The U.S. Labor Market During the Beginning of the Pandemic Recession (NBER Working Paper 27159)
Summary
NBER Working Paper No. 27159, The U.S. Labor Market during the Beginning of the Pandemic Recession, by Tomaz Cajner, Leland D. Crane, Ryan A. Decker, John Grigsby, Adrian Hamins-Puertolas, Erik Hurst, Christopher Kurz, and Ahu Yildirmaz, dated May 2020, Revised August 2020. The abstract states that the authors use weekly administrative payroll data from the largest U.S. payroll processing company to measure the U.S. labor market during the first four months of the pandemic. It states that aggregate employment fell by 21 percent through late-April and rebounded somewhat through late-June, that employment for workers in the lowest wage quintile was still 20 percent lower relative to mid-February levels as of late June, and that businesses have cut nominal wages for almost 7 million workers.
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
- Issuer
- Source documents
- Document type
- The U S Labor Market During The Beginning Of The Pandemic Recession
- Case
- The U S Labor Market During The Beginning Of The Pandemic Recession
Cite this document
“The U.S. Labor Market During the Beginning of the Pandemic Recession (NBER Working Paper 27159).”
Original at www.nber.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/the-u-s-labor-market-during-the-beginning-of-the-pandemic-recession/
, SHA-256 84f154a0a857fd4258acae0eaa7abf54ddfe472c6acb70c91141ec9d62518e53.Full text
NBER WORKING PAPER SERIES
THE U.S. LABOR MARKET DURING THE BEGINNING OF THE PANDEMIC
RECESSION
Tomaz Cajner
Leland D. Crane
Ryan A. Decker
John Grigsby
Adrian Hamins-Puertolas
Erik Hurst
Christopher Kurz
Ahu Yildirmaz
Working Paper 27159
http://www.nber.org/papers/w27159
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
May 2020
We thank Sinem Buber, Mita Goldar and Matthew Levin from ADP for their support on the
project. We also thank Steve Davis, Jan Eberly and Jonathan Parker for comments on prior drafts.
As part of the University of Chicago data use contract, ADP reviewed the paper prior to
distribution with the sole focus of making sure that the paper did not release information that
would compromise the privacy of their clients or reveal proprietary information about the ADP
business model. The views expressed in the paper are the authors' and do not necessarily reflect
the views of ADP or the National Bureau of Economic Research. Additionally, the analysis and
conclusions set forth here are those of the authors and do not indicate concurrence by other
members of the research staff or the Board of Governors.
At least one co-author has disclosed a financial relationship of potential relevance for this research.
Further information is available online at http://www.nber.org/papers/w27159.ack
NBER working papers are circulated for discussion and comment purposes. They have not been peer-
reviewed or been subject to the review by the NBER Board of Directors that accompanies official
NBER publications.
© 2020 by Tomaz Cajner, Leland D. Crane, Ryan A. Decker, John Grigsby, Adrian Hamins-Puertolas,
Erik Hurst, Christopher Kurz, and Ahu Yildirmaz. All rights reserved. Short sections of text, not to
exceed two paragraphs, may be quoted without explicit permission provided that full credit, including
© notice, is given to the source.
The U.S. Labor Market during the Beginning of the Pandemic Recession
Tomaz Cajner, Leland D. Crane, Ryan A. Decker, John Grigsby, Adrian Hamins-Puertolas,
Erik Hurst, Christopher Kurz, and Ahu Yildirmaz
NBER Working Paper No. 27159
May 2020, Revised August 2020
JEL No. E24,E3,J21,L25
ABSTRACT
Using weekly administrative payroll data from the largest U.S. payroll processing company, we
measure the evolution of the U.S. labor market during the first four months of the global
COVID-19 pandemic. After aggregate employment fell by 21 percent through late-April,
employment rebounded somewhat through late-June. The re-opening of temporarily shuttered
businesses contributed significantly to the employment rebound, particularly for smaller
businesses. We show that worker recall has been an important component of recent employment
gains for both re-opening and continuing businesses. Employment losses have been concentrated
disproportionately among lower wage workers; as of late June employment for workers in the
lowest wage quintile was still 20 percent lower relative to mid-February levels. As a result,
average base wages increased between February and June, though this increase arose entirely
through a composition effect. Finally, we document that businesses have cut nominal wages for
almost 7 million workers while forgoing regularly scheduled wage increases for many others.
Tomaz Cajner Adrian Hamins-Puertolas
Federal Reserve Board Federal Reserve Board
20th Street and C Street, NW 20th Street and C Street NW
Washington, DC 20551 Washington, DC 20551
tomaz.cajner@frb.gov Adrian.B.Hamins-Puertolas@frb.gov
Leland D. Crane Erik Hurst
Federal Reserve Board Booth School of Business
20th Street and C Street, NW University of Chicago
Washington, DC 20551 Harper Center
leland.d.crane@frb.gov Chicago, IL 60637
and NBER
Ryan A. Decker erik.hurst@chicagobooth.edu
Federal Reserve Board
20th Street and C Street, NW Christopher Kurz
Washington, DC 20551 Board of Governors of the Federal Reserve System
ryan.a.decker@frb.gov 20th and C Streets NW
Mail Stop 82
John Grigsby Washington, DC 20551
Department of Economics christopher.j.kurz@frb.gov
University of Chicago
1160 East 58th Street Ahu Yildirmaz
Chicago, IL 60637 Automatic Data Processing
jgrigsby@uchicago.edu ahu.yildirmaz@adp.com
A data appendix is available at http://www.nber.org/data-appendix/w27159
Introduction
We use administrative data from ADP—one of the world’s largest providers of cloud-based
human resources management solutions—to measure detailed changes in the U.S. labor mar-
ket during the first few months of the Pandemic Recession.1 In the current pandemic, data
from ADP have many advantages over existing data sources. First, ADP processes payroll
for about 26 million U.S. workers each month (about 20% of total U.S. private employment).
As discussed in Cajner et al. (2018), Cajner et al. (2020) and Grigsby et al. (2019), the
ADP data are representative of the U.S. workforce along many labor market dimensions.
The sample sizes are orders of magnitudes larger than most household surveys, which mea-
sure individual labor market outcomes at monthly frequencies. Second, the ADP data are
available at weekly frequencies. As a result, statistics on the labor market can be observed
in almost real time. This facilitates high-frequency analysis such as examining employment
responses when states lift closure restrictions on certain industries. Third, the ADP data
link both workers and firms which permits study of worker recall. The data also include
worker and firm characteristics that allow for the estimation of the distributional effects of
the recession across demographic group, industry, firm size, and location. Finally, the data
include administrative measures of wages which are free from measurement error facilitating
the study of nominal wage adjustments. Collectively, the ADP data allow for a detailed anal-
ysis of high-frequency changes in labor market conditions in the first months of the current
Pandemic Recession, complementing the data produced by U.S. statistical agencies.
We find that paid U.S. private sector employment declined by 21 percent between mid-
February and late-April 2020 and then rebounded partially thereafter. As of late June, U.S.
employment is still 13 percent below February levels. About 30 percent of the employment
decline through mid-April was driven by business shutdowns. However, some of these busi-
nesses started coming back during May and June, albeit at a smaller size. About one-third
of the increase in U.S. paid employment since the late-April trough can be attributed to the
re-opening of businesses that temporarily closed. Employment declines through April were
largest for businesses with fewer than 50 employees, with closures and re-openings playing
an even larger role for this size group. We also document that re-entering businesses are
primarily bringing back their original employees. Finally, we find that despite a staggering
fifty percent of all continuing businesses substantively shrinking between February and June,
1
Importantly, our series are constructed from the ADP microdata and are distinct from the National Em-
ployment Report (NER), the monthly employment series published jointly by ADP and Moody’s which has
the goal of predicting BLS employment numbers. The ADP microdata tracked the last recession remarkably
well; Appendix Figure A1 shows that monthly employment changes using ADP microdata closely match the
monthly employment changes reported in the BLS’s Current Employment Statistics (CES) survey during
the 15 years prior to the Pandemic Recession.
1
over ten percent of businesses actually grew during this time period.
Importantly, we also show employment declines were disproportionately concentrated
among lower-wage workers. Segmenting workers into wage quintiles, we show that more
than 35 percent of all workers in the bottom quintile of the wage distribution lost their job—
at least temporarily—through mid-April. The comparable number for workers in the top
quintile was only 9 percent. Through late June, bottom quintile workers still had employment
declines of 20 percent relative to February levels, but many previously nonemployed workers
had been re-called to their prior employer. We also find that employment declines were about
4 percentage points larger for women than for men. Very little of the differences across wage
groups or gender can be explained by business characteristics such as firm size or industry.
Finally, we show that that states that re-opened earlier had larger employment gains in the
re-opening sectors.
The massive decline in employment at the lower end of the wage distribution implies
meaningful selection effects when interpreting aggregate data. For example, we document
that average wages of employed workers rose sharply—by over six percent—between Febru-
ary and April, consistent with official data. However, all of this increase is due to the
changing composition of the workforce. After controlling for worker fixed effects, worker
base wages during the beginning of the recession have been flat. Moreover, we find evidence
that businesses are much less likely to increase the wages of their workers and slightly much
more likely to cut the wages of their workers during the first four months of the Pandemic
Recession. We find that nearly 7 million continuously employed workers received a nominal
wage cut between March and June 2020.2
Section I. Data and Methodology
We use anonymized administrative data provided by ADP. ADP is a large international
provider of human resources services including payroll processing, benefits management, tax
services, and compliance. ADP has more than 810,000 clients worldwide and now processes
payroll for over 26 million individual workers in the United States per month. The data
allow us to produce a variety of metrics to measure high-frequency labor market changes
2
Our paper complements many recent papers which use a variety of different data sources to track labor
market outcomes during the recent recession. A sampling of those papers includes: Bartik et al. (2020a),
Bartik et al. (2020b) Barrero et al. (2020), Bick and Blandin (2020), Brynjolfsson et al. (2020), Chetty et al.
(2020), Dingel and Neiman (2020), Coibion et al. (2020), Kahn et al. (2020) and Kurmann et al. (2020). As
discussed above, our ADP data have advantages over the data used in many of these other papers in that
they are nationally representative, have large sample sizes, track both employment and wages, and allow for
the joint matching of individual workers to individual businesses. For overlapping questions, our findings
are mostly similar to the results in these other papers. When results differ, we discuss further in the text.
2
for a large segment of the U.S. workforce. A detailed discussion of the data and all variable
definitions can be found in the paper’s online appendix.
We use two separate anonymized data sets—one measuring business level outcomes
and another measuring employee level outcomes—to compute high-frequency labor mar-
ket changes. The business-level data set reports payroll information during each pay period.
Each business’ record is updated at the end of every pay period for each ADP client.3 The
record consists of the date payroll was processed, employment information for the pay period,
and many time-invariant business characteristics such as NAICS industry code. Business
records include both the number of paychecks issued in a given pay period (“paid” employ-
ees) and the total number of individuals employed (“active” employees). Paid employees
include any workers issued regular paychecks during the pay period as well as those issued
bonus checks or any other payments. Active employees include paid employees as well as
workers with no earnings in the pay period (such as workers on unpaid leave or workers who
are temporarily laid-off).
The data begin in July 1999 but are available at a weekly frequency only since July
2009. As shown in Cajner et al. (2018), ADP payroll data appear to be quite representative
of the U.S. economy; the data modestly overrepresent the manufacturing sector and large
businesses, but we emphasize that coverage is substantial across the entire industry and size
distribution. While some forms of selection into ADP cannot be observed (i.e., certain types
of firms choose to contract with ADP), we ensure representativeness in terms of observables
by reweighting the data to match Statistics of U.S. Businesses (SUSB) employment shares
by firm size and 2-digit NAICS industry; a further discussion can be found in the online
appendix. For businesses that do not process payroll every week (for example, businesses
whose workers are paid biweekly), we create weekly data by assuming the payroll in the
missing intermediate period is what is observed in the next period for which the business
processes payroll. We then build a weekly time series of employment for each business.4
The business-level data report payroll aggregates for each business. For a very large subset
of businesses, we also have access to their anonymized de-identified individual-level employee
3
Note that we use the terms “business” and “firm” throughout the paper to denote ADP clients. Often,
entire firms contract with ADP. However, sometimes establishments or units within a firm contract separately.
The notion of business in our data is therefore a mix of Census Bureau notions of an establishment (i.e., a
single operating business location) and a firm (i.e., a collection of establishments under unified operational
control or ownership).
4
The methodology we adopt for this paper differs slightly from that used in our previous work with the
ADP business-level data (e.g., Cajner et al. (2018) and Cajner et al. (2020)). In particular, in light of the
extreme employment changes during the beginning of the Pandemic Recession, in the present work we do not
seasonally adjust the data, and we measure employment changes of surviving businesses, closing businesses,
and re-opening businesses relative to mid-February levels rather than constructing longer-term time series.
3
data.5 That is, we can see detailed anonymized payroll data for individual workers. As with
the business data, all identifying characteristics (names, addresses, etc.) are omitted from our
research files. Workers are provided an anonymized unique identifier by ADP so that workers
may be followed over time. We observe various additional demographic characteristics such
as the worker’s age, gender, tenure at the business, and residential state location. We also
can match the workers to their employer. As with the business-level data described above,
we can observe the industry and business size of their employers.
The benefits of the employee data relative to the business data described above are three-
fold. First, we can explore employment trends by worker characteristics such as age, gender,
initial wage levels, and worker residence state. This allows us to discuss the distributional
effects of the current recession across different types of workers. Second, the individual-level
data allow us to measure additional labor market outcomes such as worker wages as well as
recall rates of a given worker as businesses start to re-open. Finally, the panel structure of
the data permits analysis of individual wage dynamics.
In all the work that follows, we will indicate whether we are using the business-level data
or the employee-level data for our analysis. Unless indicated otherwise, we report weighted
results.6
Section II. Employment Changes in the Pandemic Re-
cession
This section presents weekly labor market indices in the United States compiled from the
ADP business-level microdata. Panel A of Figure 1 shows our estimated aggregate employ-
ment changes spanning the payroll week covering February 15 through late June.7 Impor-
tantly, this panel shows employment changes at both continuing businesses and businesses
that have shut down (i.e., those not issuing any paychecks during regularly scheduled pay
periods), where shutdowns could reflect either permanent or temporary inactivity. The in-
5
Unlike the business-level data, the data for our employee sample skew toward employees working in
businesses with at least 50 employees. This is the same data used in Grigsby et al. (2019). While the data
come from employees mostly in businesses with more than 50 employees, there is representation in this data
for employees throughout the business size distribution. Again, we weight these data so that they match
aggregate employment patterns by industry and firm size from the SUSB.
6
For all aggregate results, the weighted-employment changes found in both data sets are nearly identical
during the beginning of the Pandemic Recession.
7
For all figures based on the business-level data, we report 2-week trailing moving averages to smooth
through volatility that results from offsetting pay frequencies across ADP businesses, the majority of which
are biweekly but not all occurring on the same weeks. Also, these results – like all results in the paper – are
weighted to match SUSB employment (treating the ADP businesses as firms). In the online appendix, we
show results weighting to the match QCEW employment (treating the ADP businesses as establishments).
4
Figure 1: Aggregate Paid and Active Employment
1.05 1.05
U.S. Employment Relative to February 15th U.S. Employment Relative to February 15th
1.00 1.00
0.95 0.95
0.90 0.90
0.85 0.85
0.80 0.80
0.75 0.75
0.70 0.70
Paid Employment Active Employment Paid Employment, Continuers Active Employment, Continuers
Panel A: All Panel B: Continuing
Businesses Businesses
Notes: Figure shows employment changes relative to February 15th within the ADP business-level
sample at the weekly frequency using 2-week trailing moving averages. The solid black line with
circles in Panel A shows the trend in paid payroll employment for all businesses. The dashed black
line with squares in Panel A shows the trend in active employment for all businesses. Panel B shows
the same patterns for businesses who continually make scheduled payroll payments throughout the
entire sample period starting on February 15th.
dices plot employment levels relative to February 15th levels without seasonal adjustment.
The figure shows the evolution for paid employees (solid line, circles) and active employees
(dashed line, squares). Between mid-February and the labor market trough in late-April,
paid employment in the U.S. fell by 20.6 percent, and active employment fell by about 11
percent. The sharper drop in paid employment is to be expected if many businesses initially
placed their workers on temporary layoff. Since mid-April, paid employment has increased
by 7.4 percentage points through June 20th. However, as of late June, paid employment in
the U.S. is still about 13 percent below its level at the start of the recession. Importantly,
the bulk of the rebound occurred in May, and the pace of job gains slowed measurably—
or perhaps even stalled—toward the end of June. As we highlight below, the employment
increases since mid-April are associated with many states starting to reopen their businesses.
The job loss numbers in the ADP data are broadly consistent with employment data
published in the BLS’s Current Employment Statistics (CES) survey for overlapping weeks.
The CES, which measures employment during the week containing the 12th of the month,
estimated employment declines of 1.0 million in March and 18.9 million in April followed
by rebounds of 3.7 million in May and 5.5 million in June (not seasonally adjusted). In
our measure of total paid employment, focusing on the pay periods corresponding with CES
reference weeks, we observe employment declines of about 1.2 million in March and 23.9
5
million in April followed by rebounds of 3.8 million in May and 5.3 million in June.8
Panel B of Figure 1 shows employment losses for continuing businesses. We define con-
tinuing businesses in a given week as those businesses who have continually made scheduled
payroll payments between February 15th and that week. Notice that paid employment for
continuing businesses declined by 16.9 percent through late-April before rebounding through
the late June, leaving paid employment 11.2 percent below mid-February levels. The dif-
ferences between Panels A and B highlight the importance of firm closures (which may be
temporary) in driving employment declines through late April and the importance of those
firms re-opening in driving the increase in employment during May and June. Continuing
firms accounted for about three-quarters of the employment losses through late April. We
further explore the importance of business shutdown and re-entry to aggregate employment
trends in Section Section V..
It is worth mentioning the active employment series shown on Figure 1. Recall that
active employment measures the number of workers in payroll databases, including those
not receiving pay in a given pay period.9 Active employment among continuing businesses
actually declined by about 0.5 million jobs during between the April and May CES reference
periods while other measures showed gains; in other words, businesses in continuous opera-
tion trimmed their active employees in the payroll databases, on net, even while aggregate
paid employees increased. This pattern hints at important gross employment flows underly-
ing the net numbers we highlight: even while employment has resumed net growth (driven
largely by the return of temporarily inactive workers), many businesses were shedding jobs.
Much attention has been given to the preservation of small businesses in the current
recession. The $2 trillion stimulus package signed into law on March 27 makes special provi-
sions to support small businesses through a large expansion in federal small business loans,
and a second tranche of small business loan appropriations was signed on April 24. Figure 2
plots the change in employment by initial business size relative to February 15th. The figure
shows that businesses with fewer than 50 employees reduced both paid employment (Panel
A) and active employment (Panel B) at a faster rate than their larger counterparts through-
out March and April. However, businesses of all sizes saw massive employment declines
during the first few months of the current recession. Businesses with fewer than 50 employ-
ees saw paid employment declines of more than 25 percent through April 18, while those
with between 50 and 500 employees and those with more than 500 employees, respectively,
8
Numbers computed using our estimated employment declines multiplied by total US private sector
employment in February 2020. The corresponding numbers for active employment were -0.7 million, -13.1
million, +1.6 million, and +4.4 million for March, April, May, and June, respectively.
9
Importantly, we do not observe active employment of firms not issuing paychecks; that is, active em-
ployment is necessarily zero among firms that have shut down.
6
Figure 2: Employment Change By Business Size
1.05 1.05
U.S. Employment Relative to February 15th U.S. Employment Relative to February 15th
1.00 1.00
0.95 0.95
0.90 0.90
0.85 0.85
0.80 0.80
0.75 0.75
0.70 0.70
1-49 Workers 50-499 Workers 500+ Workers 1-49 Workers 50-499 Workers 500+ Workers
Panel A: Paid Panel B: Active
Employment Employment
Notes: Figure shows employment changes relative to February 15th within the ADP business-
level sample at the weekly frequency with two-week trailing moving averages. Panel A shows the
trend in payroll employment for each size grouping. Panel B shows the same patterns for active
employment.
saw declines of 15-20 percent during that same time period and reached troughs a week
or two later than the smallest businesses.10 Notably, the growth in paid employment since
late April has been much larger for smaller businesses. Between late April and late June,
smaller businesses increased employment by 17 percent (of February 15th levels). Businesses
with more than 50 employees increased employment by between 4 and 7 percent during the
same time period. As employment is rebounding, it is the smaller firms that are primarily
increasing employment. Again, as we highlight below, much of this differential growth for
smaller firms is due to the re-opening of smaller firms that temporarily shuttered during the
state imposed shutdowns.
Figure 2 hides interesting heterogeneity across businesses even within size classes. In
Figure 3, we report the entire distribution of employment changes within and across business
size classes, limiting our focus to businesses that survive through this time period (continuers)
so we can study a meaningful growth distribution. For each initial employment size class,
we report percentiles of employment change between February 15 and June 20th, where
percentiles are constructed from the employment-weighted business distribution.
Starting on the left-hand side of Figure 3, the 10th percentile business within every size
class saw declines of at least 50 percent, with the largest class (at least 500 employees) seeing
10
The somewhat jagged variation in employment changes for the larger businesses is an artifact of the
heterogeneity of varying payroll frequencies. In our employee level data, we can control for the pay frequency
of a given worker exactly and such small week-to-week variations are smoothed out.
7
Figure 3: The Distribution of Employment Change by Business Size Among Survivors
40
20
Employment change (in percent)
0
-20
-40
-60
-80
-100
10 20 30 40 50 60 70 80 90
Percentiles
1-49 employees 50-499 employees 500+ employees
Notes: Figure shows change in paid employment by initial business size and employment change at
each decile of the paid employment-weighted change distribution using the businesses-level sample.
Shutdown businesses are excluded. Change in employment is measured between February 15th and
June 20th.
a decline of about 98 percent. These are large firms that essentially shut down only keeping
a handful of original employees on payroll. Even the smallest business size class (1-49) saw
substantial declines. The facts that (1) small business saw even more overall employment
declines (as highlighted in Figure 2) and (2) employment changes in the bottom decile of
continuing firms were smaller for small businesses suggest that most of the total decline
in employment for businesses with fewer than 50 employees is due to business closures.
Conversely, all business size groups experienced positive growth at the 90th percentile. Even
during the Pandemic Recession, some firms added net employment.
Between the extremes, we also observe a wide range of businesses whose employment is
close to unchanged. Among the smallest size group at least 10 percent of businesses had
little employment change (those spanning the 60th through the 70th percentiles). Similarly
a large swath of mid-size and larger businesses experienced only modest changes (those
spanning the 60th through the 80th percentiles saw changes of less than 7 percent). Taken
together, Figure 3 reveals striking heterogeneity in the experiences of businesses, even within
size classes.11 The median surviving small (less than 50) business declined 5 percent, while
the medium and large median declines were 8 and 11 percent, respectively.
The results by firm size are not overly surprising in light of the industry results docu-
11
We observe qualitatively similar results when focusing on active employment instead of paid employment,
though the distribution of changes in all directions is notably narrower.
8
mented next. The industries that were hit hardest in the beginning of the pandemic recession
also tend to be the industries with the smallest businesses as documented by Hurst and Pugs-
ley (2011). Table 1 shows employment changes by two-digit NAICS industries during two
time periods: February 15-April 25th (the aggregate employment trough, prior to states
starting to re-open) and February 15-June 20th (i.e., the entire period). These results are
shown in columns 1 and 2 of the table, respectively. The largest declines in employment
were in sectors that require substantive interpersonal interactions. Through late-April, paid
employment in the “Arts, Entertainment and Recreation” and “Accommodation and Food
Services” sectors (i.e., leisure and hospitality) both fell by more than 45 percent while em-
ployment in “Retail Trade” fell by almost 30 percent. The “Other Services” industry, which
includes many “local” or neighborhood businesses like laundromats and hair stylists, also
experienced declines in employment of 25.1 percent through late-April. Despite a boom in
emergency care treatment within hospitals, the “Health Care and Social Assistance” indus-
try experienced a 16.5 percent decline in employment through late April. Industries that
employ higher-educated workers—like Finance/Insurance—saw smaller initial employment
declines.
Since bottoming in late April, most sectors have seen some recovery in employment.
Much of the relatively larger increases are in sectors where re-openings have occurred. For
example, most states started re-opening manufacturing and construction sectors in early
May. Both of these sectors saw employment gains of about 20 and 70 percent, respectively,
of their initial employment losses. Large recoveries are also seen in some of the sectors
that saw the largest initial declines, such as accommodation and food services, retail trade,
and other services. Businesses in these three sectors started opening up during May as
many states started to lift restrictions on restaurants, retail outlets, and personal service
businesses such as barbershops, beauty parlors and nail salons. Despite states re-opening
and employment rebounding slightly, employment in these sectors still remains significantly
depressed relative to mid-February levels. Notice, as travel has still remained depressed and
schools remain closed, employment in the transportation and education sectors have not
seen the rebound found in retail trade or food services. Another sector which saw a large
rebound is health care and social assistance which has recovered nearly 40 percent of lost
employment as hospitals and other health providers have attempted to start returning to
normal activities.12
12
In the online appendix, we provide a table of the weekly employment and wage changes by two-digit
sector from February through June to facilitate the calibration of various model of the Pandemic Recession.
9
Table 1: Paid Employment Changes By 2-Digit Industry (in percent)
Feb 15 - Feb 15 -
Industry April 25 June 20
Arts, Entertainment and Recreation -50.7 -31.7
Accommodation and Food Services -45.4 -26.8
Retail Trade -28.9 -17.5
Other Services -25.1 -14.6
Transportation and Warehousing -21.8 -20.5
Real Estate, Rental and Leasing -21.0 -16.9
Information Services -17.7 -11.3
Wholesale Trade -17.5 -12.8
Administrative and Support -16.8 -15.5
Health Care and Social Assistance -16.5 -10.3
Educational Services -16.2 -18.9
Construction -13.8 -4.0
Manufacturing -12.6 -10.4
Professional, Scientific, and Tech Services -12.1 -8.3
Finance and Insurance -1.2 -4.9
Notes: Table shows total (i.e., inclusive of shutdowns) decline in paid employment through April
25th, 2020 (column 1) and through June 20th, 2020 (column 2) for all firms in each two-digit
NAICS industries. All changes are relative to February 15th, 2020. Data from the business-level
sample. Weekly data (without two-week moving average).
Section III. Distributional Effects across Workers
In this section, we document the heterogeneity in job loss across different types of workers
using our employee sample. We begin by exploring the labor market outcomes for workers at
different points of the base wage distribution at the beginning of the current downturn. We
first segment workers by their initial place in the wage distribution. Specifically, we use early
February data to define wage quintiles for our analysis based on a worker’s administrative
base hourly wage. We pool together hourly and salaried workers when making our quintiles.
For hourly workers, we use their exact hourly wage. For salaried workers, we assume the
workers work 40 hours per week when computing their hourly wage. For weekly (biweekly)
salaried individuals, this is just their weekly (biweekly) base administrative contracted earn-
ings divided by 40 (80). We hold these thresholds fixed throughout all other weeks of our
analysis. The nominal thresholds for the quintiles are 13.5, 16.41, 24.53 and 32.45 dollars
10
Figure 4: Employment Changes By Initial Wage Quintile and Gender
1.05 1.05
1.00 1.00
Employment Change Relative to February 1st Employment Change Relative to February 1st
0.95 0.95
0.90 0.90
0.85 0.85
0.80 0.80
0.75 0.75
0.70 0.70
0.65 0.65
0.60 0.60
0.55 0.55
2/1/20 2/8/20 2/15/20 2/22/20 2/29/20 3/7/20 3/14/20 3/21/20 3/28/20 4/4/20 4/11/20 4/18/20 4/25/20 5/2/20 5/9/20 5/16/20 5/23/20 5/30/20 6/6/20 6/13/20 6/20/20 2/1/20 2/8/20 2/15/20 2/22/20 2/29/20 3/7/20 3/14/20 3/21/20 3/28/20 4/4/20 4/11/20 4/18/20 4/25/20 5/2/20 5/9/20 5/16/20 5/23/20 5/30/20 6/6/20 6/13/20 6/20/20
Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Male Female
Panel A: By Wage Quintile Panel B: By Gender
Notes: Figure shows changes in employment through the beginning of the Pandemic Recession by
initial wage quintile (Panel A) and by gender (Panel B). Employment declines measured relative
to early February. Data for this figure use the employee sample.
per hour.13
Panel A of Figure 4 shows the employment changes for workers in different wage quintiles
relative to early February. As seen from the figure, employment declines in the initial stages
of this recession are disproportionately concentrated among lower wage workers. Workers in
the bottom quintile of the wage distribution experienced a staggering 37 percent decline in
employment between early March and late April. Employment for this group has partially
rebounded through late-June, but their employment still remains depressed by 20 percent
relative to mid-February levels. Conversely, employment of workers in the top quintile of the
wage distribution declined 10 percent through the end of April. Only about 4 percent of these
top earning workers remain out of work through late June. The employment losses during
the Pandemic Recession are disproportionately concentrated among lower wage workers.
How much of the larger decline in employment among low-wage workers can be attributed
to the industrial composition of the COVID-19 shock? Low-wage workers are more likely to
work in restaurants, retail, and leisure services and are also more likely to work in smaller
businesses. To assess whether differential exposure to the recession by business characteristics
(industry and business size) or worker characteristics (age and location) can explain the
13
These cutoffs match well the distribution of wages in the 2019 March Supplement of the Current Popu-
lation Survey (CPS). Computing hourly wages as annual earnings last year divided by annual hours worked
last year, the 20th, 40th, 60th, and 80th percentile of hourly wages (measured in nominal dollars per hour)
in the 2019 CPS were 12.0, 17.1, 24.0, and 36.1 (author’s calculation).
11
differential pattern across either gender or the wage distribution, we further exploit the
panel nature of our data and estimate a linear probability model of monthly employment for
a given worker at a given firm on wage quintile dummies and detailed controls for industry
and business size.14 Specifically, we measure whether the employee is paid at that firm at
the beginning of each given month.
The baseline separation probability between February and March is 6.1 percentage points
higher for bottom quintile earners than for top quintile earners. After controlling for only
wage quintile fixed effects, bottom quintile earners were 21.5 percentage points less likely
to be employed by their February employer in the first two weeks of April relative to top
quintile earners, reflecting the patterns in Panel A of Figure 4. Including industry and firm
size fixed effects reduces the gap in excess separation rates between bottom quintile earners
and top quintile earners only slightly to 19.1 percentage points. Therefore, a differential firm
size and industry mix can explain 12.2 percent (1 – 19.1/21.5) of the gap in job loss between
low-wage and high-wage workers during the beginning of this recession, but a substantial gap
remains even after accounting for firm size and industrial composition. However, including
controls for worker age further reduces the gap in excess separation probabilities between
low-wage and high-wage workers to 16.5 percent. As highlighted in the online appendix,
younger workers were more likely to be displaced during the early part of the recession,
and younger workers systematically have lower wages. Overall, we conclude that there is a
substantial difference in the behavior of low- and high-wage workers during the early stages
of the Pandemic Recession. Only a small amount of these differences can be accounted for
by differences in industry, business size, and age.
Panel B of Figure 4 plots employment changes by gender. Through late April, women
experienced a decline in employment that was 4 percentage points larger than men (22
percent vs 18 percent). The gap has remained roughly constant through late June. These
patterns stand in sharp contrast to prior recessions where men experienced larger job declines.
Historically, male dominated industries such as construction and manufacturing contracted
the most during recessions. However, as noted above, this recession is hitting a different set
of industries including retail, leisure and hospitality industries. Can the differential industry
declines explain the gender differences in employment losses? In the appendix, we again
exploit the panel nature of our data to assess this question. Less than 0.5 percentage points
of the 4 to 5 percentage point difference can be explained by industry. In other words,
even within detailed industries, women are experiencing larger job declines relative to men.
The fact that industry or other firm characteristics do not explain the gender difference
14
The online appendix discusses the details of this specification as well as plotting the coefficients and
standard errors from the regression output.
12
in employment declines is interesting in its own right. Future research using household
level surveys with additional demographic variables can explore whether other facets of the
Pandemic—such as the increased need for childcare—explains some portion of the gender
gap in employment losses during this recession.
Section IV. Wage Changes during the Pandemic Re-
cession
Figure 5 shows the trends in wages in the economy during the pandemic recession. The
solid line creates a wage index by measuring the mean contract per-period wage rate of all
working individuals in the economy.15 Since the start of the recession, observed average
wages in the ADP sample grew by nearly 6 percent through mid-May. As highlighted in
Solon et al. (1994), the changing composition of workers over the business cycle can distort
measures of wage cyclicality.16 As seen from Panel A of Figure 4, workers at the bottom of
the wage distribution were much more likely to have employment reductions than those at
the top of the wage distribution. From March through the end of April, the sample became
more selected towards higher-earning individuals, while the reverse happened thereafter.
To assess the importance of this selection, we again exploit the panel nature of the ADP
data. In particular, we compute individual wage growth for a sample of continuing workers
between pay periods t and t+1. By considering individual wage growth rather than levels, we
restrict attention to workers who are in the sample in consecutive periods, thereby purging
the wage series of the principal form of selection. We then produce a selection-adjusted
wage index by chain-weighting this average wage growth from the reference week ending
February 15. The result of adjusting for selection in this way is shown in the dashed line
in Figure 5. Two things are of note. First, despite the rapid nominal wage growth for
the average employed worker (solid line), there is essentially no nominal wage growth for
continuing workers during this period (dashed line). In other words, all of the observed
aggregate wage growth is due to selection. Second, the selection effects are largest through
late April when employment declines were largest. Since late April there has been a decline
in aggregate average unadjusted wages as employment has disproportionately increased for
lower wage workers. These patterns also reveal themselves within industries. In the online
15
Contract per-period wages are the contracted per hour wage rate for workers paid hourly and the
contracted per-period weekly or bi-weekly earnings for salaried workers (depending on pay frequency). The
online appendix outlines the wage concept in greater detail.
16
Grigsby (2019) documents that measured growth in average wages has become countercyclical during
the last few recessions. He documents that the changing selection of workers during the recent recessions
has been responsible for the observed countercyclicality of wages.
13
Figure 5: Trend in Base Wages, Controlling for Selection
1.08
1.06
Average U.S. Wage Relative to February 15th
1.04
1.02
1.00
0.98
0.96
0.94
No Adjustment Selection Adjusted
Notes: Figure shows trends in weekly wages during the beginning of the Pandemic Recession. The
solid line (circles) averages base wages across all employed workers in each period. The dashed line
(triangles) controls for selection by measuring the base wage of a given worker over time.
appendix, we present employment, mean base wage, and selection-adjusted wage indices by
2-digit NAICS industry. In every industry, selection-adjusted base wages are much flatter
than average wages, with the selection-adjusted wage falling in many industries. The flat
composition-adjusted wages in Figure 5 suggest that nominal wage growth has actually
slowed. Normally, over a few month period, nominal wages increase as some workers receive
their regularly scheduled wage adjustments, while wage cuts are exceedingly rare.
This was not the case at the beginning of the Pandemic Recession. Figure 6 provides a
summary comparison of the employee wage adjustment patterns in 2019 (darker bars) and
2020 (lighter bars). Panel A of Figure 6 plots the share of continuously-employed workers
who receive a base wage cut in our sample. Specifically, it plots the share of workers who
were employed with a given firm in both March and June who saw declines in their base per-
period pay rate between March and June. The first column shows that 6.2% of job-stayers
saw wage declines between March and June of 2020. This stands in stark contrast to the
patterns for 2019, when just 1.6% of workers saw wage cuts between these same months.
Base wage cuts are a remarkable feature of the labor market in the Pandemic Recession.
Of course, many workers separated from their job over this time period. Overall, we
find that 5.3% of all workers in our sample who were employed in March (regardless if they
14
Figure 6: Probability of Base Wage Cut and Freeze in 2019 and 2020 by Base Wage Quintile
Sample: Workers at Firms that Usually Adjust Wages in March - June
16 70
% With Base Wage Freeze
% With Base Wage Decrease
20 30 40 50 60
4 8 12
10
0 0
All 1st 2nd 3rd 4th 5th All 1st 2nd 3rd 4th 5th
2019 2020 2019 2020
Panel A: Probability of Wage Cut Panel B: Probability of Wage Freeze
Notes: Figure shows the probability of a wage cut (Panel A) and the probability of a wage freeze
(Panel B) for different wage quintiles. Panel A includes all workers employed with the same firm in
both March and June. Panel B restricts the sample to firms that made 75 percent of their annual
wage changes for their employees in 2019 during March, April, May and June.
remained employed through June) saw a base wage cut between March and June. Given that
total U.S. employment at the beginning of March was 129 million workers, this amounts to
approximately 6.8 million workers receiving base wage cuts in addition to the tens of millions
more who lost their job and remain out of work.
Of course, firms may choose to forgo scheduled wage increases without actually cutting
workers’ wages. Studying such wage freezes is made more complicated by the fact that most
continuously employed workers only receive one base wage change per year. However, as
highlighted in Grigsby et al. (2019), most firms adjust their base wages annually in a given
month. For example, some firms always provide annual base wage adjustments in April
while others do their adjustments in July. To study wage freezes during the beginning of the
Pandemic Recession, we create a sample of firms who did at least 75 percent of their 2019
base wage changes in March, April, May and June. These are firms for which March-June
are their normal base wage adjustment months.
We plot the probability that workers receive a wage freeze (i.e. zero base wage change) in
these firms in Panel B of Figure 6. Column 1 shows that these firms adjusted the base wages
of roughly 64 percent of their continuously employed workers from March through June of
2019 (i.e,. kept the wages fixed of 36 percent of their employees). This number is similar
to the decade-long average of base wage changes within the firms in the ADP employee
sample found in Grigsby et al. (2019). Moreover, essentially all base wage changes in 2019
were increases; these firms only decreased the nominal wages of 0.7% of their workers during
15
these months of 2019. However, during the same four months in 2020, these same firms froze
the wages of 58% of their workers. In addition to the millions of base wage cuts observed in
the last four months, millions of workers have seen zero base wage changes at firms due to
make their annual wage adjustment.17
The 6.3% of workers receiving nominal base wage cuts during the Pandemic Recession is
of similar magnitude to the 6% found by Grigsby et al. (2019) during the Great Recession.
However, many firms have yet to make their scheduled wage adjustments in 2020 and it
remains to be seen whether such base wage cuts will continue. Similarly, during the Great
Recession, over half of workers still received nominal wage increases. So far during the
Pandemic Recession, base wages are increasing much less and decreasing slightly more than
they did during the Great Recession.
The remaining columns of Figure 6 shows the probability of a wage cut (Panel A) and
the probability of a wage freeze (Panel B) for workers in different initial wage quintiles.
Wage freezes were more common throughout the wage distribution in 2020 relative to 2019.
However, while employment losses were concentrated among low wage workers (Figure 4),
nominal wage cuts were disproportionately concentrated among higher wage workers. Over
three quarters of all nominal wage cuts were concentrated in workers in the top two deciles
of the wage distribution. For this sample, 13.4 percent of all workers in the top wage quintile
received a nominal wage reduction between March and June.
Section V. Business Shutdown, Re-Entry and Worker
Recall
So far, most of our results combine employment changes for businesses that suspend opera-
tions (whether temporarily or permanently) and businesses that continue operating. Sepa-
rating these groups is useful: a primary determinant of the speed of recovery from this crisis
may be the extent to which irreversible dis-investments occur. This question has come to
the forefront recently as employment has increased. Are business closures permanent? How
much of the employment increase has occurred as the result of businesses re-opening or firms
recalling workers that were temporarily laid-off?
Table 2 shows the decomposition of aggregate employment growth into the contributions
from: continuers (employment at firms that operated continuously since February 15), entry
(employment at firms that did not exist in our sample on February 15), shutdown (initial em-
ployment at firms that shut down at any point since February 15), and reentry (employment
17
Nine percent of workers at these firms saw wage cuts between March and June of 2020. The full histogram
of base wage changes for these firms in 2019 and 2020 are shown in the online appendix.
16
Table 2: Decomposition of Employment Growth in Shutdown and Re-Entering Businesses
New Ever Shutdown
Week Continuers Entry Shutdown Re-entry Total
2/15/2020 0.0 0.0 0.0 0.0 0.0
2/22/2020 0.5 0.1 -0.1 0.0 0.5
2/29/2020 0.1 0.3 -0.3 0.0 0.1
3/07/2020 -0.1 0.5 -0.6 0.0 -0.1
3/14/2020 -0.6 0.6 -1.0 0.1 -0.9
3/21/2020 -4.4 0.7 -1.9 0.1 -5.4
3/28/2020 -10.7 0.8 -3.4 0.1 -13.2
4/04/2020 -15.0 1.0 -4.8 0.1 -18.8
4/11/2020 -15.0 1.1 -5.7 0.2 -19.4
4/18/2020 -14.8 1.2 -6.1 0.2 -19.6
4/25/2020 -15.8 1.2 -6.4 0.5 -20.6
5/02/2020 -15.8 1.3 -6.6 0.8 -20.3
5/09/2020 -14.4 1.5 -6.8 0.9 -18.8
5/16/2020 -12.5 1.6 -6.9 1.2 -16.6
5/23/2020 -11.7 1.8 -7.0 1.6 -15.4
5/30/2020 -11.6 2.0 -7.2 2.1 -14.7
6/06/2020 -10.5 2.2 -7.4 2.4 -13.3
6/13/2020 -9.7 2.3 -7.5 2.6 -12.4
6/20/2020 -10.3 2.3 -7.9 2.7 -13.1
Notes: Table shows decomposition of total employment growth into employment contributions from continu-
ously operating firms, newly entering firms, firms that were shutdown at some point since February 15, and
firms who used to be shutdown but subsequently re-entered. Data from the business-level sample. Percentages
expressed in terms of February 15 employment.
at firms that shut down at some point since February 15 but are now open). The sum of
these four contributions equals total aggregate paid employment growth. To create this table
we use our business-level sample and report all contributions as percent of total employment
as of February 15th. We define a firm as “shutting down” if they issue no paychecks during
a week in which we would expect them to do so (given past pay frequency patterns). We
define a firm as “re-entering” if it has shut down and starts paying its workers again.
On June 20th, job losses at firms that operated continuously contributed 10.3 percentage
point out of 13.1 percent total employment decline since mid-February (as highlighted in
Figure 1).18 As of June 20th, 7.9 percent of mid-February employment was shut down at
some point and 2.7 percent of (mid-February) employment that was previously shut down
returned by June 20th. In other words, one-third of employment in firms that have ever
18
The numbers in column 1 of Table 2 differ slightly from those in Panel B of Figure 1 due to different
normalizations. In Figure 1 we normalize the series by employment in continuing firms in February 2020
while in this table we normalize the series by total employment across all firms in February 2020.
17
shutdown at some point during the first few months of the pandemic has returned by late
June. The difference between shutdown and re-entry columns in Table 2 measures the
employment in firms that have remain closed and the change in employment at re-entering
firms (the former component accounts for most of the difference). Newly entering firms
added 2.3 percent (of February levels) to employment through the late June, though we
caution against interpreting these figures in terms of genuine new business formation.19
Between April 25th and June 20th, aggregate employment increased by 7.5 percentage
points (relative to February 15th levels). About three-quarters of that growth (5.5 out of
7.4 percentage points) was due to employment gains in continuing firms, and about one-
third (2.2 out of 7.4 percentage points) was due to employment contributions of firms that
re-opened.20 The findings suggest that the re-opening of temporarily shuttered businesses
contributed meaningfully to aggregate employment gains during May and June.
Panel A of Figure 7 shows the dynamics of employment at currently shut down (solid line)
and re-entered (dashed line) businesses for all businesses during the recession. Specifically,
the solid lines in each panel measure the employment lost in currently shutdown firms during
each pay period. The dashed line shows the employment gains coming from re-entering firms.
Panel A shows the employment losses associated with business shutdown for the US economy
peaked in late-April. Since then, as highlighted in Table 2, some of these shuttered firms
have re-opened contributing to aggregate employment growth. As of late June, there is
still 4 percent of February employment in firms that are still shutdown. Notably, however,
the decline in shutdown employment from its April peak is smaller in magnitude than the
employment generated by reentry (about 2 percentage points versus nearly 4 percentage
points), reflecting the fact that shutdowns continued to occur even after mid-April.
In the Appendix, we additionally show that firm shutdown disproportionately affected
low wage workers. By the end of April, approximately three times as many bottom quintile
workers were in firms that have shut down than were top quintile workers. This partly reflects
differences in firm closure by industry: firms in the entertainment and food/accommodation
industries were most likely to shut down in our sample.
The remaining panels in Figure 7 show firm shutdown and re-entering patterns by business
size. Business shutdown was much more prominent among smaller firms, with shutdown firms
contributing about 15 percent of the initial employment decline by late April among those
19
Note that entry does not necessarily correspond to new firm formation; it could simply capture existing
firms newly contracting with ADP or firms that existed at some point in the past, were closed during February,
and later re-opened again, e.g. seasonal businesses. We do note, however, that Census Bureau data on new
business applications with planned wages described by Haltiwanger (2020) indicate that application rates
returned to their 2019 pace by early June.
20
The employment gains from newly entering businesses during May and early June and the employment
losses from businesses that newly shuttered in May and early June roughly offset each other.
18
Figure 7: Employment in Shutdown and Re-Entering Firms
0.16 0.16
Share of February 15th Employment Share of February 15th Employment
0.14 0.14
0.12 0.12
0.10 0.10
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0.00 0.00
Shut Down Re-Entered Shut Down Re-Entered
Panel A: All Panel B: 1-49 Emps
0.16 0.16
Share of February 15th Employment Share of February 15th Employment
0.14 0.14
0.12 0.12
0.10 0.10
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0.00 0.00
Shut Down Re-Entered Shut Down Re-Entered
Panel C: 50-499 Emps Panel D: 500+ Emps
Notes: Figure shows the share of February 15th employment at firms that were shut down as of
each date (solid line), and the share of February 15 employment at firms that had shut down and
then re-entered (dashed line). The sample of firms is defined as of mid-February and are followed
over time. We define shut downs to be where a firm processes no payroll.
businesses with fewer than 50 employees. However, many of these small businesses have
re-opened through late June. As seen in Figure 2, total paid employment in small businesses
increased by nearly 17 percentage points (relative to February levels) between mid April and
late June. Employment growth in firms that temporarily shuttered—i.e., shuttered and then
reopened—contributed about half of the employment gains in among businesses with less
than 50 employments since mid-April. Businesses with 50 to 499 employees saw lower, but
still notable, levels of shutdown, peaking around 5 percent of initial employment. Shutdown
has been subdued among the largest firms, though it is noteworthy that the series has not
19
Figure 8: Employee Recall in Re-Entering Firms
1 1
.8 .8
Cumulative Probability Cumulative Probability
.6 .6
.4 .4
.2 .2
0 0
0 .5 1 1.5 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
Re-Opening Employment Relative to Size Feb 1-14 Employment Share of Returning Employees Present in February 1-14
Panel A: Re-Entering Panel B: Share of Employees
Firm Size Who Were Recalled
Notes: Figure shows the firm weighted CDF of re-entering business size (Panel A) and share
of initial business employees recalled (Panel B). The sample of firms is those firms that existed
in early February 2020 and that temporarily shut down. Temporary shutdowns are those firms
that shut down and then subsequently reopened (i.e., resumed paycheck issuance). Employment
measurements are through late June.
peaked and continues to gradually increase through late June.
When businesses re-enter, they may not hire back all of their pre-existing workforce.
Figure 8 explores this possibility. Panel A plots the distribution of firm employment at
re-entry relative to the firm’s employment during early February, weighting each firm by
their initial size. The figure shows that about 60% of returning firms are smaller than they
were in the beginning of February. The median re-entering firm re-opened with 86% of
their initial employment, while the mean firm only has 78% of its initial employment. We
present versions of this figure weighting firms by their February employment in the Appendix.
That employment weighted figure shows that returning large firms disproportionately return
smaller: the employee-weighted median re-entering firm has 65% of its February employment.
Although firm re-entry is contributing to a recovery in overall employment, these re-entering
firms are operating below their initial capacity. In part, this may be due to firms allowing
individuals to return to work in stages, in order to minimize social contact in the office.
Monitoring this sub-capacity operation will be important to the overall recovery dynamics.
When businesses return, they can choose to either re-hire their prior workforce or seek
new employees. Panel B of Figure 8 shows the share of returning businesses’ workforce
that was previously employed with that same business in the first two weeks of February.
Such workers represent “recalls.” Again, the distribution this figure is weighted by initial
business size. Almost half of re-entering firms have their new workforce comprised of at
20
Figure 9: Growth of Continuing Businesses
1 1
Hiring Outside Emp Hiring Outside Emp
Shedding Initial Emp Hiring Initial Emp
.8 .8
Cumulative Probability Cumulative Probability
.6 .6
.4 .4
.2 .2
0 0
0 .2 .4 .6 .8 1 1.2 1.4 -1 -.8 -.6 -.4 -.2 0 .2 .4 .6 .8 1
Firm Size Relative to Early February Share of Employment Growth From Initial Employees
Panel A: Size Relative Panel B: Share of Growth
to February 1-14 Accounted for by Recall
Notes: Figure shows how continuing businesses have grown through the early stages of recovery. Panel A
plots the distribution of business employment in late June relative to the first two weeks of February. Panel
B plots firm-level distribution of the share of firm growth of continuing businesses accounted for by recall
of previously employed workers. Firm growth measured between the firm’s trough employment after March
11 and late June. Recall defined as hiring workers who were employed by the firm in the first two weeks of
February. Throughout we restrict attention to firms whose trough employment occurred after March 11. In
Panel B, we limit attention to firms which add at least 10 employees.
least 90% employees who worked in the firm in early February. Hardly any firms re-enter
without having their workforce comprised of at least half workers who were with the firm
in early February. The results in this figure suggest that the overwhelming majority of re-
entering businesses are seeking to avoid costly search by simply rehiring from their initial
workforce. Again, most of these businesses are still well below their initial size so as the
recovery continues they may be able to bring back more of their initial workers.
As we highlight throughout, firm shutdown has not been the only source of employment
declines at the beginning of this recession. Continuing employees have also seen enormous
employment declines followed by small employment increases over the last few months. As
these continuing firms recover, they too face a choice of whether to rehire existing employees
or seek outside employment. Panel A of Figure 9 plots the distribution of the current firm
size for firms that contracted during the beginning of the recession but then subsequently
started growing again. Specifically, we consider the growth in employment between the week
in which a continuing firm has its lowest observed employment (after some contraction) and
the final week of our sample: the end of June. We then calculate the firm’s current size
relative to their size in mid-February. The figure shows that median growing continuing firm
is currently at a size that is ten percent lower than their mid-February level. Consistent with
the patterns in Figure 3, roughly 15 percent of these growing firms are now larger than they
21
were in mid-February.
Panel B shows the share of trough-to-peak employment growth for continuing businesses
accounted for by recalling previously employed workers. For each growing continuing firm,
we calculate the share of this employment growth accounted for by growth in workers who
were employed by the firm in the first two weeks of February. Note that this share can
be negative if the business continues to shed existing workers while simultaneously hiring
new outside workers. Finally, to remove noise from small-growth firms, we consider only
continuing firms that grow by at least 10 workers from their trough to peak.21 The figure
shows that 90% of firms grow at least in part by recalling existing workers. Almost 10% of
continuing firms hired exclusively from recall.22 However, the complement of these findings
is also interesting. Almost 10% of continuing firms are growing from external hires, even
as they shed their initial workforce. Even in these uncertain times, there remains some
worker churn. The fact that workers are being reallocated among existing business during
the Pandemic Recession is consistent with the findings in Barrero et al. (2020).
Overall, firm shutdown was an important driver of employment losses at the beginning
of the Pandemic recession, and firm re-opening is likewise contributing to the labor market
recovery. However, re-entering firms operate at far below capacity, only hiring back a fraction
of their prior workforce. Although both continuing and shutdown firms principally recall their
prior employees to spur growth at this stage of the recovery, many continuing firms are also
looking toward external labor markets for their hiring.
Section VI. Employment Gains and State Re-Openings
Figure 10 explores the effects of states re-opening certain sectors on employment. To facilitate
exposition, we create two groups of states - a set of large states that broadly opened in late
April or early May and a set of large states that broadly opened in late May and early June.23
For the first set of states opening early we pool together data from Florida, Georgia and
Texas. These states opened restaurants and lifted stay at home orders between April 24th
and May 4th. For the second set of states opening later we pool together data from Illinois,
Pennsylvania, Virginia and Washington. The late states opened restaurants and lifted stay
21
We present the distribution of these recall shares for firms that grow by at least 1 or 5 employees in the
appendix.
22
These findings are broadly consistent with the results in Fujita and Moscarini (2017) showing the im-
portance of employee recall in prior recessions.
23
We focus on large states because there is less noise in employment fluctuations at the state-by-industry
level within the ADP data. We use our employee sample for this analysis so we can measure state of residence.
There are small differences in aggregate employment declines by sector between our business sample and our
employee sample given the slightly different sampling frames.
22
Figure 10: State Reopening and Employment: Accommodation/Restaurant Sector
1.10
Employment Change Relative to February 1st
1.00
0.90
0.80
0.70
0.60
0.50
0.40
1/25/20 2/1/20 2/8/20 2/15/20 2/22/20 2/29/20 3/7/20 3/14/20 3/21/20 3/28/20 4/4/20 4/11/20 4/18/20 4/25/20 5/2/20 5/9/20 5/16/20 5/23/20 5/30/20 6/6/20 6/13/20 6/20/20
Large States Opening Late (IL, PA, VA, WA) Large States Open Early (FL, GA, TX)
Notes: Figure plots employment in NAICS Industry 72 (Accommodation and Food Service) in a
set of large states that opened in late April/early May (FL, GA, TX) (dashed line) vs a set of
large states that opened in late May/early June (IL, PA, VA, WA) (solid line). Data come from
the ADP employee sample.
at home orders after May 31st. Our results focus on one sector where reopening had the
most direct effect: the Food and Accommodation sector (NAICS 72).
The figure shows that employment in Restaurants and Accommodations fell similarly
through Mid-April in both state groupings. Starting in late April, employment in this sector
within the states opening early increased faster than employment in the states opening later.
The states start to to diverge during the week of April 18th which was a week prior to the
Georgia re-opening of in person dining. Given that the state openings were announced in
advance, firms started ramping up some employment prior to the actual date of opening.
This qualitative pattern is not overly surprising.
The quantitative patterns are, however, noteworthy. First, even in the states that opened
early, employment in this sector is still over 20 percent below February levels as of late-June.
Opening, per se, does not guarantee employment will fully rebound in these sectors. If
individuals are concerned about contracting the virus in public places, the demand for these
types of services may remain depressed even as these sectors start to re-open. Second,
employment in these sectors within states that opened late started to increase even prior to
those states re-opening. The increase was modest but suggests that demand was increasing
(perhaps for takeout meals) even prior to official re-openings. These demand effects could
interact with disease trends within the state that could also prompt states to lift stay at home
orders. Additionally, the expectation of re-opening likely resulted in some firms bringing
back workers early to prepare for serving in person customers. Researchers seeking to attach
23
a causal quantitative interpretation of employment gains associated with state reopening
should do so cautiously. Finally, employment in this sector has almost converged between the
two groups of states as of late-June. Again, this suggests that once states reopen employment
in previously constrained sectors will rise but demand forces will still prevent employment
from returning to pre-recession levels.
Section VII. Conclusion
In this paper, we use high-frequency payroll data from ADP to track the behavior of the
labor market in early part of the Pandemic Recession. The data show an unprecedented
collapse in employment from mid-February through late April with employment falling by
a 21 percent relative to early February levels. As states started to reopen, employment
rebounded partially. As of late June, employment is still 13 percent below February levels.
The employment declines as of late June are massive relative to past recessions; during the
Great Recession employment troughed at 7 percent of pre-recessionary levels. Despite the
rebound, not only is the U.S. currently experiencing the worst employment losses since the
Great Depression, but the pace of job gains has slowed measurably.
Our results highlight that the employment losses were disproportionately concentrated
among smaller firms and lower wage workers. Much of the fiscal stimulus implemented during
the early part of the recession was targeted towards these groups. Job losses, in percentage
terms, have converged between smaller and larger businesses as of late June (relative to pre-
recession levels). Many previously shuttered businesses—particularly smaller businesses—
have re-opened during May and June bringing back laid off workers. This could be consistent
with government stimulus provided through programs like the Paycheck Protection Program
(PPP) allowing some smaller businesses to survive the beginning of the recession during the
mandated shutdowns. However, further research will be needed to try to causally isolate the
effects of PPP on small business employment.
One finding of ours that needs to be monitored going forward is how the Pandemic
Recession is affecting wages of workers who did not get displaced. During the first few
months of the recession, we have shown that many workers are receiving nominal cuts to
their contracted wage while many others are receiving pay freezes. The extent of nominal
wage cuts and wage freezes are large relative to non-recessionary years and are even larger
than what was observed in the Great Recession. How broader measures of compensation
adjust—including components such as bonuses, performance pay and fringe benefits-—is
worth monitoring as the recession continues. Our initial findings suggest that both the
employment adjustments and the wage adjustments are large relative to prior recessions.
24
References
Barrero, Jose Maria, Nicholas Bloom, and Steven J Davis, “COVID-19 Is Also a
Reallocation Shock,” Working Paper 27137, National Bureau of Economic Research May
2020.
Bartik, Alexander W., Marianne Bertrand, Feng Lin, Jesse Rothstein, and Matt
Unrath, “Early Impacts of COVID on the Labor Market,” mimeo 2020.
Bartik, Alexander W, Marianne Bertrand, Zoë B Cullen, Edward L Glaeser,
Michael Luca, and Christopher T Stanton, “How Are Small Businesses Adjusting
to COVID-19? Early Evidence from a Survey,” Working Paper 26989, National Bureau
of Economic Research April 2020.
Bick, Alexander and Adam Blandin, “Real-Time Labor Market Estimates During the
2020 Coronavirus Outbreak,” Working Paper June 2020.
Brynjolfsson, Erik, John J Horton, Adam Ozimek, Daniel Rock, Garima Sharma,
and Hong-Yi TuYe, “COVID-19 and Remote Work: An Early Look at US Data,”
Working Paper 27344, National Bureau of Economic Research June 2020.
Cajner, Tomaz, Leland Crane, Ryan A. Decker, Adrian Hamins-Puertolas,
Christopher Kurz, and Tyler Radler, “Using Payroll Processor Microdata to Measure
Aggregate Labor Market Activity,” FEDS Working Paper 2018-005, Board of Governors
of the Federal Reserve System (U.S.) January 2018.
, Leland D. Crane, Ryan A. Decker, Adrian Hamins-Puertolas, and Christo-
pher Kurz, “Improving the Accuracy of Economic Measurement with Multiple Data
Sources: The Case of Payroll Employment Data,” in Katharine G. Abraham, Ron S.
Jarmin, Brian Moyer, and Matthew D. Shapiro, eds., Big Data for 21st Century Eco-
nomic Statistics, University of Chicago Press, July 2020.
Chetty, Raj, John N. Friedman, Nathaniel Hendren, Michael Stepner, and the
Opportunity Insights Team, “How Did COVID-19 and Stabilization Policies Affect
Spending and Employment? A New Real Time Economic Tracker Based on Private Sector
Data,” Working Paper June 2020.
Coibion, Olivier, Yuriy Gorodnichenko, and Michael Weber, “Labor Markets During
the COVID-19 Crisis: A Preliminary View,” Working Paper 27017, National Bureau of
Economic Research April 2020.
Dingel, Jonathan I and Brent Neiman, “How Many Jobs Can be Done at Home?,”
Working Paper 26948, National Bureau of Economic Research April 2020.
Fujita, Shigeru and Giuseppe Moscarini, “Recall and Unemployment,” American Eco-
nomic Review, December 2017, 107 (12), 3875–3916.
Grigsby, John, “Skill Heterogeneity and Aggregate Labor Market Dynamics,” Working
Paper, University of Chicago 2019.
25
, Erik Hurst, and Ahu Yildirmaz, “Aggregate Nominal Wage Adjustments: New
Evidence from Administrative Payroll Data,” Working Paper 25628, National Bureau of
Economic Research March 2019.
Haltiwanger, John, “Applications for New Businesses Contract Sharply in Recent Weeks:
A First Look at the Weekly Business Formation Statistics,” mimeo, University of Maryland
2020.
Hurst, Erik and Benjamin Pugsley, “What Do Small Businesses Do?,” Brookings Papers
on Economic Activity, 05 2011, Spring.
Kahn, Lisa B, Fabian Lange, and David G Wiczer, “Labor Demand in the Time
of COVID-19: Evidence from Vacancy Postings and UI Claims,” Working Paper 27061,
National Bureau of Economic Research April 2020.
Kurmann, Andre, Etienne Lale, and Lien Ta, “The Impact of COVID-19 on U.S.
Employment and Hours: Real-Time Estimates with Homebase Data,” mimeo 2020.
Mongey, Simon, Laura Pilossoph, and Alex Weinberg, “Which Workers Bear the
Burden of Social Distancing Policies,” BFI Working Paper 2020-51, Becker Friedman
Institute Working Paper April 2020.
Solon, Gary, Robert Barsky, and Jonathan Parker, “Measuring the Cyclicality of Real
Wages: How Important is Composition Bias,” Quarterly Journal of Economics, 1994, 109
(1), 1–25.
26
Online Appendix:
“The U.S. Labor Market during the Beginning of the
Pandemic Recession”
(Not for Publication)
Appendix A Data Description
We use anonymized administrative data provided by ADP. ADP is a large international
provider of human resources services including payroll processing, benefits management, tax
services, and compliance. ADP has more than 810,000 clients worldwide and now processes
payroll for over 26 million individual workers in the United States per month. The data
allow us to produce a variety of metrics to measure high-frequency labor market changes for
a large segment of the U.S. workforce.
Subsection I. Business Level Data
We use two separate data sets to measure high-frequency labor market changes. In this
section we introduce a business-level data set, while the subsequent section covers a worker-
level data set.24 The business-level data set reports payroll information during each pay
period. Each business’ record is updated at the end of every pay period. The record consists
of the date payroll was processed, employment information for the pay period, and many
time-invariant business characteristics such as NAICS industry code.25 Business records
include both the number of individuals employed (“active” employees) and the number
of paychecks issued in a given pay period (“paid” employees). Active employees include
wage earners with no hours in the pay period, workers on unpaid leave, workers who are
temporarily laid-off and the like. Paid employees include any wage or salary workers issued
regular paychecks during the pay period as well as those issued bonus checks or any other
payments.
24
When accessing the microdata, we follow a number of procedures to ensure confidentiality. Business
names are not present in the data.
25
Note that we use the term “business” throughout the paper to denote ADP clients. Often, entire
businesses contract with ADP. However, sometimes establishments or units within a firm contract separately.
The notion of business in our data is therefore a mix of Census Bureau notions of an establishment (i.e., a
single operating business location) and a business (i.e., a collection of establishments under unified operational
control or ownership).
27
The data begin in July 1999 but are available at a weekly frequency only since July 2009.
As shown in Cajner et al. (2018), ADP payroll data appear to be quite representative of
the U.S. economy, though the data modestly overrepresent the manufacturing sector and
large businesses (as compared to the SUSB universe of firms). We address these issues by
reweighting the data as explained below. The process of transforming the raw data into
usable aggregate series is complex, and we refer the interested reader to Cajner et al. (2018)
for details of the creation of the ADP-Federal Reserve Board (ADP-FRB) high frequency
employment series for additional information. In short, for businesses that do not process
payroll every week (for example, businesses whose workers are paid biweekly), we create
weekly data by assuming the payroll in the missing intermediate period is what is observed in
the next period the business processes payroll. We build a weekly time series of employment
for each business, estimating employment at the business each Saturday.26
In our baseline analysis we treat ADP payroll units as firms for weighting purposes. As
a result, we use 2017 SUSB employment counts by firm size and two-digit NAICS as the
target population (2017 is the latest year available). We use six size bins defined by 1-19,
20-49, 50-99, 100-299, 300-499, and 500 or more employees.27 For the analysis in this paper,
we keep the weights fixed throughout the COVID-19 period.28
Since the primary focus of this paper is on weekly data, it is worth noting the distribution
of pay frequencies in the ADP data. As of March 2017, 22 percent of ADP clients were issuing
paychecks weekly, 46 percent biweekly, 21 percent semi-monthly, and 11 percent monthly
(in terms of employment, these shares are 23 percent, 55 percent, 18 percent, and 4 percent,
26
Technically, the employment concept is business employment for the pay period that includes the Sat-
urday in question, as we cannot observe changes within pay period. Lacking any information on events
within a pay period, we assume that businesses adjust their employment discretely at the beginning of each
pay period and that employment is constant within the pay period. This assumption is consistent with the
typical practice of human resource departments, according to which job start dates often coincide with the
beginning of pay periods. It is also analogous to the CES methodology, which asks for employment for the
pay period including the 12th of the month.
27
We also conducted all reported exercises using a weighting scheme with one additional bin at 1000+
employees. This makes only modest differences for the results (and only affects firms in the “large” size
category): at most, total employment for large firms differs by less than 2 percent of February 15 employment,
with most weeks (including the trough week and the most recent week) differing by less than 1 percent.
For overall (all-sizes) employment, the trough and most recent weeks differ by less than 0.5 percent. The
employment series become more volatile under the 7-bin weighting scheme, however, so we prefer the 6-bin
scheme.
28
Formally, let wj be the ratio of SUSB employment in a size-industry cell j to employment from ADP
data in cell j, where SUSB employment are from 2017 (the latest year available) and ADP employment
are fixed in the week ending February 15 for weighting purposes. Then weighted employment for any firm
i in cell j is given by wj ei,j,t , where e is firm employment. We calculate aggregate employment declines
based on the change in total weighted employment. Exercises using percentile changes calculate firm-level
growth rates then evaluate percentiles on the full employment-weighted distribution where employment is
SUSB-weighted employment.
28
respectively). These fractions are not far from what the BLS reports.29
Finally, it is worth noting that we only measure employment declines once we observe
a business’s regularly scheduled payroll. This can mean that there is some lag in our mea-
surement. For example, suppose a business pays all of its workers biweekly. We will observe
the business’s payroll in week t and then again in week t + 2. Suppose the business lets 20
percent of its workers go in week t + 1. We would not be able to infer this paid employment
decline until week t + 4, since those workers worked some in the t + 2 pay period. Given that
the payroll would be missing in t + 4, we attribute the job loss occurring in t + 2. All of this
is to say that our measurement may, at times, be shifted a week or two relative to when a
hire or separation took place. This is part of our motivation for focusing on the pay period
employment concept, discussed above.
Subsection II. Worker Level Data
The business-level data reports payroll aggregates for each business. For a very large subset
of businesses, we also have access to their anonymized de-identified individual-level employee
data.30 That is, we can see detailed anonymized payroll data for individual workers. As with
the business data, all identifying characteristics (names, addresses, etc.) are omitted from our
research files. Workers are provided an anonymized unique identifier by ADP so that workers
may be followed over time. We observe various additional demographic characteristics such
as the worker’s age, gender, tenure at the business and residential state location. We also
can match the workers to their employer. As with the business-level data described above,
we can observe the industry and business size of their employers.
The benefits of the employee data relative to the business data described above are three-
fold. First, we can explore employment trends by worker characteristics such as age, gender,
and initial wage levels. This allows us to discuss the distributional effects of the current
recession across different types of workers. Second, the individual-level data allow us to
measure additional labor market outcomes such as wages per worker as well as recall rates
of a given worker as businesses start to re-open. Finally, the individual level data allows us
to measure the state where a worker livers allowing us to compute high frequency local labor
market measures as the economy recovers.31
29
See BLS (2019) “Length of pay periods in the Current Employment Statistics survey.”
30
The data for our employee sample skew towards employees working in businesses with at least 50 em-
ployees. This is the same data used in Grigsby et al. (2019). While the data come from employees mostly in
businesses with more than 50 employees, there is representation in this data for employees throughout the
business size distribution. Again, we weight these data so that it matches aggregate employment patterns
by industry and business size.
31
The business level data set tracks the location of the firm. However, for larger firms, this is often the
location of the headquarters and not the local establishment.
29
The individual-level data allows us to observe the worker’s contractually obligated pay
rate as well as their gross earnings during the pay period. For hourly workers, the per-period
contract pay rate is simply the worker’s base hourly wage. For salaried workers, the per-
period contract rate constitutes the pay that the worker is contractually obligated to receive
each pay period (e.g., weekly, biweekly, or monthly). For workers who are paid hourly, we
also have administrative records of how many hours they worked during the pay period. For
workers who are salaried, the hours are almost always set to 40 hours per week for full-time
workers and some fraction of 40 hours per week for part-time workers. For example, workers
who are half-time are usually set to 20 hours per week. As a result, the hours for salaried
workers are more indicative of full-time status than actual hours worked.
When reporting hours, employment, and wage statistics using the employee-level sample,
we also weight the data to ensure that it is representative of the U.S. population by 2-digit
industry and business size. To create the weights for this part of our analysis, we use data
from the U.S. Census’ 2017 release of the Statistics of US Businesses. Specifically, we weight
the ADP data so that it matches the share of businesses by 2-digit NAICS industry and
business size. As highlighted in Grigsby et al. (2019) the weighted employee-level data is
representative of the U.S. labor market on many dimensions.
To construct employment indices, we exploit the high-frequency nature of the ADP data.
To facilitate our measurement using the employee data, we limit our attention to workers
paid weekly or biweekly for these analyses to avoid time aggregation issues. These account
for about 80 percent of all employees in our employee sample. Unsurprisingly, this is nearly
identical to the share of weekly and biweekly employees in the business-level sample described
above.32 Biweekly workers are generally paid either on every even week (e.g. the 4th, 6th,
and 8th week of the year) or on every odd week. We designate biweekly workers to be “even
biweekly” workers if their regularly scheduled paychecks are disbursed on even weeks, or “odd
biweekly” workers if their regularly scheduled paychecks are disbursed on odd weeks. We
then sum all paychecks—earnings and hours—in a two-week period to the nearest subsequent
even week for even biweekly workers, and the nearest subsequent odd week for odd biweekly
workers. We additionally sum all paychecks in a given week for all weekly workers. The
result of this is an individual-by-week panel. We then produce separate indices for weekly,
biweekly-even, and biweekly-odd employees and then combine the indices into an aggregate
employment index. We use these indices when computing employment changes by worker
characteristics (age, sex, worker location, and wage percentile). We compute hours and
wage indices similarly. However, the panel nature of our data allows us to make indices for
32
Our preliminary evidence suggests that workers paid semi-monthly look nearly identical to those paid
bi-weekly.
30
hours worked and wages following a given worker over time. This allows us to control for
the changing selection of the workforce at the aggregate level over this period. To account
for most of our data sample is paid bi-weekly, we lag the employment measure from the
employee sample by two-weeks. For example, the payroll week of February 15th measures
employees who worked during the week of February 1st.
Appendix B Additional Results
In this section, we show various other results from our analysis. We start by documenting
gross job creation and job destruction rates from our employee sample. Next we show trends
in hours worked for continuing workers. We then explore patterns by age. We also highlight
trends in employment by business size using our employee data where we can control for the
timing of pay-check receipt exactly. We then discuss our regressions explaining differences
in employment changes by wage quintile and gender. Finally, we conclude with additional
results associated with firm closures, firm openings and worker recall.
Subsection I. Historical Comparison of ADP Data to BLS CES
ADP data tracked the last recession remarkably well; Appendix Figure A1 compares the
monthly change in employment in the unbenchmarked ADP-FRB series (constructed by
Cajner et al. (2018)) to the Bureau of Labor Statistics (BLS) Current Employment Statistics
(CES) series from January 2006 through February 2020. The two series pick up the same
underlying signal—aggregate U.S. payroll growth.
Figure A1: Historical Monthly Change in Private Payroll Employment: ADP-FRB and CES
400
200
Thousands of Jobs
0
-200
-400
-600
CES
-800 ADP-FRB active employment,
not benchmarked
-1000
Jan2006 Jan2008 Jan2010 Jan2012 Jan2014 Jan2016 Jan2018 Jan2020
Notes: Source CES, ADP, and Cajner et al. (2018). CES data benchmarked to the QCEW.
31
Table A1: Aggregate Employment Patterns under Two Weighting Schemes
SUSB QCEW
Total paid employment change, February 15 to April 25 -20.6% -22.5%
Share of decline contributed by business closure 0.18 0.27
Total paid employment gain, April 25 to June 20 7.4% 10.3%
Notes: Table shows aggregate employment patterns as implied by two different weighting schemes.
The SUSB column reports the main results of the paper, which rely on firm-based weights from the
Census Bureau’s 2017 Statistics of U.S. Businesses. The QCEW column reports alternative results
treating ADP business units as establishments and weights from the March 2019 BLS Quarterly
Census of Employment and Wages. All figures expressed as percents of February 15 employment.
Subsection II. The Importance of Weights
The aggregate numbers shown in Section Section II. are based on the ADP sample with
SUSB weights to ensure representativeness in terms of industry and firm size. As noted in
our data description, however, some business units in ADP may be more akin to establish-
ments (i.e., single operating locations that may be part of a larger firm) than to firms. An
alternative approach to mapping ADP data to the U.S. business universe would be to treat
ADP business units as establishments and then apply establishment-based weights; indeed,
some previous work with ADP data takes this approach (e.g., Cajner et al. (2020)). Ap-
pendix Table A1 compares aggregate employment patterns under SUSB weights (i.e., those
used in our main results in Section Section II.) and QCEW weights (i.e., those that treat
ADP businesses as establishments rather than firms). Aggregate series based on QCEW
weights show a somewhat larger decline (and rebound) in aggregate employment; the reason
is that establishment-based weights have more activity in smaller units than do firm-based
weights, and as we documented above, smaller ADP businesses have seen a deeper de-
cline (and stronger rebound) in employment than have larger businesses.33 Moreover, the
QCEW weighting scheme produces a stronger post-trough rebound in total employment,
again consistent with the stronger rebound observed among smaller ADP units that have
larger weights in QCEW. As a result, the two weighting schemes suggest a similar total
decline from February 15 through June 20.
Subsection III. Hours Worked For Continuing Workers
Appendix Figure A2 shows the decline in hours of continuing hourly workers during the
beginning of the recession using the ADP employee sample. We create an index of hours
33
The result that weights matter in our data differs from Chetty et al. (2020) who find that the private
sector data samples they work with track relevant national benchmarks without reweighting.
32
Figure A2: Hours Worked Index for Continuing Hourly Workers
1.10
1.08
1.06
U.S. Hours Relative to Feb 15th
1.04
1.02
1.00
0.98
0.96
0.94
0.92
0.90
1/25/20 2/1/20 2/8/20 2/15/20 2/22/20 2/29/20 3/7/20 3/14/20 3/21/20 3/28/20 4/4/20 4/11/20 4/18/20 4/25/20 5/2/20 5/9/20 5/16/20 5/23/20 5/30/20 6/6/20 6/13/20 6/20/20
Hours Continuing Hourly Workers
Notes: Figure shows the change in hours worked for hourly workers who remain continuously
employed between pay periods. We create a chain weighted index of the hours changes. The index
is relative to hours worked during the week of February 15th. The data come from the employee
sample and are weighted such that the sample is representative by business size crossed with 2-digit
NAICS industry.
of continuing hourly workers relative to February 15th. There has been little adjustment of
hours worked for hourly workers throughout the recession. Essentially all of the employment
adjustment has occurred on the extensive margin of labor supply.
Subsection IV. Employment Declines By Age
Appendix Figure A3 plots paid employment changes by age bin using our employee sample.
Employment changes are relative to February 15th. The figure shows an inverted-U between
age and employment declines through late April. The youngest workers—those between the
ages of 21 and 30—and the oldest workers—those 60+—saw the largest employment declines
in the first two months of the recession. Young workers, however, have seen their employment
recover more sharply than any other group betweeen late April and late June. All age groups
from 21-60 years old have employment which is approximately 10 percentage points below
their February levels, while workers over 60 had employment that is 15 percentage points
below February as of the end of June.
33
Figure A3: Employment Changes by Age
1.05
1.00
Employment Change Relative to February 1st
0.95
0.90
0.85
0.80
0.75
0.70
0.65
0.60
0.55
2/1/20 2/8/20 2/15/20 2/22/20 2/29/20 3/7/20 3/14/20 3/21/20 3/28/20 4/4/20 4/11/20 4/18/20 4/25/20 5/2/20 5/9/20 5/16/20 5/23/20 5/30/20 6/6/20 6/13/20 6/20/20
21-30 31-40 41-50 51-60 60+
Notes: Figure shows paid employment declines by different age ranges. All changes relative to
February 15th, 2020. Employee sample is used for this analysis. Data are weighted so that the
sample matches aggregate employment shares by 2-digit industry cross business size.
Subsection V. Controlling for Industry in Explaining Different
Employment Declines Across Demographic Groups
To assess whether differential exposure to the recession by business characteristics (industry
and business size) or worker characteristics (age and location) can explain the differential
pattern across either gender or the wage distribution, we exploit the panel nature of our data
and estimate the following linear probability model with OLS:
Eijt = αq(i)t + βt Xijt + ijt (A1)
where Eijt is an indicator equal to one if worker i is employed by firm j in the first two
weeks of month t, and q(i) is the base wage quintile of worker i as of the first two weeks of
February. The quintile-by-month fixed effect αq(i)t captures the employment probability of
quintile q workers in month t. We include observable business- and worker-level controls Xijt
in some regressions, and allow the relationship between Xijt and employment probabilities
to differ in each month.
Our sample for this regression is the set of workers, paid either weekly or biweekly, who
are ever employed by an ADP firm from February-May. Because of this, the estimated
coefficients may be interpreted as capturing workers’ differential separation probabilities,
adjusted by the probability of returning to employment after a job loss. As with the analysis
34
in Figure 4 of the main text, we define the quintiles in early March based on the aggregate
wage distribution within the ADP sample and hold the quintile boundaries fixed when sorting
workers into the quintiles during early February.
We control for worker i’s wage quintile at the beginning of the period to allow the baseline
separation rate to differ for workers in different quintiles q. Our variable of interest is how
the employment probabilities of each quintile change during the beginning of the recession.
This is captured by the coefficients αqt . We then ask how these αqt coefficients change as we
include various business and worker controls, Xijt .
Appendix Figure A4 plots the αqt coefficients across various specifications of equation
(A1). Panel A shows the estimated coefficients for March – the baseline separation probabil-
ities – while Panels B and C show the employment probabilities in April and May. The black
points (circles) show our estimates including no additional Xijt controls, along with a 95%
confidence interval using standard errors clustered at the 3-digit NAICS-by-firm size level.
We omit the coefficient from quintile 5 (the top wage quintile). As a result, all coefficients
should be interpreted as the employment declines in quintile q relative to the decline of quin-
tile 5. The baseline separation probability between February and March is approximately
6 percentage points higher for bottom quintile earners than that of top quintile earners.
After controlling for wage quintile fixed effects, bottom quintile earners were 21.5 percentage
points less likely to be employed in April than were top quintile earners, reflecting the pat-
terns in Figure 4. In May, bottom quintile workers were 37.2 percentage points less likely to
be employed than top quintile workers.34 The employment probabilities rise monotonically
throughout the base wage distribution.
Recent research by Mongey et al. (2020) suggests that low-income workers tend to work
in “social” sectors and the large decline observed in these sectors will result in job loss
concentrated among lower wage workers. The gray points (triangles) explore this hypothesis
by introducing firm size and 2-digit NAICS industry fixed effects as additional regressors.
Including industry fixed effects reduces the gap in excess separation rates between bottom
quintile earners and top quintile earners in April and May only slightly to 19.1 and 33.1
percent, respectively. Therefore, a differential size and industry mix can explain 12.2%
(April) and 11.0% (May) of the gap in job loss between low-wage and high-wage workers
during the beginning of this recession, but a substantial gap remains even after accounting
for industrial composition. The explanatory power is primarily embedded in the industry
fixed effects
34
The results in Table A4 are not directly comparable to those in Figure 4 given that we are focusing on
a balanced panel of businesses. Some of the additional employment decline we highlight in Figure 4 is due
to business shutdown.
35
Figure A4: Probability of Employment by Wage Quintile and Sex, Conditional on Observ-
ables
0
Employment Probability in March,
Employment Probability in April,
Relative to Top Quintile Relative to Top Quintile
-.06 -.04 -.02 0 -.2 -.15 -.1 -.05
-.08 -.25
1st 2nd 3rd 4th 1st 2nd 3rd 4th
Wage Quintile Wage Quintile
No Controls Add Ind and Size FE No Controls Add Ind and Size FE
Add Demographic Cont. Add State FE Add Demographic Cont. Add State FE
Panel A: March Employment Panel B: April Employment
by Wage Quintiles by Wage Quintiles
0 0
Employment Probability in May, Female Employment Probability,
Relative to Top Quintile
Relative to Men
-.3 -.2 -.1
-.04 -.02
-.4
-.06
1st 2nd 3rd 4th
Wage Quintile April May
No Controls Add Ind and Size FE No Controls Add Ind and Size FE
Add Demographic Cont. Add State FE Add Demographic Cont. Add State FE
Panel C: May Employment Panel D: Female Employment
by Wage Quintiles
Notes: Figure plots estimated employment probabilities by wage quintile (Panels A-C) and sex (Panel D)
conditional on observable characteristics. Panels A-C plots the estimated αqt from equation (A1) for March,
April and May, respectively. The omitted category is top quintile workers. Base wages defined according
to the distribution of wages in the first two weeks of February. Panel D plots the γt for April and May, as
estimated from equation (A2), and shows the employment probability of women relative to men. The black
lines (circles) show estimates with no controls. The gray lines (triangles) control for firm size and 2-digit
NAICS industry fixed effects. The blue lines (squares) add controls for age and (for Panels A-C) sex. The
read lines (diamonds) also control for state of residence. Error bars report 95% confidence interval using
standard errors clustered at the 3-digit NAICS cross firm size level.
The blue dashed points (squares) additionally include fixed effects for worker demo-
graphics; namely, 5-year age bins and gender. This reduces the gap in excess separation
probabilities between low-wage and high-wage workers to 16.4 (April) and 29.1 (May) per-
cent. As seen below, younger workers and women were more likely to be displaced, and
younger workers systematically have lower wages. This additional reduction in excess sepa-
36
rations suggests that the differential age and industry composition of low-wage workers can
jointly explain between one-fifth and one-quarter of the gap between low- and high-wage
worker employment behavior during the early stages of the Pandemic Recession. Finally,
the column with the red markers (diamonds) include state fixed effects. Doing so reinflates
the gap between top and bottom quintile workers to 17.2 (April) and 30.1 (May) percent,
suggesting that low-wage workers are disproportionately in states which do not have large
employment declines.
Finally, we repeat the exercise by sex, estimating the following linear probability model:
Eijt = γt · F emalei + θt Xijt + ηijt . (A2)
Here, γt represents the employment probability of women relative to men in month t,
after controlling for observables Xijt . These γt are plotted in Panel D of Appendix Figure
A4. The figure shows that including controls for industry, firm size, age, or location do not
meaningfully affect the result that women were less likely to be employed in April and May
than men.
Subsection VI. Additional Results on Nominal Wage Adjustments
Appendix Figure A5 shows the histogram of wage changes for workers in firms that histor-
ically provide annual wage changes to their employees in March - June. Specifically, the
sample includes any worker in a firm that did at least 75 percent of their 2019 base wage
adjustments between March and June. Panel A shows the unconditional distribution of
wage changes in 2019 while Panel B shows unconditional distribution of wage changes in
2020. Panels C and D show the corresponding distributions conditional on a wage change
occurring. As seen in the main text, almost 60 percent of workers received a nominal wage
change in these firms during 2019 while only 40 percent of workers got a wage change in
these firms during the same months in 2020. About one-fifth of the wage changes in 2020,
conditional on a wage change occurring, were negative (Panel D).
It is worth discussing in more detail our concept of nominal wage changes used in the
above figure and in Section Section IV. of the main text. The ADP data measures many
forms of worker compensation. We focus on measures of a worker’s contract – or base –
wage. A worker’s base wage is their contracted hourly wage (if the worker is paid hourly)
or their contracted earnings per pay period (if the worker is paid salary). Specifically, a
salaried worker’s base earnings per pay period is their contracted annual salary divided by
the number of pay-periods per year. For example, for salaried workers that are paid bi-
weekly, it is their contracted base annual salary divided by 26. The notion of base wage
37
Figure A5: Distribution of Base Wage Changes for Continuing Workers Over Time
Sample: Workers at Firms that Usually Adjust Wages in March - June
60 60
50 50
40 40
Percent Percent
30 30
20 20
10 10
0 0
-50 -40 -30 -20 -10 0 10 20 30 40 50 -50 -40 -30 -20 -10 0 10 20 30 40 50
March-June 2019 Base Wage Change (%) March-June 2020 Base Wage Change (%)
0.7% had wage decline 9.3% had wage decline
Panel A: 2019 (Unconditional) Panel B: 2020 (Unconditional)
40 40
30 30
Percent Percent
20 20
10 10
0 0
-50 -40 -30 -20 -10 0 10 20 30 40 50 -50 -40 -30 -20 -10 0 10 20 30 40 50
March-June 2019 Base Wage Change (%) March-June 2020 Base Wage Change (%)
0.7% had wage decline 9.3% had wage decline
Panel C: 2019 (Conditional) Panel D: 2020 (Conditional)
Notes: Figure shows distribution of base wage change for continuously employed workers. Sample
is restricted to firms that made 75 percent of their annual wage changes for their employees in
2019 during March, April, May and June. Panel A and B show the unconditional wage changes
for workers in those firms between March and June 2019 and between March and June 2020,
respectively. Panels C and D show the corresponding distributions conditional on a non-zero wage
change. The data we show here are the weighted distribution of wage changes and come from the
employee sample.
is distinct from a workers gross earnings. A worker’s gross earnings can include variation
in hours worked (for workers paid hourly) as well as other forms of compensation including
overtime premiums, bonuses, commissions and performance pay. As highlighted in Grigsby
et al. (2019), over 95% of compensation for the median worker comes from base earnings.
38
Figure A6: Firm Shutdown and Re-Entry by Employee Wage Quintile
Share of Initial Quintile Employment in Exited Firms
.5
Share of Quintile Employment
in Exited Firms that Have Returned
0 .02 .04 .06 .08
.1 .2 .3 .4
0
Feb1 Mar1 Apr1 May1 Jun1 Jul1 Feb1 Mar1 Apr1 May1 Jun1 Jul1
1st 2nd 3rd 4th 5th 1st 2nd 3rd 4th 5th
Panel A: Employment Lost Through Panel B: Share of Shutdown Employment
Firm Shutdown Re-called by Re-Entry
Notes: Figure plots the share of employment lost to firm shutdown (Panel A) and subsequently recalled
through firm re-entry (Panel B) by worker base wage quintile. Data from our employee sample.
Subsection VII. Additional Results on Firm Shutdown, Firm-Rentry,
and Worker Recall
In this section we detail additional results on firm shutdown and re-entry. Throughout this
section, as in the main text, we restrict attention to businesses that existed in the ADP data
as of the first two weeks of February. Appendix Table A2 reports business closures and re-
opening by 2-digit NAICS industry using the business-level data. This is the analog of Table
2 in the main text but broken down by industry instead of the aggregate economy. Column 1
presents the share of February employment that was employed in a business in that industry
that closed at any point during our sample period (through June 20). Sectors that were hit
particularly hard, such as Accommodation and Food Services and Arts, Entertainment and
Recreation had a large number of employees in businesses that closed at some point during
the early period of the Pandemic Recession.
Column 2 shows the share of employment lost to business closure by industry that has
returned by the end of June, as businesses re-opened. Some sectors have seen sharp rebounds
from business re-opening. For instance, 38.8% of the employment lost to closure in the
Accommodation and Food Services has returned as of the end of June. In contrast, just
6.7% of the employment lost to closure in Mining sector has returned, in part due to the
confluence of pandemic-related shutdowns and a collapse in oil and gas prices.
As a result of these industry differences, business closure has disproportionately affected
workers at different points in the wage distribution. Appendix Figure A6 shows the time
series of the share of February employment lost to business closure by base wage quintile
39
Table A2: Employment Lost and Recovered through Firm Shutdown and Re-Entry, by 2-
digit NAICS Industry
Share of Feb Employment Share of Exited Employment
2-digit NAICS Industry in Firms that Exited that has Re-entered
by June 20 by June 20
72: Food/Accomodation 14.8 38.8
81: Other Services 14.3 48.1
71: Arts/Entertainment 12.5 32.2
21: Mining 6.8 6.7
23: Construction 6.6 39.9
51: Information 6.4 18.9
11: Agriculture 6.2 39.5
62: Health Care 6.0 42.8
44-45: Retail Trade 6.0 34.7
61: Education 5.9 31.5
42: Wholesale Trade 5.2 24.8
48-49: Transportation/Warehousing 4.9 11.9
53: Real Estate 4.4 27.2
54: Professional Services 4.3 28.8
56: Admin/Support Services 4.2 15.7
31-33: Manufacturing 3.6 27.4
55: Management of Companies 3.1 7.0
52: Finance/Insurance 2.4 15.9
22: Utilities 0.8 18.2
Notes: Table plots the employment losses and gains from shutdown and re-entry by 2-digit NAICS industry.
Column (1) shows the share of February 15 employment in firms that ever close through June 20. Column
(2) shows the share of this lost employment that has been recovered through re-entry by the end of May.
using our employee sample. As in the main text, we define wage quintile cutoffs as of the
first two weeks of February. The solid dark blue plots the patterns for the bottom quintile
of workers. This line in Panel A shows that over 8% of February’s bottom quintile workers
were in firms that had closed by the end of April.35 By contrast, only 3% of February’s top
quintile workers were in firms that exited by the end of June.
Panel B reports the share of the employment lost to firm shutdown that had returned.
Although bottom quintile workers were most affected by exit, they have also recovered over
50% of the employment lost to shutdown. This is roughly triple the 17% of recovered
employment in the top quintile. This in large part reflects the re-opening of bars, restaurants,
retail, and construction businesses as states have begun to open up. These industries tend to
35
These figures were produced using ADP’s employee-level sample, which is underweight small firms. As a
result, the estimated total losses from business closure are less in this table than those reported in the main
text.
40
Figure A7: Employee Recall in Re-Entering Firms, Firms Weighted by February Employment
1 1
.8 .8
Cumulative Probability Cumulative Probability
.6 .6
.4 .4
.2 .2
0 0
0 .5 1 1.5 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
Re-Opening Employment Relative to Feb 1-14 Employment Share of Returning Employees Present in February 1-14
Panel A: Re-Entering Panel B: Share of Employees
Firm Size Who Were Recalled
Notes: Figure shows the employment weighted CDF of re-entering business size (Panel A) and share
of initial business employees recalled (Panel B). The sample of firms is those firms that existed
in early February 2020 and that temporarily shut down. Temporary shutdowns are those firms
that shut down and then subsequently reopened (i.e., resumed paycheck issuance). Employment
measurements are through late June. All trends weighted by firms’ February employment, and are
reweighted to be representative of the SUSB industry × firm size mix.
employ many low wage workers. As a result, their closure and re-opening disproportionately
affect low wage workers.
Appendix Figure A7 plots the distribution of re-entering firm size (Panel A) and the share
of employees at re-entering firms which had been previously employed by that firm (Panel
B), weighting distributions by firms’ February employment. The corresponding figures in the
main text were firm weighted as opposed to employee weighted. We see that the re-entering
firm of the median worker had 59% of their February employment, while nearly every firm
re-entered smaller than their initial size after employment weighting. As in the main text,
the vast majority of re-entering firms hire back almost entirely workers who had previously
been employed by the firm.
Finally, Appendix Figure A8 reproduces Figure 9 in the main text, but looking at contin-
uing firms that increase employment from their trough by at least 1 (Panel A) or 5 (Panel B)
employees. That is, it plots the share of employment growth from continuing firms’ troughs
through the end of June that is accounted for by employees who were previously employed
by the firm. In the main text, we restrict our analysis to firms that increase employment
from their trough by at least 10 employees. As seen in this appendix figure, our results are
not sensitive to this cutoff.
41
Figure A8: Share of Continuing Firm Trough-to-Peak Employment Growth Accounted for
by Recalling Previously Employed Individuals, alternative firm growth cutoffs
1 1
Hiring Outside Emp Hiring Outside Emp Hiring Outside Emp Hiring Outside Emp
Shedding Initial Emp Hiring Initial Emp Shedding Initial Emp Hiring Initial Emp
.8 .8
Cumulative Probability Cumulative Probability
.6 .6
.4 .4
.2 .2
0 0
-1 -.8 -.6 -.4 -.2 0 .2 .4 .6 .8 1 -1 -.8 -.6 -.4 -.2 0 .2 .4 .6 .8 1
Share of Employment Growth From Initial Employees Share of Employment Growth From Initial Employees
Panel A: Growth of at Panel B: Growth of at
least 1 Employee least 5 Employees
Notes: Figure plots the firm-level distribution of the share of firm growth accounted for by recall of previously
employed workers in firms that continually paid workers from the first two weeks of February through the
end of June. Firm growth measured between the firm’s trough employment after March 11 and the end of
June. Recall defined as hiring workers who were employed by the firm in the first two weeks of February.
Panel A restricts attention to firms which add at least 1 employee, while Panel B restricts attention to firms
that add at least 5 employees.
Subsection VIII. Employment Declines and The Ability to Work
at Home
There has been a lot of discussion about the ability of a worker to be able to work from
home as a form of insurance against job loss during the current pandemic driven recession.
For example, Dingel and Neiman (2020) and Mongey et al. (2020) both create measures of
a worker’s ability to work at home using detailed occupation-level task data. Dingel and
Neiman (2020) provide measures of workers’ ability to work at home at the level of 3-digit
NAICS industry.36 Their measure ranges from zero to 1 with a larger number implying
that more workers in that industry can work at home. Appendix Figure A9 shows a scatter
plot using the industry data between the Dingel-Neiman “stay at home” measure and the
decline in paid employment in that 3-digit industry through late April (Panel A) and mid
June (panel B) using our ADP employee sample. As always, all changes are relative to mid
February employment levels. As seen from both panels of the figure, there are a few 3-digit
industries that saw employment increases since mid February including non-store retailers,
which include online retailers (NAICS 454) and delivery services (NAICS 492).
36
Dingel and Neiman (2020) provide multiple measures for their work at home index. We use their
”teleworkable emp” measure. The patterns in Figure A9 are similar regardless of their measure used.
42
Appendix Figure A9 highlights a slight positive relationship between industry-level em-
ployment declines and the ability to work at home over both periods. The solid line in
Panel A is a fitted regression line with a slope coefficient of 0.15, a standard error of 0.08,
and an adjusted R-squared of 0.03. The dashed line is a fitted regression line excluding the
leisure and hospitality industries. The figure shows that the industries that saw the largest
employment declines through late April were, on average, industries where workers are not
able to do their tasks at home. These industries are in the bottom left quadrant of the
figure. However, there are also many industries where workers were not able to do their
tasks at home that saw only modest employment declines (the upper left quadrant of the
figure). Additionally, outside of the industries with the lowest work at home measures (work
at home share greater than 0.3), there was very little relationship between the ability to
work at home and industry employment declines. Even the industries where most workers
can work at home had employment declines of 15 percent on average between early March
and late April. In Panel B, we show the relationship to work at home and employment
declines through mid-June. The fitted regression line is slightly smaller at 0.094 but is no
longer statistically significant at standard levels (standard error = 0.71). It should be noted
that 3-digit industry variation may be too crude a measure to pick up the importance of the
ability to work at home in explaining employment declines. The ability to work at home is
an occupation-level variable as opposed to an industry-level variable. However, the patterns
in Appendix Figure A9 suggest that the ability to work at home is not the primary factor
explaining cross-industry variation in employment declines.
Appendix C Industry Time Series Data
The next set of tables show employment changes and selection adjusted wage changes by
week for various industries during the Pandemic Recession. The employment numbers come
from our firm level sample and the selection adjusted wage numbers come from the employee
sample. For the selection adjusted wage numbers, we track the wages of the same worker in
a given industry from week t to t + 1 and create a chain-weighted index for each industry.
See the main text for additional details. The data can be used to calibrate models with
multiple sectors during the pandemic. We do not attempt to seasonally adjust these patterns;
as a result, some industries, such as agriculture and construction, exhibit large seasonal
employment swings.
43
Figure A9: Relationship Between Dingel-Neiman Work at Home Measure and Actual Paid
Employment Declines, 3 Digit NAICS Industry Variation
221 .4
.2 237
492
492
221
454 454
Percent Employment Decline Percent Employment Decline
115 622
444 325 522 524
113 322
493 324 211 525
482
327 531 334
423 541 511 518 425
447 311 321623484 335 237 518
445
212 238483 621 323 533 115 238
482622 325
236 339443 999424 445 444 524
511
332 562 488 523 321322
484324 335 443 531 522 541
331 337 213 561 813 425 311 327 483
811
332
314
336 334423 515 533
447 113
212326 12 621 339
481441
314
312
811 515 551611 623
441 488 999
211 517
333 624 337 236
−.4 −.2 0
323 813
−.4 −.2 0 .2
517 519 331 333 561 551 525
326 481 562 624 424 523
486 712
532 442
446
812 453 313
313 812 453315 519
712 611
336 446 316213
485 512 713
316 722 532
315 448 512
722 451 442 452 493
711 721487
451 486
−.6 452
721
485 711
448 713
487 −.6
0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1
Dingel−Neiman Work at Home Share Dingel−Neiman Work at Home Share
Panel A: Employment Declines Panel B: Employment Declines
Through Late-April Through Mid-June
Notes: Figure shows the variation in the Dingel-Neiman ”Work at Home” index and the decline
in paid employment at the three-digit NAICS level through late April (Panel A) and mid June
(Panel B) using our employee sample. The leisure and hospitality industries are designated with
diamond while all other industries are denoted with circles. The solid line is a fitted regression line
across all industries. The dashed line is a fitted regression line through the industries excluding
the leisure and hospitality industries. The slopes of the two regression lines are 0.15 (standard
error = 0.08) and 0.13 (standard error = 0.08), respectively in Panel A and 0.09 (standard error =
0.07) and 0.08 (standard error = 0.07) in Panel B. Industry points are labeled with their NAICS
industry code.
44
Table A3: Cumulative Changes in Employment, Mean Wages, and Selection-Adjusted Wages, relative to week ending February
8, by 2-digit NAICS (Table 1)
11: Agriculture 21: Mining 23: Construction
# Base Selection-Adj. # Base Selection-Adj. # Base Selection-Adj.
Week Emp. Wage Wage Emp. Wage Wage Emp. Wage Wage
Feb 1 -2.1 0.1 -0.5 -14.6 0.1 -0.1 1.1 -0.1 10.9
Feb 8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Feb 15 4.2 -0.1 0.1 -3.5 -0.1 0.2 0.1 -0.0 0.1
Feb 22 5.3 0.0 0.2 -2.0 0.0 0.1 1.7 -0.2 0.1
Feb 29 7.9 -0.7 0.2 -9.8 -0.7 0.1 -0.6 -0.1 0.1
Mar 7 7.4 -0.6 0.4 1.0 -0.6 0.4 0.8 0.1 0.3
Mar 14 6.1 -0.1 0.5 -2.6 -0.1 0.7 -0.9 0.2 0.4
Mar 21 5.1 0.6 0.4 -20.2 0.6 0.8 -3.5 0.5 0.3
45 Mar 28 0.1 1.3 0.6 -29.0 1.3 0.5 -9.6 2.6 0.2
Apr 4 0.1 1.0 0.7 -25.8 1.0 0.1 -12.7 3.2 0.2
Apr 11 0.4 1.2 0.8 -18.8 1.2 -0.6 -14.8 3.5 0.3
Apr 18 5.0 1.9 0.9 -20.3 1.9 -1.2 -14.2 4.1 0.3
Apr 25 5.4 2.9 0.9 -32.3 2.9 -1.8 -13.6 4.2 0.3
May 2 7.6 3.0 1.0 -33.4 3.0 -2.3 -13.0 4.2 0.4
May 9 6.5 4.4 0.9 -32.3 4.4 -2.5 -9.4 4.3 0.6
May 16 10.9 4.8 0.9 -24.6 4.8 -3.4 -5.6 4.4 0.7
May 23 10.2 6.4 1.2 -24.0 6.4 -3.8 -5.8 4.0 0.7
May 30 10.5 7.1 1.2 -23.0 7.1 -4.3 -3.9 3.8 0.7
Jun 6 12.0 6.9 1.2 -23.4 6.9 -4.3 -2.6 3.7 0.8
Jun 13 12.1 6.7 1.5 -23.6 6.7 -4.1 -2.2 3.9 0.9
Jun 20 8.5 6.9 1.5 -26.5 6.9 -4.0 -3.9 2.4 1.0
Table A4: Cumulative Changes in Employment, Mean Wages, and Selection-Adjusted Wages, relative to week ending February
8, by 2-digit NAICS (Table 2)
31-33: Manufacturing 42: Wholesale Trade 44-45: Retail Trade
# Base Selection-Adj. # Base Selection-Adj. # Base Selection-Adj.
Week Emp. Wage Wage Emp. Wage Wage Emp. Wage Wage
Feb 1 -0.3 -0.2 -4.4 0.4 -0.2 -5.0 0.2 0.3 -0.5
Feb 8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Feb 15 -2.8 0.5 0.1 0.3 0.5 0.1 -4.3 0.8 0.1
Feb 22 -0.1 1.3 0.1 0.5 1.3 0.2 -4.7 0.8 0.2
Feb 29 -1.8 0.8 0.1 -2.0 0.8 0.0 -6.1 0.8 0.3
Mar 7 -3.1 0.3 0.2 -1.2 0.3 -0.0 -0.6 0.6 0.2
Mar 14 -1.6 0.9 0.3 -4.1 0.9 0.1 -6.4 0.8 0.3
Mar 21 -4.4 1.4 0.1 -9.1 1.4 0.0 -13.0 0.8 0.3
46 Mar 28 -13.6 1.6 -0.1 -11.1 1.6 -0.2 -22.4 1.9 0.5
Apr 4 -16.4 1.8 0.0 -15.1 1.8 -0.4 -27.9 2.4 0.6
Apr 11 -13.0 2.8 0.1 -14.7 2.8 -0.4 -26.7 4.6 0.7
Apr 18 -14.1 4.0 0.0 -13.1 4.0 -0.7 -27.5 5.6 0.7
Apr 25 -15.0 3.6 -0.1 -17.2 3.6 -0.9 -32.0 7.7 0.8
May 2 -11.5 2.7 -0.1 -15.9 2.7 -1.2 -30.8 8.9 0.7
May 9 -12.3 3.0 -0.0 -12.1 3.0 -1.1 -26.6 8.8 0.7
May 16 -12.0 3.6 -0.1 -12.0 3.6 -1.1 -20.3 8.6 0.7
May 23 -11.8 3.8 -0.2 -9.8 3.8 -1.3 -23.5 6.4 0.6
May 30 -11.1 3.2 -0.4 -11.5 3.2 -1.5 -20.2 5.2 0.6
Jun 6 -7.1 3.1 -0.6 -9.8 3.1 -1.6 -19.7 5.6 0.6
Jun 13 -8.9 3.5 -0.4 -9.9 3.5 -1.4 -15.1 5.2 0.7
Jun 20 -12.9 3.4 -0.4 -12.6 3.4 -1.3 -21.0 4.3 0.8
Table A5: Cumulative Changes in Employment, Mean Wages, and Selection-Adjusted Wages, relative to week ending February
8, by 2-digit NAICS (Table 3)
48-49: Transportation/Warehousing 51: Information 52: Finance/Insurance
# Base Selection-Adj. # Base Selection-Adj. # Base Selection-Adj.
Week Emp. Wage Wage Emp. Wage Wage Emp. Wage Wage
Feb 1 8.6 0.5 -3.4 -3.6 0.5 -0.4 6.7 0.4 -5.3
Feb 8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Feb 15 11.3 3.8 0.0 1.9 3.8 0.9 5.3 0.5 0.5
Feb 22 8.1 2.7 0.1 3.2 2.7 0.9 9.5 1.2 0.0
Feb 29 10.5 1.5 -0.1 -6.7 1.5 -0.4 10.7 0.6 0.0
Mar 7 11.8 0.4 0.1 14.9 0.4 -0.8 3.3 0.0 0.2
Mar 14 3.6 -1.0 0.1 -5.3 -1.0 0.5 5.8 1.1 0.3
Mar 21 4.7 0.5 0.2 -6.8 0.5 0.6 -6.3 2.1 -0.2
47 Mar 28 -6.0 5.2 0.1 -18.8 5.2 0.2 -11.8 3.3 -0.2
Apr 4 -6.6 11.8 0.3 -11.3 11.8 0.2 -3.7 3.7 -0.1
Apr 11 -7.8 14.0 0.6 -9.6 14.0 0.6 9.8 2.8 -0.5
Apr 18 -16.1 16.0 0.5 -17.8 16.0 0.6 2.6 3.2 -0.9
Apr 25 -12.9 21.1 0.4 -16.1 21.1 -0.4 4.1 3.2 -1.0
May 2 -12.4 14.8 0.5 -18.7 14.8 -0.6 4.0 3.4 -1.1
May 9 -15.0 17.5 0.5 -17.7 17.5 -0.4 9.9 3.7 -1.1
May 16 -15.2 15.3 0.6 5.2 15.3 -0.4 9.3 4.0 -1.2
May 23 -14.2 17.4 0.6 -13.3 17.4 -0.4 7.8 4.1 -1.3
May 30 -14.4 14.1 0.6 -2.1 14.1 -0.7 5.4 3.8 -1.9
Jun 6 -12.2 11.1 0.7 -12.0 11.1 -1.4 9.0 3.3 -2.4
Jun 13 -16.0 4.5 0.6 -6.4 4.5 -1.3 -2.6 3.5 -2.1
Jun 20 -11.5 4.7 0.7 -9.6 4.7 -0.8 0.2 3.6 -1.7
Table A6: Cumulative Changes in Employment, Mean Wages, and Selection-Adjusted Wages, relative to week ending February
8, by 2-digit NAICS (Table 4)
53: Real Estate 54: Professional Services 55: Management of Companies
# Base Selection-Adj. # Base Selection-Adj. # Base Selection-Adj.
Week Emp. Wage Wage Emp. Wage Wage Emp. Wage Wage
Feb 1 -1.5 0.1 -0.1 -2.9 0.1 -3.6 -2.8 0.1 -0.1
Feb 8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Feb 15 -2.9 0.1 0.2 -0.2 0.1 0.1 -11.2 0.4 0.1
Feb 22 -7.9 0.3 0.3 2.5 0.3 0.3 -10.1 0.2 0.1
Feb 29 -9.8 0.3 0.1 -2.0 0.3 0.2 -15.9 -0.2 0.2
Mar 7 -4.0 -0.0 0.2 -0.9 -0.0 0.0 -10.8 -0.5 0.3
Mar 14 -7.6 0.3 0.4 -2.7 0.3 0.1 -0.2 0.4 0.5
Mar 21 -13.3 -0.3 0.5 -4.8 -0.3 -0.1 -1.9 1.9 0.4
48 Mar 28 -27.9 0.0 0.3 -11.3 0.0 -0.4 -25.8 3.3 -0.4
Apr 4 -25.5 1.1 0.3 -8.3 1.1 -0.6 -17.5 4.2 -0.3
Apr 11 -22.9 1.9 0.5 -7.9 1.9 -0.6 -24.9 6.3 -0.1
Apr 18 -22.8 2.7 0.5 -9.2 2.7 -0.6 -11.3 8.2 -0.2
Apr 25 -23.3 2.9 0.6 -12.2 2.9 -0.6 -10.2 8.0 -0.6
May 2 -21.3 2.7 0.7 -11.2 2.7 -0.7 -9.0 7.8 -0.7
May 9 -21.2 2.4 0.8 -6.9 2.4 -1.1 -11.0 7.9 -0.9
May 16 -18.6 2.4 0.8 -7.7 2.4 -0.9 -13.4 7.8 -1.0
May 23 -21.6 2.9 0.7 -7.0 2.9 -0.7 -11.7 7.3 -1.2
May 30 -21.6 2.6 0.6 -7.9 2.6 -1.2 -10.1 7.5 -1.2
Jun 6 -22.9 1.1 0.6 -4.7 1.1 -1.9 -12.8 7.6 -1.2
Jun 13 -19.4 1.1 0.7 -7.3 1.1 -1.8 -11.6 7.6 -1.1
Jun 20 -19.3 0.5 0.9 -8.5 0.5 -1.5 -13.7 7.9 -1.0
Table A7: Cumulative Changes in Employment, Mean Wages, and Selection-Adjusted Wages, relative to week ending February
8, by 2-digit NAICS (Table 5)
56: Admin/Support Services 61: Education 62: Healthcare
# Base Selection-Adj. # Base Selection-Adj. # Base Selection-Adj.
Week Emp. Wage Wage Emp. Wage Wage Emp. Wage Wage
Feb 1 3.0 0.1 -0.6 -11.0 0.1 -0.0 -1.9 0.6 -0.6
Feb 8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Feb 15 5.6 -0.2 0.2 -8.7 -0.2 -0.1 0.1 0.1 0.1
Feb 22 2.4 0.4 0.3 1.9 0.4 -0.1 -0.9 0.4 0.2
Feb 29 4.5 0.7 0.2 -5.4 0.7 0.0 -3.5 0.4 0.2
Mar 7 3.0 -0.0 0.2 5.7 -0.0 0.2 -0.7 0.0 0.3
Mar 14 3.2 1.0 0.3 -9.0 1.0 0.2 -3.6 0.1 0.5
Mar 21 -5.1 0.5 0.3 -16.7 0.5 0.1 -8.4 0.5 0.5
49 Mar 28 -11.2 1.3 0.3 -23.2 1.3 -0.9 -15.3 0.6 0.5
Apr 4 -16.1 3.2 0.3 -21.2 3.2 -0.9 -15.8 1.8 0.6
Apr 11 -11.2 2.8 0.1 -24.0 2.8 -1.7 -13.4 2.4 0.8
Apr 18 -15.9 3.3 0.2 -18.3 3.3 -1.8 -13.5 2.7 0.8
Apr 25 -12.2 4.8 0.3 -23.5 4.8 -1.9 -16.5 3.1 1.0
May 2 -12.2 5.1 0.4 -16.7 5.1 -1.8 -14.4 3.3 1.1
May 9 -11.7 5.3 0.2 -22.8 5.3 -1.9 -14.5 3.2 1.3
May 16 -12.3 5.7 0.2 -21.1 5.7 -1.8 -11.3 3.0 1.3
May 23 -10.6 6.8 0.4 -29.2 6.8 -1.8 -10.8 3.1 1.3
May 30 -12.2 9.3 0.4 -23.6 9.3 -2.2 -8.1 2.9 1.4
Jun 6 -7.5 11.3 0.1 -29.8 11.3 -2.2 -10.0 2.8 1.3
Jun 13 -9.2 11.2 0.2 -22.2 11.2 -1.9 -8.8 2.4 1.4
Jun 20 -10.7 10.9 0.4 -25.9 10.9 -1.9 -10.2 2.5 1.5
Table A8: Cumulative Changes in Employment, Mean Wages, and Selection-Adjusted Wages, relative to week ending February
8, by 2-digit NAICS (Table 6)
71: Arts/Recreation 72: Food/Accommodation 81: Other Services
# Base Selection-Adj. # Base Selection-Adj. # Base Selection-Adj.
Week Emp. Wage Wage Emp. Wage Wage Emp. Wage Wage
Feb 1 6.0 -0.0 -2.5 5.7 -0.0 -1.2 -0.4 -0.3 -0.1
Feb 8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Feb 15 9.7 0.7 0.1 -0.2 0.7 0.2 -1.3 0.0 0.4
Feb 22 10.1 -0.1 0.2 3.5 -0.1 0.2 1.2 0.3 0.3
Feb 29 13.3 -0.2 0.3 3.3 -0.2 0.4 -0.4 0.4 0.3
Mar 7 3.7 -0.4 0.2 -2.2 -0.4 0.5 -0.6 0.2 0.4
Mar 14 4.9 -0.1 0.2 -5.1 -0.1 0.7 -3.1 0.2 0.6
Mar 21 -8.7 1.4 0.4 -19.1 1.4 0.6 -9.7 0.5 0.6
50 Mar 28 -24.2 3.0 0.2 -31.5 3.0 0.4 -19.7 2.3 0.6
Apr 4 -36.0 5.3 -0.4 -39.0 5.3 0.2 -24.4 4.8 0.6
Apr 11 -41.9 8.3 -0.7 -42.4 8.3 0.2 -28.0 6.7 0.5
Apr 18 -48.1 9.3 -1.0 -45.7 9.3 0.5 -25.7 8.5 0.4
Apr 25 -45.9 10.7 -1.2 -45.6 10.7 0.7 -26.0 8.2 0.6
May 2 -43.6 10.6 -1.1 -43.9 10.6 0.8 -26.1 8.1 0.6
May 9 -41.1 10.4 -1.1 -40.3 10.4 1.0 -24.0 7.6 0.7
May 16 -40.7 9.5 -1.0 -37.0 9.5 1.0 -20.7 7.6 0.8
May 23 -39.2 9.0 -1.3 -35.1 9.0 1.0 -21.5 6.3 0.8
May 30 -34.9 7.4 -0.9 -33.6 7.4 1.1 -17.9 4.8 0.5
Jun 6 -33.0 6.1 -0.7 -30.5 6.1 1.1 -17.1 4.1 0.5
Jun 13 -26.1 4.2 -0.5 -27.4 4.2 1.1 -16.2 4.4 0.7
Jun 20 -25.1 3.7 -1.3 -26.9 3.7 1.1 -15.7 4.6 0.9
File and source
- File
- the-u-s-labor-market-during-the-beginning-of-the-pandemic-recession.pdf
- Size
- 1,026,977 bytes
- SHA-256
- 84f154a0a857fd4258acae0eaa7abf54ddfe472c6acb70c91141ec9d62518e53
- Original
- www.nber.org