Brown et al. (2019) for a different way to assign ownership. They use firm-owner observations, weighted by
their ownership shares. Their method produces estimates more reflective of the equity distribution among the top
four owners which may matter if only owner groups with a majority of equity influence firm behavior.

13
The demographics by sex for the business owner(s) using the 2019 ABS are: 61.2% male,
19.8% female, and 15.0% equally owned. In terms of race, 83.0% are White, 10.3% Asian, 2.2%
Black, and less than 1% for AIAN. In terms of ethnicity, 5.8% are Hispanic and 89.2% are non-
Hispanic. Finally, 87.5% are non-Veteran owned.”* For NHOPI and Veteran-owned businesses,
we must rely on the 2018 publication due to suppressions; less than 1% are NHOPI owned and

6.1% are Veteran owned.

By matching ABS survey data to the Small Business Pulse Survey (SBPS) responses, it is
possible to generate SBPS estimates by these owner characteristics. The 2018 and 2019 ABS
matched responses are used to construct ownership characteristics for the matched SPBS
responses. As noted above, the 2018 ABS surveyed over 800,000 businesses, resulting in the
largest possible number of responding businesses for matching; additional years of the ABS that
resample businesses that responded in 2018 are not asked the owner characteristics questions.
SBPS responses not matched to the 2018 ABS are matched to the 2019 ABS (which recall

surveyed over 300,000 businesses).

The SBPS survey estimates by owner characteristics are the result of matching SBPS
response data to the 2018 and 2019 ABS survey data.”° Unique business identifiers are used to
match the SBPS and ABS response files. Using phase 1 data, approximately 28% of SBPS
responses are matched to the ABS; match rates for subsequent phases are similar. (The SBPS has
an average response rate of about 25%; thus, our matched sample on a non-weighted basis is
about 7% of the SBPS target population). Ownership shares and sex, race, ethnicity, and veteran
status values from the ABS are assigned to SBPS reporters. Majority ownership is calculated
using the same methodology as the ABS; an ownership share of more than 50% in total for a
given sex, race, ethnicity, or veteran status is required for the firm to be considered owned by
that sex, race, ethnicity, or veteran status. Businesses may be included in multiple race

categories as in addition to sex, ethnicity, and veteran status categories. If no majority sex, race,

24 Again, with the with the caveat that the SBCS uses Census Bureau data to re-weight their results, we can compare
to the SBCS. The gender breakdown for the SBCS is: 65% male, 21% female, and 15% equally owned. The SBCS
combines race and ethnicity and reports: 82% Non-Hispanic White, 11% Non-Hispanic Asian, 5% Hispanic, and 2%
Non-Hispanic Black or African American.
°5 The SBPS mail file was drawn from the 2018 Business Register (BR); the 2018 (2019) ABS sample was drawn
from the 2016 (2017) BR, likely leading to a lower match rate than if all were drawn from the same BR year.

14
ethnicity, or veteran status ownership can be established, or the SBPS records cannot be matched
to the ABS, these are considered not classified and are not included in the owner characteristics

For the SBPS-ABS matched file, once owner characteristics flags are created using the
above methodology, estimates and standard errors for the estimates are created using the SBPS
survey estimates methodology.”’ Estimates consist of the weighted percent of businesses
responding to each answer for each question on the SBPS, produced for each week of the SPBS
phases | - 3. Weights used for these tabulations are the SBPS weights adjusted for survey non-
response. Standard errors are calculated using the delete-a-group jack knife estimation. Estimates
for release include those by sex (female, male, Equally owned), race (AIAN, Asian, Black or
African American, NHOPI, White), ethnicity (Hispanic, non-Hispanic), and veteran status

(Veteran, Non-Veteran).

The use of the SBPS tabulation weights to create estimates by owner characteristics
necessitates additional caution when interpreting estimates and micro data-based results, as the
underlying data is the result of combining two samples with different target populations and
strata; results based on this matched data may not be applicable to the larger universe. Ideally,
our matched sample would be adjusted to the true target population for each owner characteristic
(i.e., we would know the ownership characteristics for all single unit small employer businesses
that existed at least as far back as 2017 and would adjust our weights such that our sample was
representative at the national level by these characteristics). While initial research is being done
at the Census Bureau to assign these characteristics to all businesses on the Business Register,

we cannot do this weight adjustment at this time.

Using the ABS-SBPS matched responses, weighted shares of SBPS businesses by sex,
race and ethnicity are calculated and presented in Table 3. Shares are constructed for each phase
and then averaged for phases 1-3. Following the ABS, firm ownership by sex, race and ethnicity
is defined as businesses having a majority ownership (>50%) by owner(s) belonging to a sex,

race, ethnicity, or veteran status. Race and ethnicity are not mutually exclusive and thus owners

6 Thus, this differs from Brown et al. (2019) since their method of assigning ownership can classify businesses
without majority status.

27 See the SBPS Methodology at https://portal.census.gov/pulse/data/#methodology. The ABS methodology is
available at Annual Business Survey Methodology (census.gov)

15
may be more than one race in addition to (non-) Hispanic. Table 3 also includes the share of
businesses in these matched sets that cannot be assigned to a majority ownership category using

the reported ABS characteristics and ownership shares.

Table 3 can be used to compare the classification of SBPS/ABS matches to published
2018 and 2019 ABS totals. Note that Table 3 includes national estimates for all employer
businesses with fewer than 500 employees, not just the single unit employer businesses with
fewer than 500 employees that are the target population of the SBPS. While we are not releasing
additional tabulations due to concerns about disclosure, we have constructed these estimates
selecting only single unit firms and found a distribution similar to the published estimates

reproduced here.

Businesses found in the SBPS-ABS matches found in Table 3 are less likely to be
female-owned, male-owned, or equally-owned by sex than the businesses in the published ABS
data, and more likely to not be classified by sex. Businesses in the SBPS-ABS matches found in
Table 3 are more likely to be not classified by race than in the published ABS results, with
almost all categories of firm ownership by specified race being lower for SBPS matched
businesses than in the published ABS. SBPS matched businesses are less likely to be Hispanic or
Non-Hispanic and more likely to be not classified by ethnicity. Last, SBPS matched businesses
are less likely to be Veteran or non-Veteran owned, and there are no businesses in the matched

file that are equally Veteran and non-Veteran owned.

Table 4 provides nonemployer data by owner characteristics for comparison to
demonstrate important differences between small employer businesses, the target of the SBPS,
and nonemployers. First, there are far more (nearly five times) the number of nonemployer
businesses than employer businesses. Second, nonemployer businesses are far more likely than
small employer businesses to be female-, Black-, or Hispanic-owned. Thus, the SBPS has a
limited ability to reflect the experience of minority-owned businesses as the majority of

minority-owned businesses are nonemployers.

Table 5 compares the representativeness by sector of the SBPS-ABS matched data relative
to the SBPS responses. Shares are constructed for each phase and then averaged for phases 1-3.
The SBPS-ABS matched responses are less likely to be in Construction (23), Retail (44), and
Food and Accommodation (72) and more likely to be in Manufacturing (31) and Wholesale (42)

16
than the SBPS overall responses (which are adjusted to be representative of the survey’s target

population nationally and at the sector level).

6. Research Data Product

We start by describing the results as provided in the public-use files. Since Buffington et
al. (2021a) find important sectoral differences for small businesses, ideally, we would provide
each of the characteristics of interest by sector (and by state). However, the sample sizes that we
have seen above for our characteristics do not support such a plan. Instead, published estimates
for each question (and standard errors) are provided at the National level by owner
characteristics as well as by rural and urban location. These are less than ideal but still provide
useful information. We attempt to provide some metrics of their use when we turn to using
microdata in Section 7. The option of using micro data is available to qualified researchers on
approved projects through the FSRDC. In the time of a fast-moving pandemic (or other similar
emergency), many interested parties are not in a position to undertake research in the FSRDC, so

it is our hope that the data provided will fill some needs.

Before we turn to the results, we note that not all differences are statistically significant
between groups or over time and may reflect survey error and not real economic differences.
Caution should be used in interpreting differences in estimates either between groups or across
time; Appendix C provides guidance for evaluating differences between these published
estimates. We note how some of these results are robust to controls such as sectors, while others

are not (which we demonstrate in Section 7).

Caution is also warranted since it is not clear what role the sample selection issues play in
these results. The matched SPBS-ABS sample used here is not representative nationally for race,
sex, ethnicity, or veteran status of ownership; for example, our matched sample has only about
half of the share of black owned businesses expected (2.1 vs. 1.1 percent according to Table 3).
Future research on the sample selection issues with the matched sample used here is needed to

understand the impact on the results found in the remainder of this paper. Understanding and

17
resolving these issues may be contingent on more general research assigning ownership

characteristics to the Business Register.

To preview our results, we find most of the differences by owners are concentrated in the
negative and more neutral outcomes. Since our focus is on changes in revenue, this means that
we see the largest differences for decreases in revenue or no change in revenue.”® In contrast, the
differences for increases in revenue tend to be smaller. (Similarly, the biggest differences for
overall sentiment are usually in large negative impact or no impact.) This is not surprising given
the near universal experience of negative outcomes from the pandemic. Time series plots for all
five core questions appear in Appendices A (location) and B (owner characteristics). In the
interest of space, we only reproduce figures for revenue changes in the body of the paper (but

discuss all results).

We start by discussing results by location. Figure 2 shows the changes in revenue over all
three phases by National (blue solid line), Rural (red dashed line), and Urban (green dash-dot
line). Small businesses in rural relative to urban have better outcomes in terms of a smaller share
facing declining revenues (top panel) and a larger share facing no change in revenues (middle
panel) but are relatively similar for shares with increasing revenues (bottom panel). The
outcomes for positive impacts and moderate negative impact are not very different than those for
urban. We see a similar pattern for employment changes but with less variation across the
location types (and even less variation for hours changes). In terms of overall sentiment, small
businesses in rural locations relative to urban locations have better outcomes in terms of having a
smaller share with large negative impact and a larger share with no impact. Finally, not
surprisingly, small businesses in rural areas have more optimistic expectations than do small
businesses in urban areas as measured by smaller share of rural businesses choosing the longest
horizon for a return to normal and having a larger share choosing little or no effect or already

returned to normal.

We now look at results by owner characteristics starting with sex. We reemphasize that

ownership of a business by a given characteristic requires ownership of more than 50% by

°8 Since one the three categories of shares of responses for revenue changes has very small numbers, the remaining
two numbers almost mechanically offset each other. This relationship is not as mechanical for overall impact where
there are five choices and the gaps are largest for “large negative impact” and “little or no impact.”

18
owners reporting that characteristic. Figure 3 shows revenue change over all three phases for
National (blue solid line), businesses owned by females (red dashed lines), and businesses owned
by males (green dash-dot lines). Small businesses owned by females relative to those owned by
males have higher shares reporting a decline in revenue (top panel) and a lower share reporting
no change in revenue (middle panel). This pattern is repeated for employment and hours (but
with smaller gaps between male-owned and female-owned businesses). In terms of overall
sentiment, relative to small businesses owned by males, small businesses owned by females have
a higher share reporting large negative impact and a lower share reporting little or no effect.
Finally, small businesses owned by females have less optimistic expectations than those owned
by males in terms of having higher shares expecting a return to normal over the longest horizon

(6 or more months) and smaller shares having had little or no effect.

Now, we turn to race starting with American Indian and Alaskan Native (AIAN)-owned
businesses in Figure 4. One notable feature of Figure 4 is the volatility of the shares of small
AIAN-owned businesses percentages over all three possible responses for revenue change. For
example, the percent of small AIAN-owned businesses with decreases in revenue in the last
week is about 80% in mid-May, is below 40% in mid-June, and is close to 60% by the end of
June. This volatility reflects the small sample size of this group in our matched sample. This

underscores the importance of using the confidence intervals when using the estimates.

Turning to small Asian-owned businesses, in Figure 5 we see that relative to the national
average, these businesses have a higher percentage of businesses with declining revenue (and
that this gap appears to be largest in phase 2). In the same vein, there is a smaller percentage of
small Asian-owned businesses with no change in revenue relative to National (and again, this
gap appears to be largest in phase 2). Finally, there is a smaller percentage of small Asian-owned
businesses in phase | with increases in revenue as compared to National. This is one of the few
groups where we can see a gap on the increase side. In terms of the overall impact, relative to
National, small Asian-owned businesses have a large percentage with a large negative effect but
a smaller percentage with a moderate negative effect (the percentages with no effect is volatile

and the percentages with moderate effect are similar to National).

Small Black-owned businesses are shown in Figure 6 which plots the percent of small

businesses by changes in revenue. All three responses show large volatility over the three phases,

19
likely reflecting the small sample size. If one instead focuses on labor market outcomes, there is
a clear pattern in phase 1 of small Black-owned businesses having both a larger percentage of
businesses with declines in hours and employment and a smaller percentage of businesses with

no change in hours and employment relative to National (see figures in Appendix B).

Figure 7 presents the results for Native Hawaiian and Other Pacific Islander (NHOPI)-
owned businesses. The results for NHOPI-owned businesses are very volatile over all three
response categories for revenue. Again, this likely reflects the very small sample for small
NHOPI-owned businesses in our matched sample. The gaps in the figures are when there is an

insufficient number of responses to allow for disclosure of results.

Closing out the results by race, Figure 8 shows the results for small White-owned
businesses as compared to National results. Recall that 76.3% of the small businesses in our
matched sample are White-owned and thus we would expect the revenue responses to be
relatively similar across the two groups. A lower share of small White-owned businesses had

revenue declines and a larger share had no change in revenue.

Next, we look at ethnicity. Figure 9 shows the three revenue responses for small
Hispanic-owned businesses and non-Hispanic-owned businesses. The responses for Hispanic-
owned businesses in the red dashed lines are volatile over response categories and over time.
However, in the first phase of the SBPS, there appears to be a slightly clearer pattern of small
Hispanic-owned businesses having a large share with declining revenues and a smaller share
with unchanging revenues relative to the National results. This pattern is also apparent for

changes in employment and hours especially in the later weeks of phase | (see Appendix slides).

Finally, we present results for veteran-owned businesses. Figure 10 shows that compared
to small non-Veteran-owned businesses, a slightly lower share of small Veteran-owned
businesses have declining revenue and a somewhat higher share have no change in revenue.
Again, the patterns for changes in employment and hours are similar to that for revenue but the
differences across owner type are muted. Relative to small non-Veteran-owned businesses, small
Veteran-owned businesses report a smaller share with large negative impact and a larger share

with no impact (but the series are very volatile). Veteran-owned small businesses have a shorter

20
expectation of a return to normal (smaller share choosing the longest horizon, larger share

choosing little or no impact).

In sum, small businesses whose revenue outcomes are more negative (higher percentage
of decreasing revenue and lower percentage of no change in revenue) are more commonly
located in urban areas, with owner(s) who are female, Asian, (and to a lesser extent) Black, and
non-Veteran. While these figures provide some insights, earlier work has shown the importance
of sectoral and temporal (and to a lesser extent, location) differences. We now turn to the

empirical exercises where we are able to control for these and other characteristics.
7. Microdata Results in an Econometric Framework

In this section, we examine how the responses to five core questions differ by business
location and business owner characteristics in an econometric framework using the micro-level
data. This framework allows us to succinctly measure the significance of the differences and to
control for other observable characteristics of businesses. From the publicly available results
from the SBPS, we know that sectoral (2-digit NAICS) and time (week collected) variation are
important, and variation across states is also relevant (but less important). We therefore control
for these dimensions in our econometric framework by adding week, state, and sector fixed
effects. In addition, we also include controls for business size (using employment size class bins)

and for response count.

Response count captures the number of times the business responded to the SBPS (with
values ranging from | to 3). Because the responses to the survey are influenced by several
selection effects, it is important to be able to control for other fixed business characteristics,
many unobserved, that may be correlated with such selection.”? The response count can control
for a variety of fixed business characteristics that matter for the outcomes of interest. It may
partially control for survival effects and self-selection into survey response. For instance, it is
plausible that businesses are more likely to respond to Census Bureau surveys when they are not

struggling to survive.°*° Nevertheless, care must be taken when interpreting this variable. For

9 Even though we have repeated observations for some businesses across the phases, we cannot add a business fixed
effect to our specifications because it would be perfectly correlated with all the other time-invariant characteristics
we are interested in (e.g. ethnicity, race, sex, and location).

3° But this may be different in times of extreme negative shocks. The SBPS included an option for write-in
responses, some businesses appeared to respond precisely to have the chance to tell their story of struggle.

21
example, a value of 3 may result because the business survived over all three phases and it was
willing to respond. However, a value of 1 could also mean that the business survived to phase 3

but simply chose not to respond to phases 1-2.

For estimation, we translate the response categories for each question into ordinal
values.*! The response categories readily lend themselves to an ordinal ranking for all questions
we are interested in. Overall sentiment response categories (large negative effect, moderate
negative effect, no effect, moderate positive effect, large positive effect) are converted to five
numerical values with higher numbers representing more positive outcomes. Revenue change,
employment change, and hours change each take on three values where the lowest value is for
decreased, followed by no change, and then increased. Finally, expectations take on six values
where lower values are for more pessimistic expectations (closing is the most pessimistic) and
higher values are for more optimistic expectations (already returned to normal is the most

optimistic).

Because our dependent variables are all ordinal, we use an ordered probit approach to
estimate the relationships we are interested in. Our first set of models considers the five
characteristics (location, sex, ethnicity, race, and veteran status) separately as explanatory
variables, together with the controls (state, 2-digit sector, week of the survey, the response count
indicatory, and employment bins*). We include an indicator for whether the business reported

receiving PPP funds as well.

We start by discussing the results of the “one-way” specifications which are summarized
in Table 6. Consider first the results for location. Given the relatively similar sectoral
distributions across urban and rural locations, we may expect our unconditional findings to hold
with our controls. Since the models control for state, the coefficients can be interpreted as
within-state differences between urban and rural locations. In other words, the specification
controls for time-invariant state policies, but not time-varying state effects such as shutdowns or
re-openings nor local policies. The results from this specification show that small businesses in

rural locations tend to have better outcomes in terms of overall sentiment and revenue changes,

31 We re-used the scheme used for SBPS indexing. See https://portal.census.gov/pulse/data/downloads/Index-
construction-for-the-Small-Business-Pulse-Survey.pdf.
32 The employment bins considered are 0-4, 5-9, 10-19, 20-49, 50-99, 100-249, 250-499, and 500+ employees.

22
and more optimistic expectations. They are also less likely to have declines in employment and

hours.

The first business owner characteristic that we examine is sex. The coefficient on female
is negative and significant for all five core concepts. That is, relative to small businesses with
male owner(s), small businesses female owner(s) are more likely to experience negative
outcomes for revenue changes and overall impact (likely due to larger negative responses) when
we control for other characteristics. They also are more likely to experience decreased
employment and hours. Finally, their expectations tend to be more pessimistic. Thus, the

unconditional patterns displayed in the figures are robust to adding controls.

Turning to race, across all models two of the owner race categories have all insignificant
coefficients relative to the omitted group (White). These categories are American Indian and
Alaskan Native (AIAN) and Native Hawaiian and Other Pacific Islander (NHOPI). Recall that
the plots for these groups were very volatile, in part due to the very small samples. One owner
category, Asian business owner(s), has all negative and significant coefficients. Thus, even when
controlling for sectoral (and other differences), these small businesses are found to have more
negative experiences. Finally, the results for small businesses with Black owner(s) indicate
negative and significant coefficients on four of the five concepts (revenue, employment, hours,
and expectations). The fifth concept, overall impact, has a negative but insignificant coefficient.
Thus, the estimation results reveal some conditional patterns obscured in the figures and show
that small businesses with Black owner(s) generally faced more negative experiences compared

to the omitted group.

We next look at estimation results for Ethnicity where the omitted group is non-Hispanic.
Only the coefficient for hours is negative and significant (all other coefficients are negative but
insignificant). Thus, we can only say that small businesses with Hispanic owner(s) saw more
negative impact on hours (likely through larger shares of businesses with decreasing hours)
relative to small businesses with non-Hispanic owner(s). Recall that the unconditional results
presented in the figures also showed the largest differences for hours (and that these were

concentrated in the latter part of phase 1).

Lastly, we look at the assigned Veteran status of the business ownership. The time series

plots weakly suggested that small businesses with a majority ownership held by Veterans fared

23
better than those with majority ownership held by non-Veterans. In terms of sectors, the ranking
of the sectors for Veteran-owned businesses are similar to those for the entire economy. From the
2018 ABS, we know that the top sectors for Veteran-owned businesses are Professional,
Scientific, and Technical Services (54) and Construction (23), sectors which did not see as large
negative impacts in the SBPS which may lead us to expect that the difference will be attenuated
when we add controls, and Health Care and Social Services (62) and Retail Trade (45), sectors
which did see large negative impacts in the SBPS. The estimation results show that these
differences between small Veteran-owned businesses and non-Veteran owned businesses

disappear once we control for other characteristics.

In sum, our ordered probit estimations using business characteristics one at a time show
that small businesses whose revenue outcomes are more negative (higher percentage of
decreasing revenue, lower percentage of no change in revenue, lower percentage of increasing
revenue) are more concentrated in small businesses located in urban areas, and with owner(s)
who are female, Asian, and Black. Business owner characteristics that did not show significant
differences (possibly due to small sample sizes) include AIAN, NHOPI, Hispanic, and Veteran.
That is, the results confirm the patterns in figures that looked strong for small businesses in rural
areas and with female or Asian owners, reveal patterns that were obscured for small businesses
with Black owners, confirm the patterns of volatility to AIAN, NHOPI, and Hispanic, and

reinforce the weak patterns for small businesses with Veteran owner(s).

Table 6 also includes the probit estimation of our five core questions on whether the
business received PPP, using our full set of controls but not including location or owner
characteristics. The receipt of PPP is positively and significantly associated with changes in
employment and hours and negatively and significantly associated with changes in revenue,
sentiment, and expectations. This is suggestive that businesses faring relatively worse during the
pandemic were more likely to request federal assistance through this program. While
employment and hours are positively associated with the receipt of PPP, we caution against

interpreting this as a causal relationship.

We highlight here the average adjusted predictions for these one-way models for which
coefficients were statistically significant in the discussion above. Table 7 presents averaged

adjusted predictions for decreasing revenue. Receiving PPP is associated with an increase the

24
probability by 7.2% of reporting decreased revenues. Rural businesses are 2.1% less likely to
report decreasing revenues than urban businesses. By owner characteristic, female-owned
businesses are 1.6% more likely to report decreasing revenues than male-owned businesses, and
Hispanic-owned businesses are 1.3% more likely than non-Hispanic-owned businesses to report
this. Asian-owned businesses are 5.6% more likely to report decreasing revenue, while Black-

owned businesses are 4.0% more likely to report decreasing revenue.

Adjusted predictions for employment change are similar across different businesses but
smaller in magnitude. Table 8 presents average adjusted predictions for decreasing employment.
Businesses that recetved PPP were 2.7% less likely to report a decrease in employment than
those that did not receive PPP. Female-owned businesses were 1.7% more likely to report a
decrease in employment than male-owned businesses, while Asian- and Black-owned businesses
were 3.1% and 3.4% more likely to report a decrease in employment than White-owned
businesses. Urban and Hispanic-owned businesses were more likely to report a decrease in
employment than rural and non-Hispanic-owned businesses, respectively, but these differences

while statistically significant are less than 1%.

Our last empirical exercise using the micro-level data is the fully saturated model whose
results are presented in Table 9. As with the one-way models, AIAN- and NHOPI-owned
businesses are associated with negative sentiment and expectations as well as revenue and
employment outcomes relative to White-owned businesses, but these results are not statistically
significant, likely due to the small number of these businesses in the matched samples as well as
in the population of business owners. Asian and Black ownership is associated with statistically
significant negative outcomes for revenue, employment, hours, and outlook; Asian ownership is

associated with statistically significant negative outcomes for sentiment as well.

Female-owned businesses are associated, statistically significantly, with more negative
outcomes relative to male-owned businesses for all five core concepts, whereas businesses
located in rural areas are associated, statistically significantly, with more positive outcomes
relative to urban businesses across all five concepts. Model coefficients for ethnicity and veteran
status are not statistically significant, except for Hispanic owned businesses being associated

with a more negative outcome for hours. Last, receiving PPP continues to have a statistically

25
significant positive relationship to hours and employment and a negative relationship to

sentiment, revenue, and expectations.

8. Conclusion and Future Research

The Small Business Pulse Survey (SBPS) provides timely information during the
COVID-19 pandemic about an important part of the firm population, small businesses. The
SBPS website allows users to see weekly results for subjects such as changes in revenue,
employment, and hours; overall impact; business expectations; operational challenges; and other
content at the national, state, and selected MSA level by sector (2-digit NAICS). A variety of
stakeholders, including policymakers, contacted the Census Bureau to request these statistics by
demographics of business owners and by rural and urban designations. We describe our attempt
to be responsive to these requests by providing estimates of the SBPS by demographics of
business owners and location. We provide a detailed description of our methodology so users can
determine the fitness for use for their specific needs. We also provide an econometric analysis of
five of the core concepts questions by owner demographics and location as further information

relevant to determining fitness for use.

Published statistics for the first three phases of the SBPS by owner characteristics as well
as by rural and urban location are available for download on the SBPS website. Downloadable
files include estimates for each question as well as standard errors. For the convenience of users,
we provide three appendices. Time series graphs for the five questions comprising the core
content (business sentiment, business expectations, and changes in revenue, employment, and
hours) are found in the appendices: Rural and Urban Published Estimates are in Appendix A,
Owner Characteristics Published Estimates are in Appendix B. Appendix C provides guidance
for evaluating differences between these published estimates since caution should be used in

interpreting differences in estimates either between groups or across time.

As is evident in our discussion, one main takeaway from the paper is the complexity of
combining survey results from two very different surveys. By contrast, it is relatively straight-
forward to combine information from the Business Register with the SBPS to provide estimates

of the rural and urban designation. The two surveys differ in terms of the target population and

26
the sampling strategy. Added to these differences is the complexity of the methodology
associated with assigning business owner characteristics in the ABS. We attempt to appropriately

highlight what these challenges mean when interpreting the estimates provided.

One limitation when considering the results of the Small Business Pulse Survey,
especially as it relates to minority-owned businesses, is that the target population for the SBPS
excludes nonemployers which account for the majority of small businesses and the majority of
minority-owned businesses in the US. The main challenge in creating SBPS results by owner
characteristics is the match between the two surveys (and the complexity of assigning business
ownership); care should be used when interpreting these data due to documented shortcomings
of the match. Future research could pay close attention to understanding sample selection issues
in linking the SBPS and the ABS; future research is needed to overcome these limitations with
the ultimate goal to make analysis of such matched samples more representative. Likewise,
caution should be used when interpreting regression results given the relatively small numbers of
minority-owned employer businesses in general as well as in data matching both
samples. Future research on assigning owner characteristics from multiple data sources to the

universe of firms Business Register is a promising avenue of inquiry.

The COVID-19 pandemic highlights the need for timely, frequent, detailed data. The
Census Bureau released some experimental data products (such as the SBPS) to meet these
needs. The products discussed in this paper are research data products as they are a first attempt
to address a complicated measurement issue with the goal of providing useful information while
encouraging feedback. These data will be available in the FSRDCs for qualified users on
approved projects to use so that other users can provide additional insights from these data. We
will use feedback from this release in determining whether a future release for the next three
phases of the SBPS is warranted. As with any research project, feedback is critical for our ability

to continually refine and improve our methods and approach.

27
Table 1. Percent of Respondent Businesses in the SBPS by Sector, Location Type

Sector Overall! Urban? Rural’
All 80 10
11 0.2 0.1 0.9
21 0.3 0.2 0.9
22 0.1 0.1 0.3
23 12.9 11.6 23.3
31 4.2 3.9 6
42 5.0 5.0 4.7
44 11.0 10.9 10.3
48 3.3 2.8 6.0
51 1.3 1.4 1.2
52 4.1 4.4 2.2
53 5.7 5.9 4.5
54 14.4 15.4 9.0
55 0.1 0.1 0.1
56 6.1 5.9 7.7
61 1.7 1.7 1.1
62 11.3 12.2 4.8
71 2.4 2.3 2.8
72 9.2 9.1 8.4
81 7.0 7.2 5.8

Notes: (1) Percent by sector and overall numbers are constructed using SBPS tabulation weights for each phase 1-3
of the SBPS and then averaged across phases. (2) Calculated as in (1) using rural and urban classification of physical
business location. Columns may not sum to 100% due to rounding. Approximately 10% of businesses reporting to
the SBPS do not have physical locations that are geocoded for rural/urban designation.

28
Table 2. Number and Percent of Employer Businesses (1-499 employees) by Sex, Race, Ethnicity,

and Veteran Status, ABS 2018 and 2019

Number of 2018 ABS Number of 2019 ABS
businesses, % businesses, %
Characteristic Ownership ABS 2018 Businesses ABS 2019 Businesses
Total Total 5,132,898 4,990,502
Sex Male 3,121,561 60.8 3,051,861 61.2
Female 997,047 19.4 987,375 19.8
Equally owned 765,615 14.9 748,444 15.0
Not classified 4.7 3.9
Race American Indian and Alaska Native 21,679 0.4 21,047 0.4
Asian 503,653 9.8 514,837 10.3
Black or African American 108,931 2.1 107,383 2.2
Native Hawaiian and Other Pacific
Islander 6,067 0.1 S S
White 4,248,686 82.8 4,141,506 83.0
Not classified 4.7 3.9
Ethnicity Non-Hispanic 4,551,127 88.7 4,453,600 89.2
Hispanic 287,842 5.6 290,877 5.8
Equally owned 45,254 0.9 42,668 0.9
Not classified 4.7 3.9
Veteran Status Veteran 313,905 6.1 N) N)
Equally owned 138,465 2.7 127,515 2.6
Non-veteran 4,431,850 86.3 4,365,772 87.5
Not classified 4.7 3.9

Source: Tables AB1700CSA01, AB1800CSAO1 are used for the Not classified statistics. Tables AB1700CSA04,

AB1800CSA04 are used for the remaining statistics. All shares are authors’ calculations. Cells marked ‘S’ are

incalculable due to suppressions in the published data.

29
Table 3. Percent of Businesses in the SBPS-ABS Matched Responses by Sex, Race, Ethnicity, and
Veteran Status

SBPS-ABS
2018 ABS Match
% %
Characteristic Ownership Businesses! _ Businesses”
Sex Male 60.8 55.6
Female 19.4 16.0
Equally owned 14.9 13.1
Not classified 4.7 15.3
Race American Indian and Alaska Native 0.4 0.5
Asian 9.8 5.9
Black or African American 2.1 1.1
Native Hawaiian and Other Pacific Islander 0.1 0.1
White 82.8 76.3
Not classified 4.7 16.0
Ethnicity Non-Hispanic 88.7 80.9
Hispanic 5.6 3.1
Equally owned 0.9 0.8
Not classified 4.7 15.3
Veteran Status Veteran 6.1 5.3
Equally owned 2.7 0
Non-veteran 86.3 76.9
Not classified 4.7 17.8

Notes: (1) Reproduced from previous table for convenience. (2) Constructed using SBPS tabulation weights for each
phase 1-3 of the SBPS and then averaged across phases. Columns may not sum to 100% due to rounding.

30
Table 4. NES-D 2017 Ownership by Sex, Race, Ethnicity, and Veteran Status

0

Characteristic | Ownership Number of ‘ ”
Businesses Businesses

Total Total 25,310,000
Sex Male 13,910,000 55.0
Female 10,550,000 41.7
Equally owned 612,000 2.4
Race American Indian and Alaska Native 84,500 0.3
Asian 1,960,000 7.7
Black or African American 2,951,000 11.7
Native Hawaiian and Other Pacific Islander 38,500 0.2
White 19,990,000 79.0
Ethnicity Non-Hispanic 21,390,000 84.5
Hispanic 3,635,000 14.4
Equally owned 44,500 0.2
Veteran status Veteran 1,421,000 5.6
Equally owned 100,000 0.4
Non-Veteran 23,550,000 93.0

Source: Nonemployer Statistics by Demographics program, Table | - Statistics for Nonemployer
Businesses by Industry, Sex, Ethnicity, Race, and Veteran Status for the U.S., States, and Metro Areas:
2017, U.S. Census Bureau. Accessed 2021-04-07, and authors’ calculations. Columns may not sum to
total due to rounding.

31
Table 5. Percent of Businesses in the SBPS-ABS Matched Responses by Sector

Sector Overall! ABS-SBPS match?
11 0.2 0.3
21 0.3 0.5
22 0.1 0.3
23 12.9 10.5
31 4.2 6.2
42 5.0 3.8
44 11.0 10.8
48 3.3 4.1
51 1.3 2.4
52 4.1 3.9
53 5.7 5.6
54 14.4 17.6
55 0.1 0.1
56 6.1 5.1
61 1.7 2.5
62 11.3 10.5
71 2.4 4.2
72 9.2 6.6
81 7.0 5.0

Notes: (1) Percents by sector are constructed using SBPS
tabulation weights for each phase 1-3 of the SBPS and then
averaged across phases. (2) Calculated as in (1) using
SBPS-ABS matched responses. Columns may not sum to
100% due to rounding.

32
Table 6. One-way Ordered Probit Estimates

Overall Revenue Employment Hours Outlook
PPP Received PPP -0.427*#* -0.190*** 0.118*** 0.0280*** -0.252***
(0.00495) (0.00446) (0.00500) (0.00469) (0.00499)
Location Rural 0.156*** 0.0562*** 0.0342*** 0.0335*** 0.143***
(0.00705) (0.00661) (0.00766) (0.00724) (0.00728)
Location not classified 0.0449*** 0.0194** -0.00352 0.0167* 0.0494***
(0.00703) (0.00674) (0.00786) (0.00705) (0.00659)
Received PPP -0.425*** -0.189*** 0.119*** 0.0286*** -0.250***
(0.00495) (0.00446) (0.00501) (0.00470) (0.00499)
Sex Female -0.0799*** -0.0420*** -0.0755*** -0.0616*** -0.0597***
(0.0107) (0.01000) (0.0114) (0.0105) (0.0100)
Sex equally owned -0.0221 0.0205 -0.0195 0.00683 -0.0179
(0.0121) (0.0115) (0.0128) (0.0119) (0.0115)
Sex not classified -0.0144** -0.0162** -0.0215*** -0.0137* -0.0203***
(0.00556) (0.00529) (0.00609) (0.00555) (0.00544)
Received PPP -0.428*** -0.190*** 0.118*** 0.0279*** -0.252***
(0.00495) (0.00446) (0.00500) (0.00469) (0.00499)
Ethnicity Hispanic -0.0169 -0.0331 -0.0409 -0.0647** -0.00407
(0.0225) (0.0215) (0.0261) (0.0230) (0.0210)
Ethnicity equally owned -0.147** -0.0137 -0.0821 -0.0238 -0.0828*
(0.0482) (0.0456) (0.0474) (0.0437) (0.0389)
Ethnicity not classified 0.00211 -0.0128** -0.00621 -0.00554 -0.00703
(0.00493) (0.00466) (0.00533) (0.00491) (0.00474)
Received PPP -0.427*#* -0.190*** 0.118*** 0.0279*** -0.252***
(0.00495) (0.00446) (0.00500) (0.00469) (0.00499)

33
Table 6. One-way Ordered Probit Estimates (continued)

Overall Revenue Employment Hours Outlook
Race AIAN -0.103* -0.101 -0.111 -0.0835 -0.0597
(0.0520) (0.0527) (0.0572) (0.0583) (0.0491)
Asian -0.0823*** -0.146*** -0.137*** -0.0883*** -0.141***
(0.0172) (0.0162) (0.0179) (0.0160) (0.0151)
Black -0.0602 -0.104** -0.147*** -0.123*** -0.0878**
(0.0343) (0.0347) (0.0404) (0.0349) (0.0321)
NHOPI -0.0380 -0.0685 -0.152 0.00904 -0.0576
(0.105) (0.0953) (0.110) (0.110) (0.0997)
Race not classified -0.00400 -0.0235*** -0.0177** -0.0114* -0.0185***
(0.00502) (0.00475) (0.00546) (0.00503) (0.00486)
Received PPP -0.427*** -0.190*** 0.118*** 0.0279*** -0.252***
(0.00495) (0.00446) (0.00500) (0.00469) (0.00499)
Veteran status Veteran 0.0174 -0.0168 0.00492 0.00656 0.0217
(0.0164) (0.0152) (0.0177) (0.0161) (0.0162)
Veteran status not
classified 0.00576 -0.0125** -0.00429 -0.00328 -0.00405
(0.00501) (0.00475) (0.00542) (0.00498) (0.00480)
Received PPP -0.427*** -0.190*** 0.118*** 0.0280*** -0.252***
(0.00495) (0.00446) (0.00500) (0.00469) (0.00499)

All models estimated using 620,000 observations and tabulation weights from the SBPS. All models include controls for state, sector, employment
size class, and week of the survey as well as an indicator for the number of times responding to the survey. Standard errors are clustered at the firm
level and are in parentheses (* p<0.05, ** p<0.01, *** p<0.001).

34
Table 7. Adjusted Predictions for Revenue Decreasing

Margin SE Z
PPP Received PPP 0.413 0.001 296.0
Did not receive PPP 0.485 0.001 479.4
Location Rural 0.445 0.002 187.7
Urban 0.466 0.001 496.3
Sex Female 0.474 0.003 137.9
Male 0.458 0.002 246.7
Ethnicity Hispanic 0.472 0.008 58.0
non-Hispanic 0.459 0.002 289.9
Race AIAN 0.494 0.020 24.4
Asian 0.511 0.006 84.6
Black 0.495 0.013 37.3
NHOPI 0.481 0.037 13.2
White 0.455 0.002 280.5
Veteran
status Veteran 0.466 0.006 82.2
non-Veteran 0.459 0.002 283.1

Adjusted predictions averaged across all responses for one-way models of
revenue for outcome decreased revenue.
Table 8. Adjusted Predictions for Employment Decreasing

Margin SE Z
PPP Did not receive PPP 0.157 0.001 153.0
Received PPP 0.130 0.001 193.5
Location Rural 0.132 0.002 86.9
Urban 0.139 0.001 212.9
Sex Female 0.151 0.002 64.5
Male 0.134 0.001 110.5
Ethnicity Hispanic 0.146 0.006 25.3
non-Hispanic 0.137 0.001 130.5
Race AIAN 0.160 0.014 11.8
Asian 0.166 0.004 39.5
Black 0.169 0.010 17.1
NHOPI 0.170 0.027 6.3
White 0.135 0.001 125.9
Veteran
status Veteran 0.137 0.004 37.0
non-Veteran 0.138 0.001 127.8

Adjusted predictions averaged across all responses for one-way models of
employment change for outcome decreased number of paid employees.
Table 9. Saturated Ordered Probit Estimates

Overall Revenue Employment Hours Outlook
AIAN -0.0949 -0.0912 -0.102 -0.0726 -0.0551
(0.0520) (0.0527) (0.0573) (0.0584) (0.0492)
Asian -0.0744*** = -0.145*** -0.136*** -0.0881*** -0.135***
(0.0172) (0.0163) (0.0179) (0.0160) (0.0151)
Black -0.0470 -0.0957** -0.140*** -0.116*** -0.0776*
(0.0344) (0.0348) (0.0404) (0.0349) (0.0323)
NHOPI -0.0391 -0.0643 -0.151 0.0124 -0.0581
(0.105) (0.0956) (0.109) (0.109) (0.0994)
Race not classified -0.0644 0.0167 -0.0440 0.00746 -0.0316
(0.0460) (0.0423) (0.0425) (0.0420) (0.0404)
Female -0.0775*** — -0.0400*** -0.0725*** -0.0591*** -0.0553***
(0.0108) (0.0101) (0.0115) (0.0106) (0.0101)
Sex equally owned -0.0236 0.0182 -0.0165 0.00897 -0.0209
(0.0126) (0.0120) (0.0135) (0.0125) (0.0120)
Sex not classified 0.0321 -0.0186 0.0385 0.0101 -0.00674
(0.0511) (0.0485) (0.0499) (0.0479) (0.0461)
Hispanic -0.0147 -0.0387 -0.0474 -0.0677** -0.00723
(0.0226) (0.0216) (0.0262) (0.0231) (0.0211)
Ethnicity equally -0.143** -0.0392 -0.0884 -0.0406 -0.0832*
owned
(0.0484) (0.0458) (0.0480) (0.0442) (0.0394)
Veteran -0.00319 -0.0278 -0.0164 -0.00615 0.00215
(0.0166) (0.0154) (0.0180) (0.0163) (0.0165)
Veteran status not 0.00789 -0.0312 -0.0332 -0.0436 0.00486
classified
(0.0251) (0.0254) (0.0270) (0.0248) (0.0244)
Rural 0.156*** 0.0549*** 0.0329*** 0.0323*** 0.142***

37
(0.00706)
Location not classified 0.0448***

(0.00703)
Received PPP -0.425***
(0.00495)

(0.00662)
0.0196**

(0.00673)
(0.00446)

(0.00767)
-0.00353

(0.00786)
0.118***
(0.00501)

(0.00725)
0.0167*

(0.00705)
0.0282***
(0.00469)

(0.00728)
0.0496***

(0.00659)
(0.00499)

All models estimated using 620,000 observations and tabulation weights from the SBPS. All models include
controls for state, sector, employment size class, and week of the survey as well as an indicator for the number
of times responding to the survey. Standard errors are clustered at the firm level and are in parentheses (*
p<0.05, ** p<0.01, *** p<0.001). Ethnicity not classified parameter estimates could not be calculated due to

collinearity.

38
lon

P3

P3

Week

Revenue
Change - Decrease

Revenue
No change

P2

National ——— Rural — - — Urban]

P2

Revenue Changes by Urban and Rural Locat

.

100 4
100 4

Figure 2

60-5

%

40

Week

National ——— Rural — - — Urban]
Revenue
Change - Increase

P3

P2

100

80 -

60-

%

40-

20-

Week
National ——— Rural — - — Urban]

Source: Authors’ calculations

41
Owner Sex

mess

Revenue Changes by Busi

.

Figure 3

P3

P3

Revenue
Change - Decrease
Week
Revenue
No change

P2

P2
National ——— Female — - — Male |

100 4
100 4

60-4

%

42

Week

National —— — Female — - — Male]
Ive

d Alaska Nat

ican Indian an

Owner Race Ameri

mess

Revenue Changes by Busi

.

Figure 4

Revenue
Change - Decrease

P3

P2

/

\

100 4

%

Week
National —— — AIAN]

Revenue
No change

P3

P2

100 4

40-

%

20-

Week
National —— — AIAN |

44
ian

Owner Race: As

mess

Revenue Changes by Busi

.

Figure 5

Revenue
Change - Decrease

P3

P2

Week
National ——— Asian]
Revenue
No change

100 4

%

20-

P3

P2

T

T

100 4

60 -

%

40-4

204

Week
National — — — Asian |

46
d Other Pacific Islander

lian an

Hawaii

Ive

iness Owner Race Nati

Revenue Changes by Bus

.

Figure 7

Revenue
Change - Decrease

\/

P3

P3

Week

Revenue
No change

—
Week
National — — — NHOP!]

T

P2
P2

100 4
100 +

80 -

60

%

40-
20-

49
Revenue
Change - Increase

P3

P2

T

100

80-

40-

20-

Week

Source: Authors’ calculations

50
Owner Veteran Status

mess

P3

Revenue
Change - Decrease

P2

Revenue Changes by Bus

.

100

%

Figure 10

Week

National ——— Veteran — - — Non-Veteran |

Revenue

No change

P3

P2

100

80

%

Week

National ——— Veteran — - — Non-Veteran |

55
Revenue
Change - Increase

P3

P2

100 4
80-
60
20-

Week

National ——— Veteran — - — Non-Veteran |

Source: Authors’ calculations

56
References

Alekseev, Georgij, Safaa Amer, Manasa Gopal, Theresa Kuchler, JW Schneider, Johannes
Stroebel, Nils C. Wernerfel. 2020. "The Effects of COVID-19 on U.S. Small Businesses:
Evidence from Owners, Managers, and Employees," NBER Working Paper, 27833,
September.

Bloom, Nicholas, Robert S. Fletcher, and Ethan Yeh. 2021. “The Impact of COVID-19 on US
Businesses” NBER Working Paper, 28314, January.

Bloom, Nicholas, Steven J. Davis, Lucia Foster, Brian Lucking, Scott Ohlmacher and Itay
Saporta-Eksten, “Business-Level Expectations and Uncertainty,” Center for Economic
Studies Working Paper No. 20-41, December 2020.

Brown, J. David Brown, John S. Earle, Mee Jung Kim, and Kyung Min Lee. 2019. “Start-ups,
Job Creation, and Founder Characteristics,” Industrial and Corporate Change, 28(6):
1637-1672.

Bucholz, Shawn, Emily Molfino, and Jed Kolko. 2020. “The Urbanization Perceptions Small
Area Index: An Application of Machine Learning and Small Area Estimation to
Household Survey Data.” U.S. Department of Housing and Urban Development Working
Paper.

Buffington, Catherine, Carrie Dennis, Emin Dinlersoz, Lucia Foster, and Shawn Klimek. 2020.
“Measuring the Effect of COVID-19 on U.S. Small Businesses: The Small Business
Pulse Survey” Center for Economic Studies Working Paper, 20-16.

Buffington, Catherine, Daniel Chapman, Emin Dinlersoz, Lucia Foster, and John Haltiwanger.
2021a. “High Frequency Business Dynamics in the United States During the COVID-19
Pandemic” Center for Economic Studies Working Paper, 21-06.

Buffington, Catherine, Jason Fields, and Lucia Foster. 2021b. “Measuring the Impact of COVID-
19 on Businesses and People: Lessons from the Census Bureau’s Experience.” AEA
Papers and Proceedings (forthcoming).

Dua, Andre, Deepa Mahajan, Ingrid Millan, and Shelly Stewart. 2020. “COVID-19’s effect on
minority-owned small businesses in the United States.” McKinsey, May 27, 2020.
https://www.mckinsey.com/industries/public-and-social-sector/our-insights/covid-19s-
effect-on-minority-owned-small-businesses-in-the-united-states

Fairlie, Robert, W. 2020. "The Impact of Covid-19 on Small Business Owners: Evidence of
Early-Stage Losses from the April 2020 Current Population Survey," NBER Working
Paper, 27309, June.

Fairlie, Robert W. and Alicia M. Robb. 2009. “Gender Differences in Business Performance:
Evidence from the Characteristics of Business Owners Survey,” Small Business
Economics, 33:375-395.

Fairlie, Robert W. and Alicia M. Robb. 2007. “Why are Black-Owned Businesses Less
Successful than White-Owned Businesses: The Role of Families, Inheritances, and
Business Human Capital,” Journal of Labor Economics, 25(2): 289-323.

Mills, Claire and Jessica Battisto. 2020.“Double Jeopardy: COVID-19’s Concentrated Health
and Wealth Effects in Black Communities.” Federal Reserve Bank of New York, August.

57
https://www.newyorkfed.org/medialibrary/media/smallbusiness/DoubleJeopardy COVID
-19andBlackOwnedBusinesses

Ratcliffe, Michael, Charlynn Burd, Kelly Holder, and Alison Fields. 2016 “Defining Rural at the
U.S. Census Bureau.” http://www2.census.gov/geo/pdfs/reference/ua/Defining Rural.pdf

58
Appendix B. Rural and Urban Published Estimates

This page intentionally left blank
Overall
Moderate Negative Effect

Overall
Large Negative Effect

P2

P3

P2

P3

—

100 +

80 -

%

100 +

80 -

%

Week

Week

National ——— Rural — - — Urban]

National ——— Rural — - — Urban]

Overall
Moderate Positive Effect

P2

Overall
No Effect

P3

P3

P2

100 +

80 5
60 -

%

404

20-

100 +

80 -
60 -

%

Week

Week

National ——— Rural — - — Urban]

National ——— Rural — - — Urban]
Overall
Large Positive Effect

P2

P3

L

100 +
80 -
60 -
404
20 -
0

%

Week

National ——— Rural — - — Urban]
Hours
Change - Decrease

Hours
Change - Increase

P3

P2

100 +
80 5
60 -

%

P3

P2

100 +
80 5
60 -

%

20-

Week

Week

National ——— Rural — - — Urban]

National ——— Rural — - — Urban]

Hours
No change

P3

P2

100 +

%

Week

National ——— Rural — - — Urban]
Expectations
2-3 months

Expectations
<= 1 month

P3

P2

100 +

80 -

60 -

%

20-

T T T T T T T T

P3

T T T
% S

TT TT
9° 7

P2

T
%
2

100 +

80 -

60 -

%

40-

20-

Week

Week

National ——— Rural — - — Urban]

National ——— Rural — - — Urban]

Expectations
6+ months

Expectations
4-6 months

P3

P2

P3

P2

—— =

100 +

804
604
40 5

%

T
%

T
%

T
S

T
4

T
%

100 +

80 5
60 -
20-
0

%

Week

Week

National ——— Rural — - — Urban]

National ——— Rural — - — Urban]
Expectations
No return to normal

P2 P3

100 -

80 -

60 4

40 -
20-
0 T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National ——— Rural — - — Urban]
Expectations
Permanently closed
P3
100 +
80
60 4
= =

40-
20-
0) “T T T T T T T T T T T T T T

T T T T
Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
9- 16 -23 -30- 6- 13-20-27 -4- 13- 9- 16-23 -30-7- 14-21 -28-4-
Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan
15 22 29 5 12 19 26 3 12 8 15 22 29 6 13 20 27 3 + 10
Week

National ——— Rural — - — Urban]

Expectations
Little or no effect
P2 P3
100 4
80 -
60 =
40 -
0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National ——— Rural — - — Urban]
Expectations
Returned to normal
P3
100 4
80 -
60 =
40 -
20 =
0 T T T T T T T T T T T T T T T T

T T
Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
9- 16 -23 -30- 6- 13-20-27 -4-

13- 9- 16-23-30-7- 14-21-28-4-

Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan
15 22 29 5

12 19 26 3 12 8 15 22 29 6 13. 20 27 3 10
Week

National ——— Rural — - — Urban]
Appendix B. Owner Characteristics Published Estimates (Sex)

This page intentionally left blank
P3

Employment
Change - Decrease

P2

100 +
80 -
60 -
40 -

%

P3

Employment
Change - Increase

P2

100 +
80 -
60 -
40 -

%

Week

Week

National ——— Female — - — Male |

National ——— Female — - — Male |

T

%

Employment
No change
P3

—

P2

%

Week

National ——— Female — - — Male |
Hours
Change - Decrease

Hours
Change - Increase

P3

P2

100 +

%

P3

P2

100 +
80 5
60 5
40 -

%

20 5

Week

Week

National ——— Female — - — Male |

National ——— Female — - — Male |

Hours
No change

P3

P2

100 +

%

Week

National ——— Female — - — Male |
Appendix B. Owner Characteristics Published Estimates (Sex equally owned)

This page intentionally left blank
L
a
r
ge
ositiv
E
ff
e
c
t

P2

P3

| 2 Sb

%

Ww
eek

N

a

tio

nal
E
qually
ow
n
ed |
Expectations
No return to normal

P2 P3
100 +
80 -
60 -

40 -
20-4

0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T

Week
National — — — Equally owned |
Expectations
Permanently closed
P3
100 +
80 -
60 -
40 -
0 —~T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
13- 9- 16-23 -30-7- 14-21-28 - 4-
Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan

9- 16 -23 -30- 6- 13-20-27 -4-
Aug Aug Aug Sep Sep Sep Sep Oct

15 22 29 5 12 19 26 3 12 8

Week

15 22 29 6 = 13

National — — — Equally owned |

20

27

3

10

%

%

100 +

80 -

60 -

40 5

20-

100

80

60

40

20

OA

Expectations
Little or no effect

P2 P3

Expectations
Returned to normal

P3

T T T T T T T T T T T T T T T T T T
Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
9- 16 -23 -30- 6- 13-20-27 -4- 13- 9- 16-23 -30-7- 14-21 -28-4-
Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan
15 22 29 5 12 19 26 3 12 8 15 22 29 6 13 20 27 3 + 10

Week

National — — — Equally owned |
Appendix B. Owner Characteristics Published Estimates (Race American Indian or Alaska Native (AIAN))

This page intentionally left blank
Expectations
No return to normal

P2 P3
100 +
80
60 4
=
40-
20-

er \

0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T

Week
National ——— AIAN |
Expectations
Permanently closed
P3
100 +
80
60 4
=
40-
0 “T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
13- 9- 16-23 -30-7- 14-21-28 - 4-
Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan

9- 16 -23 -30- 6- 13-20-27 -4-
Aug Aug Aug Sep Sep Sep Sep Oct

15

22

29

5

12

19

26 3 12 8 #15

Week

22 29

|——— National]

6

13

20

27

3

10

%

%

Expectations
Little or no effect
P2 P3
100 4
80 -
60 -
40 -
\
0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National —— — AIAN |
Expectations
Returned to normal
P3
100 4
80 -
60 =
40 -
20 =
0 —T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan

9 - 16 - 23 - 30

-6-13-20-27-4-

13- 9- 16-23-30-7- 14-21-28-4-

Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan

15

22 29 5 12 19 26 3 12 8 15 22 29 6 13. 20 27 3 10
Week
National ——— AIAN |
Appendix B. Owner Characteristics Published Estimates (Race Asian)

This page intentionally left blank
Expectations

No return to normal
P2 P3
100 +
80 -
60 -
40 -
20 7 ~ “7
0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National — — — Asian |
Expectations
Permanently closed
P3
100 +
80 -
60 -
40 -
0 “T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan

9- 16 - 23-30 - 6- 13 -20-27-4-

Aug Aug Aug Sep Sep Sep Sep Oct
15 22 29 5 12 19 26 3

13- 9- 16 -23 -30-7- 14-21 -28- 4 -
Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan
12 8 15 22 29 6 13 20 27 3 = 10

Week

National — — — Asian |

%

%

Expectations
Little or no effect
P2 P3
100 +
80 +
60 +
40 -
20 -
0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National — — — Asian |
Expectations
Returned to normal
P3
100 +
80 +
60 +
40 -
0 —T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
9- 16 -23 -30- 6- 13-20-27 -4- 13- 9- 16-23 -30-7- 14-21 -28-4-
Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan

15 22 29 5 12 19 2 3 12 8 15 22 29 6 13 20 27 3 10
Week
National —— — Asian |
Appendix B. Owner Characteristics Published Estimates (Race Black or African American)

This page intentionally left blank
Revenue
Change - Decrease

Revenue
Change - Increase

P3

P2

100 +

%

—

P3

NZ

P2

100 +
80 -
60 -

%

40 5

National — — — Black |

Week
No change

National — — — Black |
Revenue

P3

Week
National — — — Black |

P2

100 +
80 -
60 -

%

40 5
204
Expectations

No return to normal
P2 P3
100 +
80 -
60 -

40 -
20-4 ;

> 4 _ - ~

0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National — — — Black |
Expectations
Permanently closed
P3
100 +
80 -
60 -
40 -
0 —~T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan

9- 16 - 23-30 - 6- 13 -20-27-4-

Aug Aug Aug Sep Sep Sep Sep Oct
15 22 29 5 12 19 26 3

13- 9- 16 -23 -30-7- 14-21 -28- 4 -
Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan
12 8 15 22 29 6 13 20 27 3 = 10

Week

National — — — Black |

%

%

Expectations
Little or no effect
P2 P3
100 +
80 +
60 +
40 -
| /
0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National — — — Black |
Expectations
Returned to normal
P3
100 +
80 +
60 +
40 -
0 —T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
9- 16 -23 -30- 6- 13-20-27 -4- 13- 9- 16-23 -30-7- 14-21 -28-4-
Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan

15 22 29 5 12 19 2 3 12 8 15 22 29 6 13 20 27 3 10
Week
National — — — Black |
Appendix B. Owner Characteristics Published Estimates (Race Native Hawaiian or Other Pacific Islander (NHOPI))

This page intentionally left blank
P3

Employment
Change - Decrease

P2

100 +
80 -
60 -

%

40 5
\
204

P3

Employment
Change - Increase

P2

100 +
80 -
60 -

%

40 5
204

Week

Week
National ——— NHOPI]

P3

Employment
No change

P2

100 +

%

Week
National ——— NHOPI]
Expectations

No return to normal
P2 P3
100 +
80 -
60 -
=
40-
20-
0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National ——— NHOPI]
Expectations
Permanently closed
P3
100 +
80 -
60 -
=
40-
0 “T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct
9- 16 -23 -30- 6- 13-20-27 -4-

Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
13- 9- 16 -23 -30-7- 14-21 -28- 4 -

Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan
15 22 29 5 12 19 26 3 12 8 15 22 29 6 13 20 27 3 + 10
Week

%

%

100

80

60

40

100

80

60

40

20

OA

Expectations
Little or no effect

P2 P3
“T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National — — — NHOPI]|
Expectations
Returned to normal
P3
T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
9- 16 -23 -30- 6- 13-20-27 -4- 13- 9- 16-23 -30-7- 14-21 -28-4-
Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan
15 22 29 5 12 19 26 3 12 8 15 22 29 6 13 20 27 3 + 10

Week

| ——— National]
Appendix B. Owner Characteristics Published Estimates (Race White)

This page intentionally left blank
Overall
Large Positive Effect

P2

P3

100 +

80 5

60 5

%

40 -

20 5

Week
National ——— White |
Revenue
Change - Decrease

Revenue
Change - Increase

P3

=

P2

100 +

%

P3

P2

100 +
80 -
60 -

%

40 5

20-

National —— — White |

Week
National —— — White |

Revenue
No change

—_—

P3

Week
National —— — White |

P2

100 +
80 -
60 -

%
Hours
Change - Decrease

Hours
Change - Increase

P3

P2

100 +
80 -

%

P3

P2

100 +
80 -
60 -
40 -

%

Week
National —— — White |

Week
National —— — White |

Hours
No change

P3

P2

100 +

%

Week
National —— — White |
Appendix B. Owner Characteristics Published Estimates (Ethnicity)

This page intentionally left blank
Overall
Moderate Negative Effect

Overall
Large Negative Effect

P3

P2

P3

P2

100 +

80 -

60 -

40 5

%

20-

100 +

80 -

%

Week

Week

P3

Overall
Moderate Positive Effect

National — — — Hispanic — - — Non-Hispanic |
P2

100 +

80 -

60 -

%

40-

20-

P3

Overall
No Effect

National — — — Hispanic — - — Non-Hispanic |
P2

100 +

80 -

60 5

%

404

20-7

Week

Week

National — — — Hispanic — - — Non-Hispanic |

National — — — Hispanic —— - — Non-Hispanic |
Overall
Large Positive Effect

P2

P3

T
4

100 +

80 5

60 5

%

40 -

20 5

T T
%

T

4
9

Week

National — —— Hispanic — - — Non-Hispanic |
Revenue
Change - Decrease

Revenue
Change - Increase

P3

Week

P2

100 +
80 -
60 -

%

P3

Week

P2

100 +
80 -
60 -

%

40 5
204

National — — — Hispanic — - — Non-Hispanic |

National — — — Hispanic — - — Non-Hispanic |

Revenue
No change

P3

Week

P2

National — — — Hispanic — - — Non-Hispanic |

100 +
80 -
60 -

%
Expectations
2-3 months

Expectations
<= 1 month

P3

P2

100 +

80 -

60 -

%

40 5

20-

P3

P2

100 +

80 -

60 -

%

40 5

20-

Week

Week

National — — — Hispanic — - — Non-Hispanic |

National — — — Hispanic — - — Non-Hispanic |

Expectations
6+ months

Expectations
4-6 months

P3

P2

100 +

P3

P2

80 5
60 +

%

—

—

100 +

80 +
60 +
40 -

%

Week

Week

National — — — Hispanic — - — Non-Hispanic |

National — — — Hispanic — - — Non-Hispanic |
Expectations
No return to normal

P2 P3
100 +
80 -
60 -
=
40-
20-
— ~——" = > = il
0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National — — — Hispanic — - — Non-Hispanic |
Expectations
Permanently closed
P3
100 +
80 -
60 -
=
40-
0 “T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
13- 9- 16-23 -30-7- 14-21-28 - 4-
Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan
10

9- 16 -23 -30- 6- 13-20-27 -4-
Aug Aug Aug Sep Sep Sep Sep Oct
15 22 29 5 12 19 26 3 12 8 15 22 29 6 13 20 27

Week

National — — — Hispanic — - — Non-Hispanic |

3

%

%

Expectations
Little or no effect

P2 P3
100 +
80 +
60 +
0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National — — — Hispanic — - — Non-Hispanic |
Expectations
Returned to normal
P3

100 +
80 +
60 +

_ — ~- —
0 —T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan

9- 16 - 23 -30- 6- 13-20-27 -4-

13- 9- 16-23-30-7- 14-21-28-4-

Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan

15 22 29 5 12 19 26 3 12 8 15 22 29 6 13. 20 27
Week

National — — — Hispanic — - — Non-Hispanic |

3. 10
Appendix B. Owner Characteristics Published Estimates (Veteran status)

This page intentionally left blank
Overall
Moderate Negative Effect

Overall
Large Negative Effect

P2

P3

P2

P3

— SS SS

—=

JN

100 +

80 -

60 -

40 4
204

%

100 +

80 -

%

Week

Week

P3

Overall
Moderate Positive Effect

National ——— Veteran — - — Non-Veteran |
P2

100 +

80 -
60 -

%

40-

20-

P3

Overall
No Effect

(|

National ——— Veteran — - — Non-Veteran |
P2

100 +

80 -
60 5

%

404

20-7

Week

Week

National ——— Veteran — - — Non-Veteran |

National ——— Veteran — - — Non-Veteran |
Overall
Large Positive Effect

P2

P3

T
4

100 +

80 5

60 5

%

40 -

20 5

T T
%

T

4
9

Week

National ——— Veteran — - — Non-Veteran |
Revenue
Change - Decrease

Revenue
Change - Increase

Week

Week

\/
\
\
i\l
|
S&S 8 & F< A
%
{
mo
a
|
)
\

%

National ——— Veteran — - — Non-Veteran |

National ——— Veteran — - — Non-Veteran |

Revenue
No change

P3

_
T
4

P2

100 +

80 -

60 -
40-
20-
0

%

Week
National ——— Veteran — - — Non-Veteran |
P3

Employment
Change - Decrease

P2

100 +
80 -
60 -
40 -
20-

%

P3

Employment
Change - Increase

P2

100 +
80 -
60 -
40 -
20-

%

Week

Week

National ——— Veteran — - — Non-Veteran |

National ——— Veteran — - — Non-Veteran |

—~

P3

Employment
No change

P2

100 +

%

40 -

20-

Week

National ——— Veteran — - — Non-Veteran |
Expectations
2-3 months

Expectations
<= 1 month

P3

P2

P3

P2

100 +

80 -

60 -

%

40 5

20-

100 +

80 -

60 -

%

40 5

20-

Week

Week

National ——— Veteran — - — Non-Veteran |

National ——— Veteran — - — Non-Veteran |

Expectations
6+ months

Expectations
4-6 months

T T

T
%,
OL
46

P3

T
“so
2
So
Ss

\
T
4
&Y
4

P2

T
%G,

%

N/

100 +

P3

P2

80 -
60 -
0

%

T
Q
>

100 +

80 -
60 -
40 -
0

%

Week

Week

National ——— Veteran — - — Non-Veteran |

National ——— Veteran — - — Non-Veteran |
%

%

100 +

80 -

60 -

40 5

100 +

80 -

60 -

40 5

20-

Expectations
No return to normal

P2 P3
=
A= ——— =
T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National ——— Veteran — - — Non-Veteran |
Expectations
Permanently closed
P3
=
“T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
9- 16-23 -30- 6- 13-20-27 -4- 13- 9- 16-23 -30-7- 14-21 -28-4-
Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan
15 22 29 5 12 19 26 3 12 8 15 22 29 6 13 20 27 3 + 10

Week

National ——— Veteran — - — Non-Veteran |

Expectations
Little or no effect

P2 P3
100 +
80 -
60 -
40-
\
0 “T T T T T T T T T T T T T T T T T T T T T T T T T T T
Week
National ——— Veteran — - — Non-Veteran |
Expectations
Returned to normal
P3

100 +
80 -
60 -
40-
20-

= = > = =

0) “T T T T T T T T T T T T T T T T T T

Aug Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan
9- 16 -23 -30- 6- 13-20-27 -4- 13- 9- 16-23 -30-7- 14-21 -28-4-
Aug Aug Aug Sep Sep Sep Sep Oct Oct Nov Nov Nov Nov Dec Dec Dec Dec Jan Jan
15 22 29 5 12 19 26 3 12 8 15 22 29 6 13 20 27 3 + 10

Week

National ——— Veteran — - — Non-Veteran |
Appendix C. User Guide for Published Research Estimates

When comparing estimates between different classes of matched SBPS respondents, it is
important to note that differences between estimates across classes of businesses (e.g., rural and
urban) may not be statistically significant. In order to determine if the difference between two
estimates are statistically significant, a user of the SBPS may use the survey estimates and
standard errors provided in the download files to construct confidence intervals.

Consider rural estimates for the first question of the SBPS in Phase 1. Question 1 asks about the
overall experience of small businesses during the pandemic; response | indicates a large negative
effect was reported by businesses. For rural businesses, the estimate is 42.9% for “Large
negative effect” with a standard error of 1.76. For urban businesses, the estimate is 52.6% for
“Large negative effect” with a standard error of 0.4.

The following can be used to test whether differences between estimates are statistically
significant. The difference between estimates is calculated as:

where d; j is the difference between estimates X; and X j- The measure of error or MOE for a 90%

confidence interval on the difference is approximately:

MOE(dj;) = 1.645 « Jor + 0?(X;)

where o7(X;) and o?(X j) are the variances for X, and X j» respectively. The variances may be
approximated by the square of the standard error for the estimates. A 90% confidence interval for
the difference is used to determine whether the difference is significantly different from zero at
the 90% confidence level. If the interval includes zero, the difference is not significantly
different from zero at the 90% confidence level.

di; + 1.645 « |o?(&) + 07(X)

The difference between the urban and rural estimates in the example given above is:

a

durban rural = 52.6% — 42.9% = 9.7%

The MOE is calculated as:

MOE(durbanrural) = 1.645 * (1.76)? + (0.4)? =2.97%

60
The 90% confidence interval for the difference is
9.76% +2.97% or [6.79%, 12.63%]

Because the interval does not include zero, we can state at the 90% confidence level that
difference between the rural and urban estimates is significantly different.

61


File and source

File

buffington-et-al-2021-ces-wp-21-24-sbps-owner-characteristics-rural-urban.pdf

Size

20,990,054 bytes

SHA-256

33416f7547970f0bc8e143db2d8599100536ad8f59cf67df701da413a70506a3

Our copy

buffington-et-al-2021-ces-wp-21-24-sbps-owner-characteristics-rural-urban.pdf

Original

www2.census.gov