Covid-Induced Economic Uncertainty, Tasks, and Occupational Demand — LIDAM Discussion Paper IRES 2022/02
Summary
In LIDAM Discussion Paper IRES 2022/02, Sotiris Blanas and Rigas Oikonomou study how Covid-induced economic uncertainty affected relative demand for different occupations in the US. They use monthly online job postings at the occupation and US state level from January 2020 to December 2020, combined with fixed occupational characteristics and monthly measures of economic uncertainty. They find that uncertainty raised relative demand for occupations high in non-routine cognitive analytical, non-routine cognitive interactive and non-routine manual tasks, especially non-essential ones, and for occupations combining analytical and social tasks. The authors say this fits routine-biased technological change and job polarisation accelerating in recessions, while uncertainty lowered demand for customer-oriented occupations such as food service and personal care.
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
Cite this document
“Covid-Induced Economic Uncertainty, Tasks, and Occupational Demand — LIDAM Discussion Paper IRES 2022/02.”
Original at sites.uclouvain.be. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/covid-induced-economic-uncertainty-tasks-and-occupational-demand-20f5e77041/
, SHA-256 8554a244e14a65d4fedd15c9d543d21ed7a4e48fec67da3b92dddb47563e4935.Full text
COVID-INDUCED ECONOMIC
UNCERTAINTY, TASKS, AND
OCCUPATIONAL DEMAND
Sotiris Blanas, Rigas Oikonomou
LIDAM Discussion Paper IRES
2022 / 02
Covid-induced Economic Uncertainty, Tasks, and Occupational
Demand∗
Sotiris Blanas† Rigas Oikonomou‡
Abstract
In this paper, we provide novel evidence of the impact of Covid-induced economic un-
certainty on the relative demand for different occupations in the US, according to a wide
range of occupational characteristics. We conduct the analysis using monthly online job
postings data at the occupation-US state level in January 2020 – December 2020 together
with data on fixed occupational characteristics and monthly measures of economic un-
certainty in the US, which are apportioned to occupation-state pairs based on pre-Covid
country-wide occupational employment shares. The analysis reveals that Covid-induced
economic uncertainty increased the relative demand for occupations with relatively high
non-routine cognitive analytical, non-routine cognitive interactive, and non-routine man-
ual task content – especially when these are also non-essential, as well as occupations
that have both relatively high non-routine cognitive analytical and social or interaction
task content. This evidence is consistent with the secular phenomenon of routine-biased
technological change resulting in job polarisation and the growing complementarity be-
tween analytical and social tasks, but also with its episodic aspect implying its accel-
eration during recessions. Additional evidence, however, shows that Covid-induced eco-
nomic uncertainty decreased the relative demand for customer-oriented occupations (e.g.
food service, personal care and service) and increased the relative demand for essential or
contact-intensive occupations with relatively high routine manual or routine cognitive task
content, as well as occupations that are both contact-intensive and essential or service-
oriented (e.g. healthcare practice and support, protective service, community and social
service). This evidence is rationalised by idiosyncratic features of the pandemic shock
(e.g. major health crisis, social distancing, lockdown). Further research in this direction
could help us to understand whether these effects are temporary or long-lasting.
Keywords: Occupational demand; Occupational characteristics; Tasks; Online
Job Postings; Covid-induced economic uncertainty; Covid-19; Pandemic
JEL Classification: E32, J23, J24, J63, O33
∗
We thank Maurizio Zanardi for his insightful comments.
†
Corresponding author: IMT School for Advanced Studies Lucca, Piazza San Ponziano 6, 55100, Lucca,
Italy. Email: sotirios.blanas@imtlucca.it.
‡
Université catholique de Louvain. Email: rigas.oikonomou@gmail.com.
1 Introduction
The Covid-19 pandemic outbreak in early 2020 has led to a global health emergency and the
imposition of restrictive measures (e.g. strict lockdowns, domestic and international travel
restrictions and bans, social distancing, isolation, contact-tracing) for the containment of the
rapid spread of the virus (Ferguson et al., 2020; Gupta et al., 2020). The upheaval in the
labour markets around the globe was imminent and at massive scale, and has persisted to
a large extent since then. The nascent literature examining the labour market implications
of the Covid-19 shock has provided some first robust evidence on its unequal effects on the
employment, earnings and hiring rates of workers with different demographic or other char-
acteristics (e.g. Adams-Prassl et al., 2020; Baylis et al., 2021; Mongey et al., 2021; Forsythe
et al., 2020).
Our goal in this paper is to advance further this literature by conducting the first compre-
hensive analysis on the effects of the Covid-19 shock on the relative demand for occupations,
according to a broad set of characteristics such as routine intensity, service-orientedness, phys-
ical contact intensity, and essentiality. To this end, we exploit monthly data from Burning
Glass Technologies (BGT) on online job postings of two-digit (SOC) occupations in all fifty-
one US states over January 2020 – December 2020 and data on occupational characteristics
from various sources.1 Importantly, in addition to shedding light upon how the Covid-19
shock has shifted the occupational composition of US labour demand, the analysis allows us
to reflect upon these shifts in relation to labour market trends that were persistent in the
US and other advanced economies in recent decades and until before the pandemic outbreak,
most notably, job polarisation.2
We start the analysis by documenting that the ratio of the country-wide number of online
job postings to total US population declined sharply in the months of the first lockdown
(March–April), recovered only partially and exhibited small fluctuations when the lockdown
was lifted (May–October), and plummeted again during the second lockdown (November–
December).3 We, then, document that although the per capita number of online job postings
1
The main advantage of online job postings data is that they provide real-time information about future
labour demand.
2
This term has been coined to describe the decrease in employment of occupations with relatively high
routine task content, which lie in the middle of the wage distribution, relative to the employment of occupations
with relatively high non-routine cognitive task content and occupations with relatively high non-routine manual
task content, which lie in the high and low end, respectively, of the wage distribution (e.g. Autor et al., 2003;
Acemoglu & Autor, 2011).
3
Sharp declines in online job postings in March–April 2020 for the US and other countries have also been
1
of all occupations under consideration declined sharply during the first lockdown, there were
important differences in their trends in the subsequent months. As a consequence, the shares
of online job postings of some occupations increased (e.g. community and social service;
healthcare practice; healthcare support), while those of other occupations decreased (e.g.
management; business and financial; legal; arts and entertainment; sales; food preparation
and service). Alongside these trends, we show that occupations differ substantially in the
intensity or status of the characteristics under consideration.
In order to investigate whether and how the Covid-19 shock has contributed to the afore-
mentioned shifts in the occupational composition of US labour demand, we regress the share
of online job postings of each occupation-state pair on the interaction between a measure of
Covid-induced economic uncertainty and a variable for a certain occupational characteristic.
We focus on economic uncertainty as it is a key driving force of firms’ hiring decisions amid the
Covid-19 shock. To capture it empirically, we rely on the monthly news-based and composite
Economic Policy Uncertainty (EPU) indices of Baker et al. (2016), and alternatively, on the
monthly sales revenue and employment growth uncertainty indices of the Survey of Business
Uncertainty (SBU).4 As these are country-level measures, we apportion them to occupation-
state pairs according to the pre-Covid (2019) share of employment of each pair in country-wide
employment of the respective occupation.5 We estimate the specification by OLS while con-
trolling for unobserved differences across occupation-state pairs (e.g. initial demographics,
human capital) by including occupation-state dummies, and for common time-varying shocks
(e.g. FED and federal government policies) by including month dummies.6
reported by other studies (e.g. Forsythe et al., 2020; Hensvik et al., 2021).
4
Similarly to postings of job vacancies, the main advantage of these measures of economic uncertainty is
that they are forward-looking, while the SBU measures have the extra advantage of capturing the subjective
beliefs of managers about the short-run prospects of their businesses and the aggregate economy. For a detailed
account of the latter measures and the associated survey, see Altig et al. (2020). Intuitively, Covid-induced
economic uncertainty, measured by either the EPU or the SBU indices, moved in opposite directions with
respect to online job postings per capita: it spiked during the first lockdown (March–April), after which it
declined but remained at high levels relative to the pre-pandemic months, and picked up slightly again during
the second lockdown (November–December). Relatedly, Baker, Bloom, Davis, & Terry (2020) have shown
that these uncertainty measures can explain a large fraction of the drop in US GDP witnessed during the first
months of the pandemic outbreak.
5
Information on the latter variable is drawn from the US Bureau of Labour Statistics (BLS).
6
We claim that the OLS estimates are unlikely to suffer from the simultaneity bias for the following reasons.
Due to the size of the US economy, it is highly unlikely that changes in the number of online job postings
of a given occupation-state pair can contribute to country-wide economic uncertainty. Also, the country-wide
employment shares of occupation-state pairs are held constant in a year that precedes the highly unpredictable
Covid-19 shock, while the variables capturing occupational characteristics are also time-invariant. To address
a possible omitted variable bias in the estimates due to differences in the type of anti-Covid measures adopted
across US states or in the timing of their imposition or relaxation, we perform a robustness check in which we
replace the month dummies with dummies for state-month pairs.
2
The econometric analysis reveals that Covid-induced economic uncertainty impacted dif-
ferentially the relative demand for occupations with different characteristics in the US. In
particular, Covid-induced economic uncertainty increased the relative demand for occupations
with relatively high non-routine cognitive analytical, non-routine cognitive interactive, and
non-routine manual task content –especially, when these occupations are also non-essential
(e.g. architecture and engineering; education). It also increased the relative demand for
occupations that have both relatively high non-routine cognitive analytical and social or in-
teraction task content (e.g. managerial). By contrast, it decreased the relative demand for
non-essential or less contact-intensive occupations with relatively high routine intensity (e.g.
office and administrative support). These findings are in accord with existing evidence show-
ing that the secular phenomenon of routine-biased technological change (RBTC) has led to
job polarisation (e.g. Autor et al., 2003; Spitz-Oener, 2006; Acemoglu & Autor, 2011; Autor &
Dorn, 2013; Goos et al., 2014), as well as with its episodic aspect pointing to the acceleration
during recessions of the adoption of new technologies undertaking routine tasks (Hershbein
& Kahn (2018); Jaimovich & Siu, 2020). In addition, the last two findings are consistent
with existing evidence on the link between RBTC and the growing complementarity between
analytical and social tasks (Borghans et al. (2014); Weinberger (2014); Deming, 2017).
Interestingly, however, we derive additional findings that are seemingly at odds with the
job polarisation phenomenon and likely explained within the new setting that has been gen-
erated since the onset of the Covid-19 shock. That is, a setting dominated by a major health
crisis and restrictive measures (e.g. social distancing, lockdown) aiming at restraining the
rapid spread of the virus. According to these findings, Covid-induced economic uncertainty
decreased the relative demand for customer-oriented occupations (e.g. food preparation and
service; personal care and service; non-essential salespersons). In recent decades and until
before the pandemic hit, such occupations had experienced relative employment gains, partly
because they were, and still are, less amenable to offshoring and automation. Also, Covid-
induced economic uncertainty increased the relative demand for essential or contact-intensive
occupations with relatively high routine manual or routine cognitive task content, for occupa-
tions that are both essential and contact-intensive, and those that are both contact-intensive
and service-oriented. These effects are likely to be particularly driven by healthcare prac-
tice and support, protective service, and community and social service occupations. Whether
these effects are temporary or will persist in the post-Covid era, remains to be explored in
3
future research.
Related literature The novel empirical evidence of this paper contributes to four streams of
literature. First, our focus on identifying the differential effects of the Covid-19 shock on the
demand for different occupations, and subsequently, to consider the routine and non-routine
task content of occupations and their degrees of customer- and service-orientedness is moti-
vated by the influential literature providing evidence on job polarisation and its determinants.
In turn, a set of findings that we derive are consistent with key evidence of two streams of this
literature. In particular, one stream shows that routine-biased technological change (RBTC)
and offshoring are major drivers of job polarisation (Autor et al., 2003; Spitz-Oener, 2006;
Acemoglu & Autor, 2011; Autor & Dorn, 2013; Goos et al., 2014). RBTC also partly explains
why cognitive analytical and cognitive interactive tasks have become increasingly complemen-
tary since the 1980s, as Borghans et al. (2014), Weinberger (2014), and Deming (2017) have
shown using US data. In fact, machines have starting undertaking some cognitive analytical
tasks, while social interaction is, at least up to now, a prerogative of humans (Autor, 2015).
Another stream of this literature establishes a link between RBTC and job polarisation during
recessions. Using data for the US, Hershbein & Kahn (2018) show that skill requirements in
online job postings increased in Metropolitan Statistical Areas (MSAs) that were dispropor-
tionately hit by the Great Recession (2007–2009), that this effect persisted through 2015, and
that it was correlated with a surge in capital investments, both at the MSA and firm levels.
Relatedly, Jaimovich & Siu (2020) show that employment losses in routine occupations in the
US since the early 1980s have been primarily observed during recessions, and that the lack of
employment recovery in these occupations accounts almost entirely for the jobless recoveries
in the aggregate.
Our approach to consider in the analysis additional occupational characteristics, most
notably, physical contact intensity, tele-work potential, and essentiality, is influenced by the
burgeoning stream of literature studying the labour market implications of the Covid-19 shock.
To our knowledge, this paper is the first to conduct a comprehensive analysis of the differential
effects of this shock on the demand for different occupations, according to wide range of
occupational characteristics. In doing so, it enriches this stream of literature by deriving a set
of findings that, as mentioned earlier, are consistent with evidence on job polarisation, but also
another set of findings that are at odds with it, and can rather be rationalised by idiosyncratic
features of the pandemic shock (e.g. major health crisis, social distancing, lockdown).
4
Using data from real-time surveys conducted in the US, UK, and Germany in March and
April 2020, Adams-Prassl et al. (2020) document smaller job losses in the latter country, which
are attributed to its firmer labour market institutions such as the short-time work schemes
that were well-established in the country pre-Covid. They also show that job and earnings
losses in the three countries were higher for workers with temporary work arrangements and
those with relatively limited tele-work potential. Disproportionate losses were also incurred
by female and non-college-graduate workers in the US and UK, but not in Germany. In a
seminar vein, using data for Canada, Baylis et al. (2021) find larger employment losses for
women in the first part of the Covid-19 pandemic outbreak due to their larger concentration
in high viral risk occupations, and for the less educated workers due to their lower tele-work
potential. Using data from the US Current Population Survey, Cortes & Forsythe (2020) find
that the Covid-19 shock has exacerbated inequalities, with the low-educated, younger, and
female workers, and those of Hispanic origin experiencing larger increases in job losses and
larger decreases in hiring rates. Similar evidence is also provided by Montenovo et al. (2020).
Further on the US, Mongey et al. (2021) find that jobs with low tele-work potential and those
with high physical proximity requirements, which were more likely to be filled by workers
with relatively low education and income, suffered relatively large employment losses, but not
those that are also essential.7
Using data on online job postings from Burning Glass Technologies, Forsythe et al. (2020)
report a collapse of over 40% in the number of job vacancies in the US by April 2020. How-
ever, they also find that essential retail and nursing maintained or decreased slightly their
job posting levels. The simultaneous rise in unemployment insurance claims made by workers
in such occupations points to a great deal of churn in the US labour market. The latter is
consistent with the evidence provided by Barrero et al. (2020) on realised and predicted job
reallocation in the US since the onset of the Covid-19 shock. A collapse of similar magni-
tude in the number of job vacancies is also reported for Sweden by Hensvik et al. (2021),
who exploit relevant data from the country’s largest online job board. The authors also pro-
vide evidence on job reallocation, with labour market searches shifting towards occupations
that have been hit less severely by the pandemic.8 Exploiting rich online job postings data
7
Large increases in employment losses during the Covid-19 shock have been reported by several other studies
(Bartik et al., 2020; Hassan et al., 2020; Cajner et al., 2020; Coibion et al., 2020; Bick & Blandin, 2021).
8
A similar shift in labour market searches has been identified for the US by Marinescu et al. (2021), with
the use of data from Glassdoor, and Bernstein et al. (2020), with the use of data from a job board focused on
tech companies (Angel-List Talent).
5
from LinkUp, Campello et al. (2020) show that new-hiring cuts during the Covid-19 shock
were more pronounced for high-skill jobs, smaller or credit-constrained firms, highly-unionised
industries, and low-income areas.
The final stream of literature to which our paper relates studies non-labour-market out-
comes of economic uncertainty induced by the Covid-19 shock. Baker, Bloom, Davis, & Terry,
2020 estimate that half of the predicted US output contraction during the months of 2020
was accounted for by Covid-induced economic uncertainty, as measured by the stock mar-
ket volatility (VIX), the news-based Economic Policy Uncertainty (EPU), and the Survey of
Business Uncertainty (SBU) indices. In addition, Baker, Bloom, Davis, & Terry (2020) show
that investments in tangible and intangible capital are extremely responsive to EPU, most
likely due to their irreversible nature and the relatively high associated sunk costs. There is
also evidence on the unprecedented impact of the Covid-19 shock on the stock market (Baker,
Bloom, Davis, Kost, et al., 2020; Leduc & Liu, 2020). Our paper emulates this group of
studies in using the EPU and SBU measures in order to capture Covid-induced economic
uncertainty. Unlike these studies, however, it is the first paper that does so with the aim to
identify its labour market implications and, in particular, its implications for occupational
demand.
Layout of the paper In Section 2, we describe the different datasets that we rely on and
the construction of variables. We also present the main trends of online job postings and
measures of economic uncertainty in the US during the Covid-19 shock, as well as the patterns
of occupations by occupational characteristic. In Section 3, we give a detailed account of
the econometric model and the identification strategy, while in Section 4, we present and
rationalise the main results and perform some robustness checks. In Section 5, we highlight
the key conclusions drawn from the main results and indicate some promising avenues for
future research.
2 Data and stylised facts
In this section, we describe the online job postings data that we employ throughout the
empirical analysis, the variables capturing a broad range of occupational characteristics, and
the variables capturing economic uncertainty induced by the Covid-19 pandemic outbreak.
6
2.1 Online job postings
The monthly online job postings data for the US cover the period January 2020 – Decem-
ber 2020 and have been assembled, compiled and made publicly available by Burning Glass
Technologies (henceforth, BGT).9 BGT collects millions of job postings from close to 40,000
online sources, such as job boards, employer sites, newspapers, and public agencies, to name
a few. The publicly available version of the dataset includes the monthly count of online job
postings by geographic area (i.e., country-wide, US state) and occupation. The fifty-one US
states are identified by their unique two-letter ISO codes. Occupations are identified by their
two-digit 2010 Standard Occupational Classification (SOC) codes. The dataset also includes
information, originally drawn from the US Census, on the monthly population of the whole
country and each US state in 2019. The key advantage of using online job postings data is
that they offer a point-in-time measure of labour demand and at a finer-grained (i.e., two-
digit occupation) level than relevant surveys do. Equally importantly, job vacancy postings
are inherently forward-looking, capturing the intention of firms to develop new employment
relationships in the short- to medium-run.10
Panel (a) of Figure 1 shows the monthly variation in the total number of online job
postings per capita in the US over January 2020 – December 2020.11 In April, when the
disease was spreading fast and lockdown and social distancing measures were in effect, online
job postings per capita declined steeply, to 0.69% from 1.08% and 1.06% in January and
February and roughly 1.04% in March.12 They partially bounced back in May and June and
fluctuated slightly until October –when lockdown measures had been lifted, declined sharply
in November –when the second wave of the pandemic broke out– albeit much less than in
April, and declined slightly further in December. Importantly, as evident in Panel (b), the
9
Note that although the Covid-19 crisis continued through 2021, BGT has made publicly available only the
data for 2020.
10
Thus, they are indicative of the subjective perceptions firms have about the short- and longer-run prospects
of their idiosyncratic overall performance, the performance of their industry, and the aggregate economy. On
the downside, such data may miss part of labour demand as not all jobs postings are advertised online, while
employers may advertise multiple job vacancies in a single posting. For instance, job vacancy postings tend
to cover a disproportionately high number of higher-skilled occupations (e.g. Hershbein & Kahn, 2018). Be
that as it may, online job postings data should be viewed as complementary to survey data, rather than their
substitutes. In fact, as Dalton et al. (2020) show, BGT’s online job postings data are very much aligned with
those collected by the US Bureau of Labor Statistics through the Job Opening and Labor Turnover Survey
(JOLTS).
11
This is calculated as the ratio of the country-wide number of online job postings to aggregate US population.
12
This translates to a 37% decline in the online job postings per capita between January and April. A decline
of 40% in the number of online job postings in the US has been reported by Forsythe et al. (2020), who also
use BGT data. A decline of a similar magnitude has also been reported for Sweden by Hensvik et al. (2021),
based on data from the country’s major online job board.
7
trend exhibited by online job postings per capita bears a very close resemblance to the trend of
the total number of monthly job vacancies to population ratio, calculated based on the JOLTS
data. In addition to the similar trends, the BGT data at our disposal have a relatively high
degree of representativeness, as online job openings account for at least half of the official
number of job openings for each month of 2020. These facts are consistent with the evidence
of Dalton et al. (2020) on the alignment between BGT and JOLTS data and confirm BGT’s
high standards of online job postings data collection and processing. In addition, although
we by no means disregard the possibility of idiosyncratic factors having contributed to the
aforementioned trend of US online job postings per capita, it is noteworthy that we obtain
very similar trends for all the other countries that are available in the BGT dataset, namely,
Australia, Canada, New Zealand, Singapore and the UK. The similarities in the trends is
particularly evident in Feb 2020 – May 2020, that is, immediately before, during, and towards
the end the first wave of the Covid-19 pandemic (Figure B1).13
Albeit informative about how US labour demand evolved in the months of 2020, Panels (a)
and (b) of Figure 1 mask the possible heterogeneity in the demand for different occupations
over that period. Indeed, Panels (a)–(v) of Figure 2 reveal that although there was hardly any
two-digit (2010 SOC) occupation that did not experience a precipitous decline in its online
job postings per capita during the first wave of the pandemic (March 2020 – May 2020), there
were remarkable differences along this dimension in the subsequent months. In particular,
the online job postings per capita of a relatively small number of occupations were mostly on
an increasing trend, especially, during the second wave of the pandemic (healthcare practice;
community and social service; protective service; farming, fishing, and forestry; production).
Also, a couple of occupations (healthcare support; transportation and material moving) had
higher online job postings per capita in the second half of 2020 than in January of the same year
(i.e., pre-Covid) and despite small declines in these during the second wave of the pandemic.
By contrast, the online job postings per capita of the rest of the occupations recovered only
partly around May–June –when the lockdown measures had been lifted, increased further or
fluctuated slightly through October, before falling precipitously again as of November due to
the second wave of the pandemic.
13
Note that we do not include the additional countries in the analysis due to lack of or incomplete data on
country-wide employment shares of occupation-geographic area pairs. As we will explain in Section 3 for the
US case, if we were to include the additional countries, we would need these data in order to apportion the
country-level economic uncertainty measures to each occupation-area pair.
8
Ja US sales revenue growth uncertainty Ja US news−based EPU Ja online job postings per capita (%)
Fen 20 3 4 5 6 7 Fen 20 100 200 300 400 500 n .7 .8 .9 1 1.1
20
M b 2 19 M b 2 19 20
a 0 a 0 Fe
A r 2019 A r 20 19 b
M pr 2 19 M pr 2 19 20
ay 0 ay 01 20
Ju 2019 Ju 20 9 M
n 1 n 1 ar
J 20 9 Ju 20 9 20
A ul 2 19 A l 2 19 20
u 0 u 0
Seg 2019 A
Seg 2019 pr
O p 2019 p 20
O 2019 20
N ct 2 19
o 0 N ct 2 19
o 0 M
D v 2 19 ay
ec 0 D v 2019
ec 1 20
Ja 2019 Ja 20 9 20
Fen 2019 Ju
M b 2 20
Fen 2019 n
a 0 M b 2 20 20
A r 2 20 a 0 20
M pr 2020 A r 20 20
ay 02 M pr 2 20 Ju
ay 02 l2
Ju 20 0 02
n 1 Ju 20 0 0
J 20 9 n 2 A
A ul 2 20 Ju 20 0 ug
u 0 A l 2 20 20
Seg 2020 u 0 20
in the US in January 2019 – June 2021.
Seg 2020
O p 2020 p Se
p
N ct 2 20 O 2020
o 0 N ct 2 20
20
D v 2 20 o 0 20
ec 0 O
(c) News-based EPU
Ja 2020 D v 2020
ec 2 ct
Fen 2020 Ja 20 0 20
20
M b 2 21 Fen 2020 N
a 0 M b 2 21 ov
A r 2 21 a 0
M pr 2021 20
ay 0 A r 20 21 20
M pr 2 21 D
(a) Online job postings per capita
Ju 2021 ay 02 ec
n 11
2
(e) Sales revenue growth uncertainty
20 20 1 02
21 21 0
9
Ja US employment growth uncertainty Ja US three−component EPU Ja # of job postings per capita
3 4 5 6 7 100 150 200 250 300 350 n .5 1 1.5 2 2.5
Fen 20 Fen 20 20
M b 2 19 M b 2 19 20
a 0 a 0 Fe
A r 2019 A r 20 19 b
M pr 2 19 M pr 2 19 20
ay 01 ay 01 20
Ju 20 9 Ju 20 9 M
n 1
Business, Stanford University; see particularly Altig et al. (2020).
J 20 9 n 1 ar
Ju 20 9 20
A ul 2 19 A l 2 19 20
u 0 u 0
Seg 2019 A
Seg 2019 pr
O p 2019 p 20
O 2019 20
N ct 2 19 M
o 0 N ct 2 19
o 0
D v 2 19 ay
ec 0 D v 2019
ec 1 20
Ja 2019 Ja 20 9 20
Fen 2019 Ju
M b 2 20
Fen 2019 n
a 0 M b 2 20 20
A r 2 20 a 0 20
M pr 2020 A r 20 20
Ju
Online job postings p.c. (BGT)
ay 0 M pr 2 20
ay 02 l 20
Ju 2020
n 1 Ju 20 0 20
J 20 9 n 2 A
A ul 2 20 Ju 20 0 ug
u 0 A l 2 20 20
Seg 2020 u 0 20
Seg 2020
O p 2020 p Se
p
N ct 2 20 O 2020
o 0 N ct 2 20
20
D v 2 20 o 0 20
(d) Composite EPU
ec 0 D v 2020 O
Ja 2020 ec 2 ct
Fen 2020 Ja 20 0 2 02
M b 2 21 0
a 0
Fen 2020 N
A r 2 21 M b 2 21 ov
M pr 2021 a 0 20
ay 0 A r 20 21 20
Ju 2021 M pr 2 21 D
n 11 ay 02 ec
(f) Employment growth uncertainty
Job openings p.c. (BLS)
20 20 1 2 02
21 21 0
(b) Online (BGT) Vs Total (BLS JOLTS)
Figure 1: Job postings per capita and Covid-induced economic uncertainty in the US
Baker et al. (2016). SBU measures in Panels (e) and (f): Atlanta FED, University of Chicago Booth School of Survey (JOLTS). EPU measures in Panels (c) and (d): https://www.policyuncertainty.com/; see particularly based on job vacancy data of the US Bureau of Labor Statistics (BLS)’s Job Openings and Labor Turnover data of Burning Glass Technologies (BGT). Total job vacancies per capita in Panel (b): Authors’ calculations Source: Online job postings per capita in Panels (a) and (b): Authors’ calculations based on online job postings (e) and (f) display the monthly variation in the sales revenue and employment growth uncertainty measures the monthly variation in US news-based and composite EPU measures in January 2019 – May 2021. Panels of monthly job vacancies to population ratio in January 2020 – December 2020. Panels (c) and (d) display – December 2020. Panel (b) displays the line of Panel (a) along with the monthly variation in the total number Notes: Panel (a) displays the monthly variation in the online job postings per capita in the US in January 2020
Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%)
n .06 .08 .1 .12 n .02 .03 .04 .05 n .016 .018 .02 .022 .024 n .0075 .008 .0085 .009 .0095 .01 n .07 .08 .09 .1 .11 .12
20 20 20 20 20
2 0 20 20 20 20
Fe Fe Fe Fe Fe
b b b b b
20 20 20 20 20
20 20 20 20 20
M M M M M
ar ar ar ar ar
20 20 20 20 20
2 0 20 20 2 0 2 0
A A A A A
pr pr pr pr pr
2 02 20 20 20 20
0 20 20 20 20
M M M M M
ay ay ay ay ay
20 20 20 20 20
20 20 20 20 20
Ju Ju Ju Ju Ju
n n n n n
20 20 20 20 20
2 0 20 20 20 20
Ju Ju Ju Ju Ju
l2 l2 l2 l2 l2
02 02 02 02 02
0 0 0 0 0
A A A A A
ug ug ug ug ug
20 20 20 20 20
20 20 20 2 0 2 0
Se Se Se Se Se
p p p p p
20 20 20 20 20
20 20 20 20 20
O O O O O
ct ct ct ct ct
20 20 20 20 20
20 20 20 2 0 2 0
N N N N N
(i) Arts, Design, Enter- (j) Healthcare Practi- (k) Healthcare Support
ov ov ov ov ov
20 20 20 20 20
2 2
(a) Management
20 20 20 0 0
(e) Life, Physical, & So- (f) Community & So-
D D D D D
ec ec ec ec ec
20 20 20 20 20
2 0 20 20 20 20
Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%)
n .02 .025 .03 .035 n .0008 .0009 .001 .0011 n .014 .016 .018 .02 .022 n .09 .1 .11 .12 .13 n .009 .01 .011 .012 .013 n .03 .04 .05 .06 .07
20 20 .0007 20 20 20 20
2 0 20 20 20 20 20
Fe Fe Fe Fe Fe Fe
b b b b b b
20 20 20 20 20 20
20 20 20 20 20 20
M M M M M M
ar ar ar ar ar ar
(m) Food Preparation (n) Building Cleaning (o) Personal Care &
20 20 20 20 20 20
2 0 2 0 20 20 2 0 2 0
A A A A A A
pr pr pr pr pr pr
2 02 2 02 20 20 20 20
0 0 20 20 20 20
M M M M M M
ay ay ay ay ay ay
20 20 20 20 20 20
20 20 20 20 20 20
Ju Ju Ju Ju Ju Ju
n n n n n n
20 20 20 20 20 20
2 0 2 0 20 20 20 20
Ju Ju Ju Ju Ju Ju
l2 l2 l2 l2 l2 l2
02 02 02 02 02 02
0 0 0 0 0 0
A A A A A A
ug ug ug ug ug ug
20 20 20 20 20 20
20 20 20 20 2 0 2 0
Se Se Se Se Se Se
p p p p p p
(q) Office & Adminis- (r) Farming, Fishing, & (s) Construction & Ex- (t) Installation, Main-
20 20 20 20 20 20
20 20 20 20 20 20
O O O O O O
ct ct ct ct ct ct
20 20 20 20 20 20
20 20 20 20 2 0 2 0
N N N N N N
(u) Production
ov ov ov ov ov ov
20 20 20 20 20 20
20 20 20 20 2 0 2 0
D D D D D D
ec ec ec ec ec ec
tainment, Sports, Me- tioners & Technical
20 20 20 20 20 20
2 0 2 0 20 20 20 20
cial Science cial Service
10
Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%)
n .05 .06 .07 .08 .09 n .01 .012 .014 .016 n .016 .018 .02 .022 .024 .026 n .015 .02 .025 .03 n .004 .005 .006 .007 .008 n .06 .07 .08 .09 .1 .11
20 20 20 20 20 20
2 0 2 0 20 20 20 20
Fe Fe Fe Fe Fe Fe
b b b b b b
20 20 20 20 20 20
20 20 20 20 20 20
M M M M M M
ar ar ar ar ar ar
(b) Business & Finan- (c) Computer & Math- (d) Architecture & En-
20 20 20 20 20 20
2 0 2 0 20 20 2 0 2 0
A A A A A A
pr pr pr pr pr pr
2 2 20 20 20 20
(v) Transportation
02 02
0 0 20 20 20 20
M M M M M M
ay ay ay ay ay ay
20 20 20 20 20 20
20 20 20 20 20 20
Ju Ju Ju Ju Ju Ju
n n n n n n
20 20 20 20 20 20
2 0 2 0 20 20 20 20
Ju Ju Ju Ju Ju Ju
l2 l2 l2 l2 l2 l2
02 02 02 02 02 02
0 0 0 0 0 0
A A A A A A
ug ug ug ug ug ug
20 20 20 20 20 20
20 20 20 20 2 0 2 0
Se Se Se Se Se Se
p p p p p p
(g) Legal
20 20 20 20 20 20
20 20 20 20 20 20
O O O O O O
ct ct ct ct ct ct
20 20 20 20 20 20
20 20 20 20 2 0 2 0
N N N N N N
ov ov ov ov ov ov
Material Moving
20 20 20 20 20 20
20 20 20 20 2 0 2 0
D D D D D D
ec ec ec ec ec ec
20 20 20 20 20 20
2 0 2 0 20 20 20 20
dia
&
& Service & Maintenance Service
Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%) Ja online job postings per capita (%)
n .02 .025 .03 .035 n .06 .08 .1 .12 .14 n .01 .011 .012 .013 .014 .015 n .018 .02 .022 .024 .026 .028 n .018 .02 .022 .024 .026 .028
20 20 20 20 20
2 0 20 20 20 20
occupation in the US to the total US population in January 2020 – December 2020.
Fe Fe Fe Fe Fe
b b b b b
20 20 20 20 20
20 20 20 20 20
M M M M M
ar ar ar ar ar
20 20 20 20 20
(h) Education,
2 0 20 20 2 0 2 0
A A A A A
pr pr pr pr pr
2 02 20 20 20 20
0 20 20 20 20
M M M M M
ay ay ay ay ay
20 20 20 20 20
20 20 20 20 20
Ju Ju Ju Ju Ju
n n n n n
20 20 20 20 20
2 0 20 20 20 20
Ju Ju Ju Ju Ju
l2 l2 l2 l2 l2
02 02 02 02 02
0 0 0 0 0
A A A A A
ug ug ug ug ug
20 20 20 20 20
20 20 20 2 0 2 0
ing, & Library
Se Se Se Se Se
p p p p p
20 20 20 20 20
20 20 20 20 20
O O O O O
ct ct ct ct ct
20 20 20 20 20
20 20 20 2 0 2 0
N N N N N
ov ov ov ov ov
Figure 2: Job postings per capita in the US by two-digit (SOC) occupation
20 20 20 20 20
20 20 20 2 0 2 0
D D D D D
ec ec ec ec ec
20 20 20 20 20
2
(p) Sales & Related
0 20 20 20 20
(l) Protective Service
Source: Authors’ calculations based on online job postings data of Burning Glass Technologies (BGT).
Train-
cial ematical gineering
trative Support Forestry traction tenance, & Repair
Notes: Monthly variation in the ratio of the number of online job postings of each two-digit (2010 SOC)
Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%)
n 9 10 11 12 13 n 3 3.5 4 4.5 5 n 1.9 2 2.1 2.2 2.3 n .9 .95 1 1.05 1.1 1.15 n 9 9.5 10 10.5 11
20 20 20 20 20
2 0 20 20 20 20
Fe Fe Fe Fe Fe
b b b b b
20 20 20 20 20
20 20 20 20 20
M M M M M
ar ar ar ar ar
20 20 20 20 20
2 0 20 20 2 0 2 0
A A A A A
pr pr pr pr pr
2 02 20 20 20 20
0 20 20 20 20
M M M M M
ay ay ay ay ay
20 20 20 20 20
20 20 20 20 20
Ju Ju Ju Ju Ju
n n n n n
20 20 20 20 20
2 0 20 20 20 20
Ju Ju Ju Ju Ju
l2 l2 l2 l2 l2
02 02 02 02 02
0 0 0 0 0
A A A A A
ug ug ug ug ug
20 20 20 20 20
20 20 20 2 0 2 0
Se Se Se Se Se
p p p p p
20 20 20 20 20
20 20 20 20 20
O O O O O
ct ct ct ct ct
20 20 20 20 20
20 20 20 2 0 2 0
N N N N N
(i) Arts, Design, Enter- (j) Healthcare Practi- (k) Healthcare Support
ov ov ov ov ov
20 20 20 20 20
2 2
(a) Management
20 20 20 0 0
(e) Life, Physical, & So- (f) Community & So-
D D D D D
ec ec ec ec ec
20 20 20 20 20
2 0 20 20 20 20
Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%)
n 2.6 2.8 3 3.2 3.4 n .09 .095 .1 .105 .11 .115 n 1.8 2 2.2 2.4 n 11 12 13 14 15 n 1.15 1.2 1.25 1.3 1.35 n 4.5 5 5.5 6 6.5
20 20 20 20 20 20
2 0 2 0 20 20 20 20
Fe Fe Fe Fe Fe Fe
b b b b b b
20 20 20 20 20 20
20 20 20 20 20 20
M M M M M M
ar ar ar ar ar ar
(m) Food Preparation (n) Building Cleaning (o) Personal Care &
20 20 20 20 20 20
2 0 2 0 20 20 2 0 2 0
A A A A A A
pr pr pr pr pr pr
2 02 2 02 20 20 20 20
0 0 20 20 20 20
M M M M M M
ay ay ay ay ay ay
20 20 20 20 20 20
20 20 20 20 20 20
Ju Ju Ju Ju Ju Ju
n n n n n n
20 20 20 20 20 20
2 0 2 0 20 20 20 20
Ju Ju Ju Ju Ju Ju
l2 l2 l2 l2 l2 l2
02 02 02 02 02 02
0 0 0 0 0 0
A A A A A A
ug ug ug ug ug ug
20 20 20 20 20 20
20 20 20 20 2 0 2 0
Se Se Se Se Se Se
p p p p p p
(q) Office & Adminis- (r) Farming, Fishing, & (s) Construction & Ex- (t) Installation, Main-
20 20 20 20 20 20
20 20 20 20 20 20
O O O O O O
ct ct ct ct ct ct
20 20 20 20 20 20
20 20 20 20 2 0 2 0
N N N N N N
(u) Production
ov ov ov ov ov ov
20 20 20 20 20 20
20 20 20 20 2 0 2 0
D D D D D D
ec ec ec ec ec ec
tainment, Sports, Me- tioners & Technical
20 20 20 20 20 20
2 0 2 0 20 20 20 20
cial Science cial Service
11
Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%)
n 5 6 7 8 9 n 1 1.2 1.4 1.6 1.8 n 2 2.2 2.4 2.6 2.8 n 2 2.5 3 3.5 n .55 .6 .65 .7 .75 .8 n 7 8 9 10 11 12
20 20 20 20 20 20
2 0 2 0 20 20 20 20
Fe Fe Fe Fe Fe Fe
b b b b b b
20 20 20 20 20 20
20 20 20 20 20 20
M M M M M M
ar ar ar ar ar ar
(b) Business & Finan- (c) Computer & Math- (d) Architecture & En-
20 20 20 20 20 20
2 0 2 0 20 20 2 0 2 0
A A A A A A
pr pr pr pr pr pr
2 2 20 20 20 20
(v) Transportation
02 02
0 0 20 20 20 20
M M M M M M
ay ay ay ay ay ay
20 20 20 20 20 20
20 20 20 20 20 20
Ju Ju Ju Ju Ju Ju
n n n n n n
20 20 20 20 20 20
2 0 2 0 20 20 20 20
Ju Ju Ju Ju Ju Ju
l2 l2 l2 l2 l2 l2
02 02 02 02 02 02
0 0 0 0 0 0
A A A A A A
ug ug ug ug ug ug
20 20 20 20 20 20
20 20 20 20 2 0 2 0
Se Se Se Se Se Se
p p p p p p
(g) Legal
20 20 20 20 20 20
20 20 20 20 20 20
O O O O O O
ct ct ct ct ct ct
20 20 20 20 20 20
20 20 20 20 2 0 2 0
N N N N N N
ov ov ov ov ov ov
Material Moving
20 20 20 20 20 20
20 20 20 20 2 0 2 0
D D D D D D
ec ec ec ec ec ec
20 20 20 20 20 20
2 0 2 0 20 20 20 20
dia
&
& Service & Maintenance Service
Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%) Ja share of online job postings (%)
n 3 3.2 3.4 3.6 3.8 4 n 9 9.5 10 10.5 11 11.5 n 1.2 1.3 1.4 1.5 1.6 n 2.2 2.4 2.6 2.8 3 n 2 2.2 2.4 2.6 2.8
20 20 20 20 20
2 0 20 20 20 20
Fe Fe Fe Fe Fe
b b b b b
20 20 20 20 20
20 20 20 20 20
M M M M M
ar ar ar ar ar
20 20 20 20 20
(h) Education,
2 0 20 20 2 0 2 0
A A A A A
pr pr pr pr pr
2 02 20 20 20 20
0 20 20 20 20
M M M M M
ay ay ay ay ay
20 20 20 20 20
20 20 20 20 20
Ju Ju Ju Ju Ju
n n n n n
20 20 20 20 20
2 0 20 20 20 20
Ju Ju Ju Ju Ju
l2 l2 l2 l2 l2
02 02 02 02 02
0 0 0 0 0
A A A A A
ug ug ug ug ug
20 20 20 20 20
20 20 20 2 0 2 0
ing, & Library
Se Se Se Se Se
p p p p p
20 20 20 20 20
20 20 20 20 20
O O O O O
ct ct ct ct ct
Figure 3: Share of job postings in the US by two-digit (SOC) occupation
20 20 20 20 20
20 20 20 2 0 2 0
N N N N N
ov ov ov ov ov
20 20 20 20 20
20 20 20 2 0 2 0
D D D D D
ec ec ec ec ec
20 20 20 20 20
2
(p) Sales & Related
0 20 20 20 20
(l) Protective Service
Source: Authors’ calculations based on online job postings data of Burning Glass Technologies (BGT).
Train-
cial ematical gineering
trative Support Forestry traction tenance, & Repair
occupation in the US in the total number of online job postings in the US in January 2020 – December 2020. Notes: Monthly variation in the share of the number of online job postings of each two-digit (2010 SOC)
The aforementioned trends had important implications for the occupational composition
of labour demand, as indicated by the evolution of the shares of online job postings of oc-
cupations over the months of 2020 in Panels (a)–(v) of Figure 3.14 With respect to January
2020 (pre-Covid shock), community and social service, healthcare practice, healthcare sup-
port, protective service, farming, fishing and forestry, production, transportation and material
moving, life, physical and social science, and construction and extraction occupations gained
higher shares in the total number of online job postings. These gains naturally translate to
losses for the rest of the occupations.
2.2 Variables for occupational characteristics
The salient heterogeneity in the evolution of the absolute and relative demands for differ-
ent occupations during the Covid-19 pandemic outbreak suggests that different occupational
characteristics might have contributed to it. For instance, the higher shares of online job
postings of community and social service and healthcare practice and support occupations
may be attributed to the fact that such occupations, despite their relatively high physical
contact requirements, are essential during a pandemic. By contrast, other occupations that
are contact-intensive but not essential experienced declines in their shares of online job post-
ings (e.g. personal care and service, food service, arts and entertainment). In addition, it is
by now well understood that occupational characteristics such as routine intensity explain the
secular declines in the relative employment of such occupations due to routine-biased techno-
logical change (RBTC) and offshoring (Autor et al., 2003; Autor & Dorn, 2013; Goos et al.,
2014), as well as the RBTC-induced declines in their relative employment during economic
downturns (e.g. Jaimovich & Siu, 2020).
In order to be able to make an in-depth investigation of how occupational characteristics
influenced firms’ hiring strategies during the Covid-19 pandemic outbreak, we construct vari-
ables capturing a wide range of occupational characteristics. To save on space, we relegate
to Appendix Section A.1 the full description of the data sources and how all measures are
constructed and give here a rather brief account. For each two-digit (2010 SOC) occupation,
we construct measures of: non-routine cognitive analytical, non-routine cognitive interactive,
non-routine manual, routine manual, and routine cognitive task content using the data of Au-
14
These are calculated as the shares of online job postings of occupations in the country-wide number of
online job postings.
12
tor et al. (2003); social, coordination, and interaction task content and degrees of customer-
and service-orientedness using the data of Deming (2017); physical contact intensity using the
data of Leibovici et al. (2020); and tele-work potential as constructed by Dingel & Neiman
(2020) using ONET data (benchmark) or manually (alternative). As the original measures
are available at more disaggregated occupation levels, we aggregate these up to the two-digit
(2010 SOC) level by calculating their unweighted means.15
In line with evidence of the extant literature, occupations such as computer science and
mathematical, architecture and engineering, and life, physical and social science have rela-
tively high non-routine cognitive analytical and interactive task content, while managerial
occupations are among those that have simultaneously relatively high social, coordination,
and interaction task content. Construction and extraction, transportation, protective ser-
vice, and building cleaning occupations have relatively high non-routine manual task content,
while healthcare practice and support are among the occupations with relatively high rou-
tine manual and routine cognitive task content. These occupations along with personal care
and service, sales, community and social care, and protective service occupations also have
relatively high degrees of customer- and service-orientedness (see Tables B1 and B2). Not
surprisingly, healthcare support and practice, food preparation and service, and personal care
and service are the most contact-intensive occupations (column (3) of Table B3), while com-
puter science and mathematical, education, training and library, legal, business and financial,
and managerial occupations have relatively high tele-work potential (columns (4) and (5) of
Table B3).
Finally, to construct a dummy variable indicating whether an occupation is essential or
not, we use the data of Mongey et al. (2021) and classify an occupation as essential if its
employment share in essential industries is at least 60%. Under this approach, the follow-
ing occupations are labeled as essential: community and social service; healthcare practice;
healthcare support; protective service; farming, fishing, and forestry; production; transporta-
tion and material moving. As some retail occupations play an essential role amid a pandemic
(e.g. salespersons in grocery stores), we also label sales and related occupations as essential
in the benchmark version of the dummy variable. In an alternative version, we treat this
occupation as non-essential (see columns (1) and (2), respectively, of Table B3).16
15
Calculating the unweighted medians produces very similar measures.
16
Note that the threshold set by Mongey et al. (2021) for the identification of essential occupations is 75%.
We abstract from doing so, as we would have to exclude from the list occupations that are to a large extent
essential amid a pandemic: community and social service, production, and transportation and material moving.
13
2.3 Covid-induced economic uncertainty
In order to capture the effect of the Covid-19 shock on firms’ hiring strategies, we use a
comprehensive measure of economic uncertainty. Our baseline measure is the news-based
Economic Policy Uncertainty (EPU) index of Baker et al. (2016) for the US.17 Alongside the
baseline measure, we draw information on the composite US EPU index.18 The base value
of the EPU measures is equal to 100. Hence, deviations above (below) this value imply that
uncertainty is above (below) average.
Panels (c) and (d) of Figure 1 display the monthly variation in the news-based and com-
posite measures of EPU in the US over January 2019 – May 2021.19 Economic uncertainty
in 2019 and January 2020 was above average, with the two measures fluctuating within the
ranges of 100–300 and 100–200, but low relative to economic uncertainty in February 2020 –
December 2020. In fact, between February and March, when the spread of the Covid-19 virus
started having the features of a pandemic, economic uncertainty rose sharply (the news-based
EPU increased from 216 to 426 and the composite EPU increased from 161 to 283). Although
it declined slightly in April, it rose again in May and reached its peak by that month. Between
June and August –when lockdown and other restrictive measures had been lifted or relaxed, it
fluctuated at lower levels, which were still higher than the pre-pandemic ones, and was again
on a rising trend as of October, when the second wave of the pandemic hit the country. It
is only in February 2021 when economic uncertainty went down to its pre-pandemic levels
and remained at such levels in the next few months. The striking differences in the level
of economic uncertainty in the US between the months marked by the Covid-19 crisis and
those that preceded it likely suggest that the rising economic uncertainty in 2020 was induced
primarily by this shock, rather than other idiosyncratic factors.20
For instance, in order for meat and diary products produced by individuals in farming occupations to reach
the final consumer, they have to be transported to the relevant store for sale. This implies that the services
of individuals in transportation occupations (e.g. truck driver) and production occupations (e.g. butcher) are
indispensable parts of the supply chain.
17
For this particular country, the news-based EPU is also available by individual policy area, namely: health
policy, fiscal policy, government spending, tax policy, monetary policy, national security, entitlement pro-
grammes, regulation, financial regulation, trade policy, and sovereign debt and currency management. For a
detailed account of the construction of these measures, see Appendix Section A.4 and Baker et al. (2016).
18
One of the components of this measure is the news-based EPU. For details about the concept of this
measure and the steps taken for its construction, see Appendix Section A.4.
19
We have included the values of EPU in all months of 2019 and the first five months of 2021 in order to
have measures of comparison when discussing the evolution of EPU over the months of 2020. The same logic
applies to the rest of the figures that display the values of variables for months that are adjacent to those of
2020.
20
The EPU measures for different policy areas exhibited very similar trends in 2020 except for trade policy
uncertainty that was on a declining path overall (Panels (a)–(k) of Figure B2). The latter is a reversal of the rise
14
In addition, we use alternative measures of economic uncertainty, whose construction is
predicated upon US firms’ probabilistic expectations about their sales revenue and employ-
ment growth four quarters ahead, rather than upon information from US newspapers. Infor-
mation on firms’ subjective expectations is originally collected through the Survey of Business
Uncertainty (SBU), which is conducted jointly by the Atlanta FED, Stanford University, and
the University of Chicago Booth School of Business.21 As shown in Panels (e) and (f) of
Figure 1, the measures of sales revenue and employment growth uncertainty exhibited very
similar trends in the months of 2020 to those of the EPU measures. The consistency in the
trends of EPU and SBU measures in the months of 2020 is extra evidence on their capacity
to capture Covid-induced economic uncertainty.22
In sum, the descriptive statistics analysis reveals that online job postings and economic
uncertainty moved in opposite directions during the Covid-19 pandemic outbreak: two slumps
and a moderate recovery in the per capita number of online job postings coincided with
equivalent spikes and a moderate decline in the measures of economic uncertainty. In addition,
there is a great deal of qualitative and quantitative heterogeneity in the variation of online
job postings per capita and share of online job postings across occupations, and substantial
differences across occupations in certain characteristics. Our goal in the econometric analysis
is to identify the differential effects of Covid-induced economic uncertainty on the demand
in trade policy uncertainty since the revision of US trade policy by the Trump administration and the initiation
of the so-called US-China “trade war”. Not surprisingly, the largest spikes in March and April corresponded
to uncertainty about health policy: the relevant EPU rose from roughly 155 in February to 771 in March to
1031 in April. Nevertheless, the sizeable spikes in uncertainty about other policy areas clearly demonstrates
the numerous negative economic side-effects of the Covid-19 pandemic outbreak. Health-related EPU also
rose slightly in December, amid the second wave of the Covid-19 pandemic. In that month, the most sizeable
spikes are observed for EPU related to entitlement programmes (e.g. unemployment insurance, welfare state
benefits) aiming at (partly) compensating for the job and income losses incurred by individuals since the onset
of the crisis, and to debt and currency management as a result of the expansionary fiscal measures coupled
with public revenue losses. By contrast, the EPU for other policy areas declined in that month. Given the
universality of the shock, we would also expect to observe similar trends in economic uncertainty at the global
level and in other individual countries. We show that this was indeed the case by plotting the Global EPU
measure and the unweighted mean of EPU measures of individual countries other than the US from January
2019 to May 2021 (Panels (a) and (b) of Figure B3). Information on these EPU measures is retrieved from the
same source as for the US measures. The individual countries for which we calculate the unweighted average
are: Australia, Brazil, Canada, Chile, China, Colombia, France, Germany, Greece, India, Ireland, Italy, Japan,
Korea, Netherlands, Russia, Spain, Singapore, UK, Sweden, and Mexico.
21
For detailed information about the survey and the SBU measures, see Altig et al. (2020) and
https://www.atlantafed.org/research/surveys/business-uncertainty.
22
Naturally, the measures of sales revenue and employment growth expectations exhibited opposite trends
(Panels (a) and (b) of Figure B4). Also, there are strong and highly statistically significant correlations between
the main (i.e., news-based) EPU measure and the other measures of economic uncertainty under consideration.
The magnitudes of the correlations are: 97% (composite EPU), 92% (Global EPU), 89% (Global EPU PPP),
86% (cross-country mean of EPU), 86% (cross-country median of EPU), 53% (SBU sales revenue growth
uncertainty), 73% (SBU employment growth uncertainty), -61% (SBU sales revenue growth expectations), and
-43% (SBU employment growth expectations).
15
for occupations that have pre-Covid occupational structure of US states and a wide range of
occupational characteristics.
3 Econometric model
In order to study the impact of Covid-induced economic uncertainty on the relative demand
for occupations with different characteristics, we estimate the following specification:
OJPost = αos + αt + βEP Uost ∗ EP Uost + βEP Ut ,char ∗ EP Ut ∗ charo
(1)
+ βEP Uost ,char ∗ EP Uost ∗ charo + ost
where OJPost is the share of online job postings of occupation o in US state s in month t of
the year 2020. The specification includes occupation-US state fixed effects, which account for
time-invariant unobserved heterogeneity across occupation-state pairs, such as initial differ-
ences in the demographics, human capital, and real disposable income of individuals. It also
includes month fixed effects, which account for time-varying shocks to all occupation-state
pairs, namely, the aggregate business cycle, the loosening of US FED’s monetary policy and
expansion of quantitative easing programmes, and the US federal government’s fiscal and
labour market policies (e.g. CARES act).
The month dummies also capture the onset of the Covid-19 shock and the timings of
the imposition and relaxation of the associated restrictive measures, albeit not their severity.
However, as shown in Section 2.3, a critical aspect of the Covid-19 shock is the elevated
economic uncertainty, which is likely to have impacted the hiring decisions of firms. For this
reason, we incorporate in the specification our baseline measure of Covid-induced economic
uncertainty, namely, the news-based EPU in the US (EP Ut ). To apportion this measure to
each occupation-state pair, we multiply it by the share of employment of occupation-state
E
pair os in country-wide employment of occupation o in 2019: EP Uost = EP Ut ∗ EoUos,2019
S,2019
.
Information on occupation-state-level employment for 2019 is drawn from the US Bureau of
Labour Statistics (BLS). By interacting this variable with a variable for a certain occupational
characteristic, we obtain the main term of interest of the specification (EP Uost ∗ charo ). The
relevant coefficient estimate, βEP Uost ,char , captures the differential effect of Covid-induced
economic uncertainty on the relative demand for a given occupation in a given US state,
16
according to its pre-Covid country-wide occupational employment share and the intensity or
status of a certain characteristic (e.g. service-orientedness, physical contact, essentiality). As
common in the empirical literature, we facilitate the economic interpretation of this coefficient
estimate by standardising the relevant variable, so that its mean equals zero and its standard
deviation equals one. We also include EP Uost and EP Ut ∗ charo in the specification, as these
are components of the interaction term that are not captured by the fixed effects. Unobserved
factors impacting the demand for occupation o in US state s in month t are captured by the
error term, ost .
Eq. 1 is estimated by OLS. The coefficient estimates of the main explanatory variables
are unlikely to be subject to a simultaneity bias for the following reasons. The country-wide
employment share of occupation-state pairs is held constant and its values correspond to
a year that precedes that highly unpredictable Covid-19 shock. Although the time-varying
country-wide EPU could impact the share of online job postings of an occupation in a certain
US state, it is highly unlikely, especially due to the size of the US economy, that the expansion
or contraction of online job postings of an occupation in a given state has the potential to
feedback into country-wide economic uncertainty. As for the variables capturing different
occupational characteristics, these are by construction time-invariant and thus, cannot be
determined by monthly changes in the online job postings of an occupation-state pair. Given
the sets of fixed effects in the main specification, a possible threat to the identification strategy
may stem from policy shocks to US states (e.g. adoption of different anti-Covid policies and
measures, time differences in the imposition or relaxation of anti-Covid measures). To address
the possibility of an omitted variable bias in the specifications, we conduct a robustness
exercise in which we replace the month fixed effects with state-month fixed effects.
4 Econometric results
4.1 Covid-induced EPU and occupational demand
We kick-off the econometric analysis by identifying the effects of Covid-induced economic
uncertainty on the absolute and relative occupational demand in US states without accounting
for differences across occupations in the intensity or status of a given characteristic. To this
end, we estimate Eq. 1 by OLS after dropping EP Ut ∗ charo and EP Uost ∗ charo and focusing
on the interpretation of the coefficient estimate of EP Uost . As the dependent variable, we use
17
the log of the number of online job postings or the number of online job postings per capita of
occupation-state pairs, which measure absolute occupational demand, or the share of online
job postings of occupation-state pairs, which measures relative occupational demand. The
results are shown in columns (1) and (4), (2) and (5), and (3) and (6), respectively, of Panel
A of Table 1. The main explanatory variable in the first three columns is constructed based
on the main (i.e., news-based) EPU, while in the last three columns, it is constructed based
on the composite EPU. In all six columns, the relevant coefficient estimate is negative and
statistically significant at 1% or 5%. This suggests that occupations in US states that were
more exposed to Covid-induced economic uncertainty experienced disproportionate declines
in the log of their number of online job postings, their online job postings per capita, and their
shares of online job postings. That is, the elevated uncertainty and the ensuing reluctance or
incapacity of firms to proceed to new hires hit disproportionately occupations in US states that
were over-represented at country level pre-Covid. In terms of size, the coefficient estimates
suggest that an increase of one standard deviation in the exposure of the average occupation-
state pair to Covid-induced economic uncertainty decreased the number of online job postings
per capita by 0.49 or 0.56 percentage points (columns (2) and (5)) and the share of online job
postings by 0.12 or 0.074 percentage points (columns (3) and (6)).23
In additional estimations, we confirm the robustness of the results to using different mea-
sures of economic uncertainty for the construction of the main explanatory variable. In par-
ticular, we use the Global EPU or Global EPU PPP (columns (1)–(3) and (4)–(6) of Panel B
of Table 1), the mean or median of EPU across countries other than the US (columns (1)–(3)
and (4)–(6) of Panel A of Table C1), the SBU sales revenue or employment growth uncertainty
(columns (1)–(3) and (4)–(6) of Panel C of Table 1), and the SBU sales revenue or employ-
ment growth expectations (columns (1)–(3) and (4)–(6) of Panel B of Table C1). In all cases,
the estimations yield very similar results qualitatively. By and large, though, using the alter-
native economic uncertainty measures yields even stronger negative effects of Covid-induced
economic uncertainty on occupation-state pairs that had higher country-wide occupational
23
Very similar results are obtained when we re-estimate the equations in columns (2) and (3) of Panel A
using the news-based EPU by policy area (e.g. health, fiscal, monetary) for the construction of the main
explanatory variable. In particular, we identify disproportionately negative effects on the per capita number of
online job postings and the share of online job postings of occupation-state pairs by Covid-induced uncertainty
in all economic areas except for trade policy and debt and currency management. Trade policy uncertainty
impacted positively the per capita number of online job postings, while it exerted no statistically significant
effect on the share of online job postings. Debt and currency management uncertainty exerted no statistically
significant effects on either the per capita number of online job postings or the share of online job postings
(Panels (a) and (b) of Figure C1).
18
employment shares pre-Covid. For instance, an increase of one standard deviation in the ex-
posure of the average occupation-state pair to Covid-induced economic uncertainty (based on
SBU employment growth uncertainty) decreased the number of online job postings per capita
by 1 percentage point (column (5) of Panel C of Table 1), while an increase of one standard
deviation in the exposure of the average occupation-state pair to Covid-induced economic
uncertainty (based on the cross-country median EPU) decreased the number of online job
postings per capita by 0.23 percentage points (column (5) of Panel A of Table C1).24
Table 1: Covid-induced economic uncertainty and occupational demand
Panel A: US news-based and composite EPU
(1) (2) (3) (4) (5) (6)
Dep. var.: log # of OJP OJP p.c. OJP share log # of OJP OJP p.c. OJP share
EP Uost (main) -0.017*** -0.0049*** -0.0012***
[0.005] [0.0004] [0.0003]
EP Uost (alt) -0.017** -0.0056*** -0.00074*
[0.007] [0.0005] [0.0004]
Observations 13,464 13,464 13,464 13,464 13,464 13,464
R2 0.987 0.929 0.953 0.987 0.929 0.953
Panel B: Global EPU
(1) (2) (3) (4) (5) (6)
Dep. var.: log # of OJP OJP p.c. OJP share log # of OJP OJP p.c. OJP share
Global EP Uost -0.021** -0.0081*** -0.0014***
[0.009] [0.0007] [0.0005]
Global EP Uost PPP -0.022** -0.0086*** -0.0014***
[0.009] [0.0007] [0.0005]
Observations 13,464 13,464 13,464 13,464 13,464 13,464
R2 0.987 0.929 0.953 0.987 0.929 0.953
Panel C: US sales revenue and employment growth uncertainty
(1) (2) (3) (4) (5) (6)
Dep. var.: log # of OJP OJP p.c. OJP share log # of OJP OJP p.c. OJP share
SBUost (sales gr. unc.) -0.011 -0.0060*** -0.00028
[0.008] [0.0006] [0.0005]
SBUost (emp. gr. unc.) -0.035*** -0.010*** -0.00068
[0.01] [0.0008] [0.0006]
Observations 13,464 13,464 13,464 13,464 13,464 13,464
R2 0.987 0.929 0.953 0.987 0.930 0.953
Notes: OLS estimations with robust standard errors in all columns of all panels. Two-digit (SOC) occupation-US state and month fixed effects
are included in the equations. Asterisks denote significance at 1% (***), 5% (**), and 10% (*).
24
To remove any remaining concerns over the capacity of the EPU measures to capture Covid-induced
economic uncertainty, we also conduct the following exercise. Instead of using their values for the months
of 2020, we use their values for the corresponding months of the year before (2019) while maintaining the
2020 values for the outcome variable. If the principal driver of economic uncertainty in the months of 2020
were not the Covid-19 shock, but other factors that were very similar to those having contributed to economic
uncertainty in the months of 2019 (when the Covid-19 shock hadn’t hit yet), then this exercise should yield
very similar results qualitatively to those Table 1. We find that this is not actually the case: the coefficient
estimates that we obtain from these estimations are positive, rather than negative as in the main estimations,
and are mostly statistically significant at 1% (Panels C and D of Table C1 and Panels (c) and (d) of Figure
C1).
19
4.2 Accounting for occupational characteristics
As highlighted in Section 2.2, occupations differ substantially in the intensity or status of
certain characteristics and this source of heterogeneity can potentially shape the effects of
Covid-induced economic uncertainty on their relative demand. Therefore, we now proceed
to the core of the analysis by estimating Eq. 1, with the share of online job postings of
occupation-state pairs as the dependent variable. In these estimations, our main focus is
on the interpretation of the coefficient estimate of EP Uost ∗ charo . We first consider the
measures of non-routine cognitive analytical, non-routine cognitive interactive, non-routine
manual, routine manual, and routine cognitive task content of occupations that we have
created using the relevant data of Autor et al. (2003) (columns (1)–(5) of Panel A of Table 2).
These occupational characteristics have been widely used in the stream of literature providing
evidence on job polarisation. However, the Covid-19 pandemic outbreak and the associated
restrictive measures (e.g. lockdown, social distancing) have generated a new setting to which
additional occupational characteristics are likely to be at least as pertinent. Therefore, we also
consider the measures of customer- and service-orientedness, physical contact intensity, and
tele-work potential that we have created using the relevant data of Deming (2017), Leibovici
et al. (2020), and Dingel & Neiman (2020), respectively, as well as the essential status of
occupations that we identify using the relevant data of Mongey et al. (2021) (columns (1)–(5)
of Panel B of Table 2).
Same as in the previous section, the negative and significant coefficient estimates of EP Uost
in columns (1)–(3) of Panel A and column (3) of Panel B suggest that Covid-induced economic
uncertainty exerted disproportionately negative effects on the relative demand for occupations
in US states that had higher country-wide occupational employment shares pre-Covid. How-
ever, the coefficient estimates of the main interaction term of interest, EP Uost ∗ charo , suggest
that positive effects were actually exerted on the relative demand for occupations that have
relatively high non-routine cognitive interactive task content (column (2) of Panel A), non-
routine manual task content (column (3) of Panel A), or physical contact intensity (column
(3) of Panel B). The first two effects are consistent with existing evidence pointing to job
and skill polarisation (e.g. Autor et al., 2003; Acemoglu & Autor, 2011; Autor & Dorn, 2013;
Michaels et al., 2014; Blanas, 2021). The third effect is likely explained by the context created
by the Covid-19 shock, as individuals in some contact-intensive occupations –most notably,
healthcare practice and support– are in the front line to tackle the major health crisis that
20
this shock has brought about.
Table 2: Covid-induced economic uncertainty and relative occupational demand
Panel A: Cognitive analytical/interactive, non-routine manual, routine cognitive/manual
(1) (2) (3) (4) (5)
Dep. var.: OJP share OJP share OJP share OJP share OJP share
EP Uost -0.0025** -0.0017*** -0.0023*** -0.0014 -0.00068
[0.001] [0.0004] [0.0004] [0.002] [0.0005]
EP Ut * cogn. analytical -0.00025
[0.0004]
EP Uost * cogn. analytical 0.0014
[0.001]
EP Ut * cogn. interactive 0.000020
[0.0003]
EP Uost * cogn. interactive 0.00086**
[0.0004]
EP Ut * non-routine manual 0.0020***
[0.0004]
EP Uost * non-routine manual 0.0019**
[0.0008]
EP Ut * routine manual 0.00013
[0.0006]
EP Uost * routine manual 0.00027
[0.002]
EP Ut * routine cognitive 0.00093***
[0.0002]
EP Uost * routine cognitive -0.00059
[0.0004]
Observations 13,464 13,464 13,464 13,464 13,464
R2 0.953 0.953 0.954 0.953 0.953
Panel B: Customer/Service-orientedness, physical contact intensity, tele-work, essential
(1) (2) (3) (4) (5)
Dep. var.: OJP share OJP share OJP share OJP share OJP share
EP Uost 0.0011 -0.0016 -0.0061*** 0.00040 -0.0022***
[0.001] [0.001] [0.002] [0.0007] [0.0003]
EP Ut * customer -0.0022***
[0.0004]
EP Uost * customer -0.0022**
[0.001]
EP Ut * service -0.00042
[0.0003]
EP Uost * service 0.00048
[0.001]
EP Ut * contact 0.00033
[0.0006]
EP Uost * contact 0.0050***
[0.002]
EP Ut * tele-work -0.00028
[0.0004]
EP Uost * tele-work -0.0020***
[0.0007]
EP Ut * essential -0.00032
[0.0004]
EP Uost * essential 0.0020***
[0.0005]
Observations 13,464 13,464 13,464 13,464 13,464
R2 0.954 0.953 0.953 0.953 0.953
Notes: OLS estimations with robust standard errors in all columns of both panels. Two-digit (SOC) occupation-US state and month
fixed effects are included in the equations. Asterisks denote significance at 1% (***), 5% (**), and 10% (*).
21
In addition, while the effect on the relative demand for essential occupations is positive,
that on the relative demand for non-essential occupations is negative (column (5) of Pane B).
This result is likely explained by the the need for unhampered provision of essential goods
and services (e.g. groceries, healthcare) by individuals in some occupations amid a major
health crisis and the imposition of unconventional restrictive measures (e.g. lockdown, social
distancing). The negative and significant coefficient estimates of the main interaction term
in columns (1) and (4) of Panel B suggest that Covid-induced economic uncertainty exerted
disproportionately negative effects on the relative demand for occupations with relatively high
degree of customer-orientedness and those with relatively high tele-work potential. The first
of the two effects is likely driven by customer-oriented occupations that were hit particularly
hard by the imposition of restrictive measures (e.g. food service, personal care and service).
Albeit surprising, similar evidence to the second effect has been provided by Forsythe et
al. (2020). This effect likely suggests that, due to the elevated uncertainty and low expected
demand (Baker, Bloom, Davis, & Terry, 2020), firms were reluctant or incapable of prioritising
the hiring of new employees who can perform a relatively large fraction of their tasks from
distance. Finally, the effects on the relative demand for occupations with relatively high
routine manual or routine cognitive task content, and for service-oriented occupations are
statistically insignificant (columns (4) and (5) of Panel A and column (2) of Panel B).
Importantly, the results of this section are robust to the use of the alternative measures
of occupational characteristics (Panels A and B of Table C2). Specifically, we aggregate the
measures of Autor et al. (2003), Deming (2017), and Leibovici et al. (2020) up to the two-
digit (2010 SOC) level by calculating the unweighted median of their values, rather than the
unweighted mean, which is the benchmark. As an alternative measure of tele-work potential,
we use the manual version of Dingel & Neiman (2020). To account for the fact that some
occupations in the sales and related occupation category are not essential (e.g. salesperson in
apparel store), we use the alternative dummy variable for the essential status of occupations,
which labels this occupation category as non-essential. In a second robustness check, we use
the measures of the abstract (non-routine cognitive), non-routine manual, and routine task
content that we create using the data of Autor & Dorn (2013).25 The estimation results
obtained bear a very close resemblance to those of the main analysis (Table C3). In an
additional check, we ensure that the estimates are not spuriously driven by time-varying
shocks to US states, most notably, differences in the type of anti-Covid measures or the
25
For the construction of these measures, see Appendix Section A.1.
22
timing of their introduction or removal by incorporating dummies for US state-month pairs
in lieu of month dummies (Table C4).
From the analysis in Table 2, we conclude that while some of our findings are consistent
with the main findings and associated rationale of the extant literature on routine-biased
technological change (RBTC) and job polarisation, other findings are novel and can primarily
be rationalised by the conditions generated by the Covid-19 pandemic outbreak. To be able
to get more insights about these, we consider in the remaining sections of the econometric
analysis different pairwise combinations of occupational characteristics.
4.2.1 The joint role of non-routine cognitive analytical and interactive tasks
Building upon the literature on RBTC and job polarisation, Deming (2017) has shown that
over 1980–2010 occupations that have both relatively high non-routine cognitive analytical
and non-routine cognitive interactive task content experienced the largest gains in employment
shares, followed by occupations that have relatively high non-routine cognitive interactive task
content and relatively low non-routine cognitive analytical task content. These gains trans-
lated to relatively large employment share losses for occupations with both relatively low
non-routine cognitive analytical and interactive task content and relatively small losses for
occupations with relatively high non-routine cognitive analytical task content but relatively
low non-routine cognitive interactive task content. Evidence on the increasing complemen-
tarity between cognitive analytical and cognitive interactive tasks has also been uncovered
by Borghans et al. (2014) and Weinberger (2014). The underlying rationale is that machines
(e.g. computers, robots, AI) do not only undertake an increasing range of routine tasks, but
also of non-routine cognitive analytical tasks. By contrast, non-routine cognitive interactive
tasks are largely shielded from automation (Autor, 2015).
Motivated by this evidence, we incorporate in the equation of column (1) of Table 2 a triple
interaction term, which is the multiplication of the main interaction term by the measure of
non-routine cognitive interactive task content, based on the relevant data of Autor et al.
(2003), or the measure of the social, coordination, or interaction task content, based on the
relevant data of Deming (2017).26 The estimation results are displayed in Panels (a)–(d) of
Figure 4. Although the effect of Covid-induced economic uncertainty on the relative demand
26
These specifications also include the interactions of the country-level EPU with individual characteristics
and pairwise combinations of these. This applies to all the specifications with triple interaction terms that
follow. Their coefficient estimates are not disclosed for the sake of exposition.
23
for occupations with both relatively high non-routine cognitive analytical and interactive or
coordination task content is statistically insignificant (Panels (a) and (c)), we identify positive
and statistically significant effects on the relative demand for occupations that have both
relatively high non-routine analytical and social or interaction task content (Panels (b) and
(d)). Hence, similarly to the 1980–2010 period examined by Deming (2017), we show that the
joint role of analytical and interactive tasks was also crucial in shaping the relative demand
for occupations amid the Covid-19 pandemic outbreak.
Figure 4: The joint role of non-routine cognitive analytical and interactive tasks
EPU (main) EPU (main)
EPU (main) * cogn. analytical EPU (main) * cogn. analytical
EPU (main) * cogn. interactive EPU (main) * social
EPU (main) * cogn. analytical * cogn. interactive EPU (main) * cogn. analytical * social
−.005 0 .005 −.02 −.01 0 .01 .02
(a) Cognitive interactive (b) Social
EPU (main) EPU (main)
EPU (main) * cogn. analytical EPU (main) * cogn. analytical
EPU (main) * coordination EPU (main) * interaction
EPU (main) * cogn. analytical * coordination EPU (main) * cogn. analytical * interaction
−.01 −.005 0 .005 .01 −.04 −.02 0 .02 .04
(c) Coordination (d) Interaction
Notes: OLS estimations with robust standard errors and two-digit (SOC) occupation-US state and month
fixed effects in all panels. These specifications also include the interactions of the country-level EPU with
individual characteristics and pairwise combinations of these. Their coefficient estimates are not disclosed for
the sake of exposition.
24
4.2.2 The joint role of occupational characteristics and essential status
Given the essential role of some occupations during the pandemic outbreak and the subsequent
lockdowns, we consider next the joint role of the main occupational characteristics with the
essential status of occupations. To this purpose, we re-estimate the equations in columns
(1)–(5) of Panel A and columns (1)–(3) of Panel B of Table 2 after multiplying their main
interaction terms by the dummy variable indicating whether an occupation is essential or not.
The results are shown in Figure 5.
Covid-induced economic uncertainty exerted positive effects on the relative demand for
non-essential occupations with relatively high non-routine cognitive analytical, non-routine
cognitive interactive, and non-routine manual task content (Panels (a)–(c)). By contrast, it
impacted positively (negatively) the relative demand for (non-)essential occupations with rel-
atively high routine manual or routine cognitive task content (Panels (d) and (e)). Although
the effects on the relative demand for non-essential occupations are in line with the existing
evidence on job polarisation, the effects on the relative demand for the essential routine-
intensive occupations are not. Instead, the latter are most likely accounted for by the health
emergency and the ensuing rising relative demand for healthcare practice and support occupa-
tions that have in fact relatively high routine manual and routine cognitive task content (see
columns (4) and (5) of Table B1). They may also be accounted for by production occupations
that also have relatively high routine manual and routine cognitive task content and some
of these (e.g. butchers, grocery store staff, warehouse workers) played a critical role for the
unhampered supply of essential products (e.g. food) to consumers amid the Covid-19 shock
and, especially, the lockdowns.
In addition, Covid-induced economic uncertainty impacted positively the relative demand
for occupations that are both essential and contact-intensive (Panel (g)). This effect sheds
more light on the result in column (3) of Table 2 and is most likely explained by the rising
relative demand for healthcare practice and support and protective service occupations, which
are both essential amid a pandemic crisis and require intensive physical contact (see columns
(1)–(4) of Table B3). By contrast, there are no statistically significant effects of Covid-induced
economic uncertainty on the relative demand for essential customer- or service-oriented occu-
pations (Panels (f) and (h)).
25
Figure 5: The joint role of occupational characteristics and essential status
EPU (main) EPU (main)
EPU (main) * cogn. analytical EPU (main) * cogn. interactive
EPU (main) * essential EPU (main) * essential
EPU (main) * cogn. analytical * essential EPU (main) * cogn. interactive * essential
−.005 0 .005 .01 −.004 −.002 0 .002 .004
(a) Cognitive analytical (b) Cognitive interactive
EPU (main) EPU (main)
EPU (main) * non−routine manual EPU (main) * routine cognitive
EPU (main) * essential EPU (main) * essential
EPU (main) * non−routine manual * essential EPU (main) * routine cognitive * essential
−.004 −.002 0 .002 −.004 −.002 0 .002 .004
(c) Non-routine manual (d) Routine cognitive
EPU (main) EPU (main)
EPU (main) * routine manual EPU (main) * customer−oriented
EPU (main) * essential EPU (main) * essential
EPU (main) * routine manual * essential EPU (main) * customer−oriented * essential
−.01 −.005 0 .005 .01 −.005 0 .005
(e) Routine manual (f) Customer orientedness
EPU (main) EPU (main)
EPU (main) * contact EPU (main) * service−oriented
EPU (main) * essential EPU (main) * essential
EPU (main) * contact * essential EPU (main) * service−oriented * essential
−.01 −.005 0 .005 .01 −.005 0 .005
(g) Physical contact intensity (h) Service orientedness
Notes: OLS estimations with robust standard errors and two-digit (SOC) occupation-US state and month fixed
effects in all panels. These specifications also include the interactions of the country-level EPU with individual
characteristics and pairwise combinations of these. Their coefficient estimates are not disclosed for the sake of
exposition.
26
4.2.3 The joint role of occupational characteristics and physical contact intensity
Since the onset of the Covid-19 pandemic outbreak, working in physical contact or close
proximity to others has become a major issue leading to the adoption of social-distancing
practices. In order to better understand the implications of this phenomenon for occupational
demand, we now look into the joint role of the main occupational characteristics and physical
contact intensity. We do so by re-estimating the equations in columns (1)–(5) of Panel A
and column (2) of Panel B of Table 2 after multiplying their main interaction terms by the
measure of physical contact intensity. The results of these estimations are shown in Figure 6.
While Covid-induced economic uncertainty exerted no statistically significant effects on
the relative demand for contact-intensive occupations with relatively high non-routine cog-
nitive analytical and interactive task content (Panels (a) and (b)), it exerted a negative and
significant effect on the relative demand for contact-intensive occupations with relatively high
non-routine manual task content and a positive and significant effect on the relative demand for
less contact-intensive occupations with relatively high non-routine manual task content (Panel
(c)). The two effects are novel and add new insights about how occupational demand during
the Covid-19 shock might have changed with respect to the recent pre-Covid period, which
was largely marked by the job polarisation phenomenon. Specifically, in the the pre-Covid
period the relative demand for low-paying service occupations (e.g. waiters, hairdressers,
plumbers) increased as their tasks were largely shielded from offshoring and automation due
to their relatively high physical contact intensity and non-routine manual task content. By
contrast, the identified effects during the Covid-19 shock are most likely explained by the
declining demand for occupations such as food service and arts and entertainment, which are
both contact-intensive and have relatively high non-routine manual task content, as well as
by the rising demand for occupations such as farming, fishing, and transportation, which are
less-contact intensive, but have relatively high non-routine manual task content (see column
(3) of Tables B1 and B3).
What is more, Covid-induced economic uncertainty exerted positive (negative) effects on
the relative demand for (less) contact-intensive occupations that have relatively high routine
manual or routine cognitive task content, or high degree of service-orientedness (Panels (d)–(f)
of Figure 6). Again, these findings point to important changes in the occupational composition
of US labour demand compared to the recent pre-Covid period. While the negative effects
are consistent with the relative employment contraction in the recent pre-Covid period of less
27
Figure 6: The joint role of occupational characteristics and physical contact intensity
EPU (main) EPU (main)
EPU (main) * cogn. analytical EPU (main) * cogn. interactive
EPU (main) * contact EPU (main) * contact
EPU (main) * cogn. analytical * contact EPU (main) * cogn. interactive * contact
−.02 −.01 0 .01 .02 −.015 −.01 −.005 0 .005 .01
(a) Cognitive analytical (b) Cognitive interactive
EPU (main) EPU (main)
EPU (main) * non−routine manual EPU (main) * routine cognitive
EPU (main) * contact EPU (main) * contact
EPU (main) * non−routine manual * contact EPU (main) * routine cognitive * contact
−.02 −.01 0 .01 .02 −.02 −.01 0 .01 .02
(c) Non-routine manual (d) Routine cognitive
EPU (main) EPU (main)
EPU (main) * routine manual EPU (main) * service−oriented
EPU (main) * contact EPU (main) * contact
EPU (main) * routine manual * contact EPU (main) * service−oriented * contact
−.05 0 .05 −.04 −.02 0 .02 .04
(e) Routine manual (f) Service orientedness
Notes: OLS estimations with robust standard errors and two-digit (SOC) occupation-US state and month
fixed effects in all panels. These specifications also include the interactions of the country-level EPU with
individual characteristics and pairwise combinations of these. Their coefficient estimates are not disclosed for
the sake of exposition.
28
contact-intensive occupations that have relative high routine manual or cognitive task content
(e.g. production and assembly line staff, administrative and clerical staff), the positive effects
are most likely explained by the increase in the relative demand for community and service,
healthcare practice and support, protective service, and essential sales occupations. Indeed,
all these occupations are highly contact-intensive and have relatively high routine manual or
cognitive task content, or high degree of service-orientedness (see columns (4)–(5) of Table
B1, column (5) of Table B2, and column (3) of Table B3).
5 Conclusion
The Covid-19 shock has caused large-scale upheaval in the labour markets around the world,
including in the US, but the adverse consequences are likely to have been borne dispropor-
tionately by some individuals, occupations, and sectors. This paper is the first to conduct a
comprehensive analysis on the effects of Covid-induced economic uncertainty on the relative
demand for occupations that differ in a broad range of characteristics. The period that we ex-
amine covers all months (January–December) of 2020, which were marked by the onset of the
Covid-19 pandemic outbreak, the rapid spread of the virus across and within countries, and
the imposition of restrictive measures for its containment (e.g. two strict lockdowns, domestic
and international travel restrictions or bans, social distancing). To conduct the analysis, we
combine monthly online job postings data at the occupation-US state level with data on fixed
occupational characteristics and monthly data on measures of economic uncertainty in the US
(Economic Policy Uncertainty, Survey Business Uncertainty). We apportion the latter mea-
sures to occupation-state pairs according to their pre-Covid (2019) country-wide occupational
employment shares.
From the analysis, we obtain results that are consistent with the secular phenomenon of
RBTC leading to “job polarisation” (e.g. Autor et al., 2003; Autor & Dorn, 2013) and stronger
complementarity between analytical and social tasks (Borghans et al. (2014); Weinberger
(2014); Deming, 2017), but also with its episodic aspect implying its acceleration during
recessions (Hershbein & Kahn, 2018). Specifically, we find that Covid-induced economic
uncertainty increased the relative demand for occupations with relatively high non-routine
cognitive analytical, non-routine cognitive interactive, and non-routine manual task content.
These effects hold particularly for occupations that are also non-essential. Positive effects
29
are also identified on the relative demand for occupations that have both relatively high
non-routine cognitive analytical and social or interaction task content.
Importantly, though, additional findings are rationalised by the peculiarity of the Covid-19
shock, that is, the major health crisis and restrictive measures (e.g. lockdown, social distanc-
ing) that it has brought about. In particular, we find that Covid-induced economic uncertainty
decreased the relative demand for customer-oriented occupations (e.g. food preparation and
service, personal care and service), while it increased the relative demand for essential or
contact-intensive occupations with relatively high routine manual or routine cognitive task
content, as well as occupations that are both essential and contact-intensive and those that
are both contact-intensive and service-oriented (e.g. healthcare practice and support, protec-
tive service, community and social service).
Whether the latter effects are temporary or will lead to structural shifts in the occupational
composition of US labour demand remains an open question and will depend on various
factors. For instance, tele-presence has become common since the onset of the pandemic and
tele-work is now viewed as a conventional human resource practice that will likely continue
even after the pandemic ends (Barrero et al., 2021). This is also evident by the rising patent
applications in the US between January 2020 and September 2020 for digital technologies that
prop-up tele-work, such as tele-presence software and equipment (Bloom et al., 2021). Tele-
presence is a form of automation (Mindell, 2015), and firms –especially, the relatively large
ones that can incur more easily the sunk costs of irreversible investments– might have already
shifted disproportionately their capital investments towards other more costly automation
technologies (e.g. industrial robots, Artificial Intelligence systems). If this is indeed the
case, then the increasingly automation-intensive future may arrive sooner than previously
anticipated and will determine accordingly the occupational composition of labour demand
in the US and other countries (Autor & Reynolds, 2020). This is an intriguing research topic
and merits in-depth investigation.
Another factor to consider is the de-densification of cities that started with the pandemic
and was facilitated by the adoption of tele-work practices (Ramani & Bloom, 2021). Until
before the pandemic hit, large cities were attractors of talented individuals in high-paying
occupations (e.g. managerial, business and financial, computer science, architecture and engi-
neering, legal) and highly-innovative sectors (e.g. ICT, pharmaceuticals, biotechnology), who
were concentrated in them in order to benefit from agglomeration effects. Their physical pres-
30
ence in large cities boosted the demand for local services such as food preparation and service,
security, cleaning, entertainment, and recreation (Moretti, 2012). This is the main reason for
the increase in the demand for the latter occupations relative to middle-paying ones that were
more amenable to automation and offshoring (Autor & Dorn, 2013). The ongoing pandemic
shock reversed this trend and if this reversal becomes permanent in the post-pandemic era,
it can have profound implications for the occupational composition of labour demand in the
medium- to long-run. This is definitely another topic which is worth shedding light upon.
Although anti-Covid vaccines have become available by now, there are large differences in
inoculation rates both across and within countries, while new variants of the virus have been
spreading around the globe.27 As the pandemic is still in full swing and the time horizon
of its end is unclear, we have to continue to closely monitor its economic and labour market
effects. Insights such as those derived from this analysis could be particularly useful to policy
makers in the US and other countries and prompt them to design and implement ambitious
medium- to long-term plans that could mitigate the adverse labour market effects of the
ongoing pandemic shock and prepare the workforce for the new state that will emerge after
the end of the current state of emergency.
27
Check, for instance: https://www.nytimes.com/interactive/2021/world/covid-vaccinations-tracker.html.
31
References
Acemoglu, D., & Autor, D. (2011). Skills, Tasks and Technologies: Implications for Employ-
ment and Earnings. In O. Ashenfelter & D. Card (Eds.), Handbook of Labor Economics
(Vol. 4, pp. 1043–1171). Elsevier.
Adams-Prassl, A., Boneva, T., Golin, M., & Rauh, C. (2020). Inequality in the Impact of the
Coronavirus Shock: Evidence from Real Time Surveys. Journal of Public Economics, 189 ,
104245.
Altig, D., Barrero, J. M., Bloom, N., Davis, S. J., Meyer, B., & Parker, N. (2020). Surveying
Business Uncertainty. Journal of Econometrics.
Autor, D. (2015). Why Are There Still So Many Jobs? The History and Future of Workplace
Automation. Journal of Economic Perspectives, 29 (3), 3–30.
Autor, D., & Dorn, D. (2013). The Growth of Low-Skill Service Jobs and the Polarization of
the US Labor Market. American Economic Review , 103 (5), 1553–1597.
Autor, D., Levy, F., & Murnane, R. (2003). The Skill Content of Recent Technological Change:
An Empirical Exploration. The Quarterly Journal of Economics, 118 (4), 1279–1333.
Autor, D., & Reynolds, E. (2020, July). The Nature of Work after the COVID Crisis: Too
Few Low-Wage Jobs (Essay). The Hamilton Project.
Baker, S., Bloom, N., & Davis, S. (2016). Measuring Economic Policy Uncertainty. The
Quarterly Journal of Economics, 131 (4), 1593–1636.
Baker, S., Bloom, N., Davis, S., Kost, K., Sammon, M., & Viratyosin, T. (2020). The
Unprecedented Stock Market Reaction to COVID-19. The Review of Asset Pricing Studies,
10 (4), 742–758.
Baker, S., Bloom, N., Davis, S., & Terry, S. (2020). COVID-Induced Economic Uncertainty
(Working Paper No. 26983). National Bureau of Economic Research.
Barrero, J. M., Bloom, N., & Davis, S. J. (2020). COVID-19 Is Also a Reallocation Shock
(Working Paper No. 27137). National Bureau of Economic Research.
Barrero, J. M., Bloom, N., & Davis, S. J. (2021). Why Working from Home Will Stick
(Working Paper No. 28731). National Bureau of Economic Research.
32
Bartik, A. W., Bertrand, M., Cullen, Z. B., Glaeser, E. L., Luca, M., & Stanton, C. T.
(2020). How Are Small Businesses Adjusting to COVID-19? Early Evidence from a Survey
(Working Paper No. 26989). National Bureau of Economic Research.
Baylis, P., Beauregard, P.-L., Connolly, M., Fortin, N. M., Green, D. A., Gutiérrez-Cubillos,
P., . . . Warman, C. (2021). The Distribution of COVID-19–related Risks. Canadian Journal
of Economics.
Bernstein, S., Townsend, R. R., & Xu, T. (2020). Flight to Safety: How Economic Downturns
Affect Talent Flows to Startups (Working Paper No. 27907). National Bureau of Economic
Research.
Bick, A., & Blandin, A. (2021). Real-Time Labor Market Estimates During the 2020 Coron-
avirus Outbreak (SSRN).
Blanas, S. (2021). The Distinct Effects of Information Technologies and Communication
Technologies on Skill Demand (Tech. Rep.). IMT School for Advanced Studies Lucca.
Bloom, N., Davis, S. J., & Zhestkova, Y. (2021). COVID-19 Shifted Patent Applications
toward Technologies That Support Working from Home. AEA Papers and Proceedings,
111 , 263–266.
Borghans, L., Weel, B. T., & Weinberg, B. A. (2014). People Skills and the Labor-Market
Outcomes of Underrepresented Groups. ILR Review , 67 (2), 287–334.
Cajner, T., Crane, L. D., Decker, R. A., Grigsby, J., Hamins-Puertolas, A., Hurst, E., . . .
Yildirmaz, A. (2020). The U.S. Labor Market during the Beginning of the Pandemic
Recession (Working Paper No. 27159). National Bureau of Economic Research.
Campello, M., Kankanhalli, G., & Muthukrishnan, P. (2020). Corporate Hiring under COVID-
19: Labor Market Concentration, Downskilling, and Income Inequality (Working Paper No.
27208). National Bureau of Economic Research.
Coibion, O., Gorodnichenko, Y., & Weber, M. (2020). Labor Markets During the COVID-19
Crisis: A Preliminary View (Working Paper No. 27017). National Bureau of Economic
Research.
33
Cortes, G. M., & Forsythe, E. C. (2020). The Heterogeneous Labor Market Impacts of the
Covid-19 Pandemic (Upjohn Institute Working Paper No. 20-327). Upjohn Institute for
Employment Research.
Dalton, M. R., Kahn, L. B., & Mueller, A. I. (2020). Do Online Job Postings Capture Job
Vacancies? An Analysis of Matched Online Postings and Vacancy Survey Data (Mimeo).
Deming, D. J. (2017, 06). The Growing Importance of Social Skills in the Labor Market. The
Quarterly Journal of Economics, 132 (4), 1593–1640.
Dingel, J. I., & Neiman, B. (2020). How Many Jobs Can Be Done at Home? Journal of
Public Economics, 189 , 104235.
Ferguson, N. M., Laydon, D., Nedjati-Gilani, G., Imai, N., Ainslie, K., Baguelin, M., . . .
Ghani, A. C. (2020). Impact of Non-Pharmaceutical Interventions to Reduce Covid-19
Mortality and Healthcare Demand (Tech. Rep.). Imperial College COVID-19 Response
Team.
Forsythe, E., Kahn, L. B., Lange, F., & Wiczer, D. (2020). Labor Demand in the Time of
COVID-19: Evidence from Vacancy Postings and UI Claims. Journal of Public Economics,
189 , 104–238.
Goos, M., Manning, A., & Salomons, A. (2014). Explaining Job Polarization: Routine-Biased
Technological Change and Offshoring. American Economic Review , 104 (8), 2509–2526.
Gupta, S., Simon, K., & Wing, C. (2020). Mandated and Voluntary Social Distancing during
the COVID-19 Epidemic. Brookings Papers on Economic Activity, 269–315.
Hassan, T. A., Hollander, S., van Lent, L., Schwedeler, M., & Tahoun, A. (2020). Firm-Level
Exposure to Epidemic Diseases: COVID-19, SARS, and H1N1 (Working Paper No. 26971).
National Bureau of Economic Research.
Hensvik, L., Le Barbanchon, T., & Rathelot, R. (2021). Job Search during the COVID-19
Crisis. Journal of Public Economics, 194 , 104349.
Hershbein, B., & Kahn, L. B. (2018, July). Do Recessions Accelerate Routine-Biased Tech-
nological Change? Evidence from Vacancy Postings. American Economic Review , 108 (7),
1737–72.
34
Jaimovich, N., & Siu, H. E. (2020, 03). Job Polarization and Jobless Recoveries. The Review
of Economics and Statistics, 102 (1), 129–147.
Kane, J. W., & Tomer, A. (2021). Valuing Human Infrastructure: Protecting and Investing in
Essential Workers during the COVID-19 Era. Public Works Management & Policy, 26 (1),
34–46.
Leduc, S., & Liu, Z. (2020, March). The Uncertainty Channel of the Coronavirus. FRBSF
Economic Letter , 2020 (07), 1–05.
Leibovici, F., Santacreu, A. M., & Famiglietti, M. (2020). Social Distancing and Contact-
Intensive Occupations (Tech. Rep.). FED St. Louis.
Marinescu, I., Skandalis, D., & Zhao, D. (2021). The Impact of the Federal Pandemic
Unemployment Compensation on Job Search and Vacancy Creation. Journal of Public
Economics, 200 , 104471.
Michaels, G., Natraj, A., & Reenen, J. V. (2014). Has ICT Polarized Skill Demand? Evidence
from Eleven Countries over Twenty-Five Years. The Review of Economics and Statistics,
96 (1), 60–77.
Mindell, D. A. (2015). Our Robots, Ourselves: Robotics and the Myths of Autonomy. New
York, NY: Viking.
Mongey, S., Pilossoph, L., & Weinberg, A. (2021). Which Workers Bear the Burden of Social
Distancing? The Journal of Economic Inequality, 19 (3), 509–526.
Montenovo, L., Jiang, X., Lozano Rojas, F., Schmutte, I. M., Simon, K. I., Weinberg, B. A.,
& Wing, C. (2020). Determinants of Disparities in Covid-19 Job Losses (Working Paper
No. 27132). National Bureau of Economic Research.
Moretti, E. (2012). The New Geography of Jobs. Houghton Mifflin Harcourt.
Ramani, A., & Bloom, N. (2021). The Donut Effect of Covid-19 on Cities (Working Paper
No. 28876). National Bureau of Economic Research.
Spitz-Oener, A. (2006). Technical Change, Job Tasks, and Rising Educational Demands:
Looking outside the Wage Structure. Journal of Labor Economics, 24 (2), 235–270.
Weinberger, C. J. (2014). The Increasing Complementarity between Cognitive and Social
Skills. The Review of Economics and Statistics, 96 (5), 849–861.
35
Appendix
A Data and variables
A.1 Variables for occupational characteristics
Occupational task content measures of Autor et al. (2003) By relying on the fourth
(1977) edition of the US Department of Labor’s Dictionary of Occupational Titles (DOT),
the authors create measures of the non-routine cognitive analytical, non-routine cognitive
interactive, non-routine manual, routine manual, and routine cognitive task content of 330
occupations in 1980.28 As occupations are identified by their occ1990dd codes, we first map
these codes to two-digit 2010 SOC codes according to the US IPUMS CPS (see Appendix
Section A.2), and then calculate the unweighted mean of the measures across occ1990dd
codes by the corresponding two-digit SOC code (Table B1).
Occupational task content measures of Autor & Dorn (2013) The authors create
measures of the abstract (i.e., non-routine cognitive), non-routine manual, and routine task
content of 330 occupations in 1980.29 Occupations are identified by their occ1990dd codes.
Therefore, we first assign two-digit 2010 SOC codes to occ1990dd codes as indicated by the
US IPUMS CPS (see Appendix Section A.2), and then calculate the unweighted mean of the
measures across occ1990dd codes by the corresponding two-digit SOC code (Table B4).
Occupational task content measures of Deming (2017) The author creates task content
measures for different O*NET 1998 occupations by considering tasks that pertain to the
DOT category of non-routine cognitive interactive tasks (social, coordination, interaction),
the degree of customer orientedness, and the degree of service orientedness (e.g. provision of
care).30 As all measures range between 1 (lowest) and 7 (highest), the author re-scales these
measures so that they range between 0 and 10. In order to produce scores at the two-digit
SOC level, we first map O*NET occupation codes to two-digit 2010 SOC codes. We do so
by using sequentially the crosswalk between O*NET codes and occ1990 codes, the crosswalk
between occ1990 codes and occ1990dd codes, and the crosswalk between occ1990dd codes
and two-digit SOC codes. The first two crosswalks are provided by Deming (2017), while
the third crosswalk is the one that we have created based on US IPUMS CPS and described
above.31 Then, we calculate the unweighted mean of the measures across occ1990dd codes by
the corresponding two-digit SOC code (Table B2).
28
Non-routine cognitive analytical tasks are those requiring quantitative reasoning, non-routine cognitive
interactive tasks are those requiring direction, control and planning, and non-routine manual tasks are those
requiring eye-hand-foot coordination. Routine manual tasks are those requiring finger dexterity, while routine
cognitive tasks are those requiring setting limits, tolerances or standards.
29
For the construction of these measures, they aggregate the respective original task scores of Autor et al.
(2003): non-routine cognitive analytical and non-routine cognitive interactive; non-routine manual; routine
manual and routine cognitive.
30
For the precise O*NET definition of each type of task, see Appendix Section A.3.
31
There is a more than 99% match between O*NET 1998 and occ1990 codes and a perfect match between
occ1990 and occ1990dd codes.
36
Essential occupations à la Mongey et al. (2021) We identify essential occupations using
the data of Mongey et al. (2021). After obtaining information on employment in two-digit
occupations from the Occupational Employment Statistics Survey (OES) 2018, the authors
calculate the employment share of each occupation in essential industries. To identify essential
industries, they rely on the relevant four-digit NAICS industry classification of Kane & Tomer
(2021). We label an occupation as essential if its employment share in essential industries is
at least equal to 60%. Under this methodology, the following occupations are labeled as
essential: community and social service; healthcare practice, healthcare support; protective
service; farming, fishing, and forestry; production; transportation and material moving. As
some retail occupations are also deemed as essential (e.g. salespersons in grocery stores), we
also label sales and related occupations as essential in the benchmark version of the dummy
variable. In the alternative version, we treat this occupation as non-essential (see columns (1)
and (2), respectively, of Table B3).
Physical contact intensity measures of Leibovici et al. (2020) The authors create a
physical contact intensity index for four-digit 2010 SOC occupations by combining individual-
level data from the 2017 American Community Survey with information from O*NET on
the extent to which an occupation requires a worker to perform job tasks in close physical
proximity to other people. A score is assigned to each possible answer, whose aggregation
yields the contact intensity index.32 Using the contact intensity scores for four-digit 2010
SOC occupations, we then calculate the unweighted mean across these by two-digit 2010 SOC
occupation.
Tele-work measures of Dingel & Neiman (2020) By relying on O*NET questions related
to the capacity of workers in different occupations to tele-work, the authors calculate the
fraction of tasks of 8-digit 2010 SOC occupations that can be performed from distance. As
a result, the occupation-specific tele-work measure ranges between zero and one, with zero
values implying that not a single task can be performed from distance and unit values implying
that all tasks can be performed from distance. Alternatively, the authors assign manually 0,
0.5 or 1 to each 6-digit 2010 SOC code. Calculating the unweighted mean of these measures
across 8-digit and 6-digit SOC codes, respectively, by the corresponding two-digit SOC code,
we obtain tele-work measures at the occupation level considered in our analysis. The scores
are displayed in columns (5) and (6) of Table B3.
A.2 Concordance between occ1990dd and two-digit SOC
The correspondence of occ1990dd codes to two-digit 2010 SOC occupation titles according to
the US IPUMS CPS is listed below:
Management (3–37); Architecture and Engineering (43–59); Computer and Mathematical
32
The possible answers and associated scores in parentheses are: “I don’t work near other people (beyond
100 ft.)” (0); “I work with others but not closely (e.g., private office)” (25); “Slightly close (e.g., shared office)”
(50); “Moderately close (at arm’s length)” (75); “Very close (near touching)” (100).
37
(64–68); Life, Physical, and Social Science (69–83, 166, 173); Healthcare Practitioners and
Technical (84–106, 203–208); Education, Training, and Library (113–165); Community and
Social Service (174–176); Legal (178, 179, 234); Arts, Design, Entertainment, Sports, and
Media (183–200); Sales and Related (243–253, 256–290); Business and Financial Operations
(253–256); Office and Administrative Support (303–391); Protective Service (415–427); Food
Preparation and Service (434–444); Healthcare Support (445–447); Building and Grounds
Cleaning and Maintenance (448–455); Personal Care and Service (456–469); Farming, Fishing,
and Forestry (473–498); Installation, Maintenance, and Repair (503–549); Construction and
Extraction (558–617, 865, 869); Production (628–699); Transportation and Material Moving
(803–859).
A.3 Definitions of different types of tasks in Deming (2017)
Social The measure of social task content of an occupation is the average of four variables: 1)
social perceptiveness (defined as “being aware of others’ reactions and understanding why they
react the way they do”); 2) coordination (“adjusting actions in relation to others’ actions); 3)
persuasion (“persuading others to approach things differently”); and 4) negotiation (“bringing
others together and trying to reconcile differences”).
Coordination The measure of coordination task content of an occupation is the average of
two variables: 1) coordinating work and activities of others (defined as “coordinating members
of a work group to accomplish tasks”); and 2) developing and building teams (“encouraging
and building mutual trust, respect, and cooperation among team members”).
Interaction The measure of interaction task content of an occupation is calculated as the
average of four variables: 1) interpreting the meaning of information to others (defined as
“translating or explaining what information means and how it can be understood or used to
support responses or feedback to others”); 2) communicating with other workers (“provid-
ing information to supervisors, fellow workers, and subordinates”); 3) communicating with
persons outside the organization (“representing the organization to customers, the public,
government, and other external sources”); and 4) establishing and maintaining relationships
(“developing constructive and cooperative working relationships with others”).
Customer-oriented The degree of customer-orientedness of an occupation corresponds to
tasks requiring knowledge of principles and processes for providing customer and personal
services, including needs assessment techniques, quality service standards, alternative delivery
systems, and customer satisfaction evaluation techniques.
Service-oriented The degree of service-orientedness of an occupation is the average of two
variables: 1) assisting and caring for others (defined as “providing assistance or personal care
to others”); and 2) service orientation (“actively looking for ways to help people”).
38
A.4 Economic policy uncertainty
A.4.1 News-based index
For the US, the news-based Economic Policy Uncertainty (EPU) index is a measure of the
monthly frequency of articles in ten leading national newspapers (e.g. New York Times,
Boston Globe) that contain at least one keyword from each of the following groups: “eco-
nomic” or “economy” (group 1); “uncertain” or “uncertainty” (group 2); “congress”, “deficit”,
“Federal Reserve”, “legislation”, “regulation”, “White House” (group 3). As the number of
articles may vary across newspapers, the monthly frequency is normalised by the total count
of articles in the corresponding newspaper and month. The data and methodology can be
found at: https://www.policyuncertainty.com/. According to the authors, the index aims at
capturing uncertainty about who will make economic policy decisions, what economic policy
actions will be undertaken and when, and the economic effects of policy actions (or inac-
tion) – including uncertainties related to the economic ramifications of “non-economic” policy
matters, e.g., military actions. Also, the index aims at capturing both near-term concerns
(e.g. when will the Fed adjust its policy rate) and longer-term concerns (e.g. how to fund
entitlement programs). For the construction of the EPU by policy area, the authors combine
the keywords in groups 1–3 with keywords that are relevant to each individual policy area.
The keywords by policy area are:
Healthcare: health care, Medicaid, Medicare, health insurance, malpractice Tort reform,
malpractice reform, prescription drugs, drug policy, food and drug administration, FDA,
medical malpractice, prescription drug act, medical insurance reform, medical liability, part
D, affordable care act, Obamacare.
Fiscal policy: Anything covered by “Taxes” or “Government Spending & Other”.
Taxes: taxes, tax, taxation, taxed.
Government spending and other: government spending, federal budget, budget battle,
balanced budget, defense spending, military spending, entitlement spending, fiscal stimulus,
budget deficit, federal debt, national debt, Gramm-Rudman, debt ceiling, fiscal footing, gov-
ernment deficits, balance the budget.
Monetary policy: Federal Reserve, the FED, money supply, open market operations, quan-
titative easing, monetary policy, fed funds rate, overnight lending rate, Bernanke, Volker,
Greenspan, central bank, interest rates, FED chairman, FED chair, lender of last resort, dis-
count window, European Central Bank, ECB, Bank of England, Bank of Japan, BOJ, Bank
of China, Bundesbank, Bank of France, Bank of Italy.
National Security national security, war, military conflict, terrorism, terror, 9/11, defense
spending, military spending, police action, armed forces, base closure, military procurement,
saber rattling, naval blockade, military embargo, no-fly zone, military invasion.
39
Entitlement Programs: entitlement program, entitlement spending, government entitle-
ments, social security, Medicaid, medicare, government welfare, welfare reform, unemployment
insurance, unemployment benefits, food stamps, AFDC, TANF, WIC program, disability in-
surance, part D, OASDI, Supplemental Nutrition Assistance Program, Earned Income Tax
Credit, EITC, head start program, public assistance, government subsidized housing.
Regulation: Anything covered by “Financial Regulation” and truth in lending, union rights,
card check, collective bargaining law, national labor relations board, NLRB, minimum wage,
living wage, right to work, closed shop, wages and hours, workers compensation, advance
notice requirement, affirmative action, at-will employment, overtime requirements, trade ad-
justment assistance, Davis-Bacon, equal employment opportunity, EEO, OSHA, antitrust,
competition policy, merger policy, monopoly, patent, copyright, federal trade commission,
FTC, unfair business practice, cartel, competition law, price fixing, class action, healthcare
lawsuit, Tort reform, tort policy, punitive damages, medical malpractice, energy policy, energy
tax, carbon tax, cap and trade, cap and tax, drilling restrictions, offshore drilling, pollution
controls, environmental restrictions, clean air act, clean water act, environmental protection
agency, EPA, immigration policy.
Financial Regulation: banking (or bank) supervision, Glass-Steagall, TARP, thrift super-
vision, Dodd-Frank, financial reform, commodity futures trading commission, CFTC, house
financial services committee, basel, capital requirement, Volcker rule, bank stress test, secu-
rities and exchange commission, sec, deposit insurance, FDIC, FSLIC, OTS, OCC, FIRREA.
Trade Policy: import tariffs, import duty, import barrier, government subsidies, government
subsidy, WTO, World Trade Organization, trade treaty, trade agreement, trade policy, trade
act, Doha Round, Uruguay Round, GATT, dumping.
Sovereign Debt and Currency Crises: sovereign debt, currency crisis, currency crash, cur-
rency devaluation, currency revaluation, currency manipulation, euro crisis, Eurozone crisis,
european financial crisis, european debt, asian financial crisis, asian crisis, Russian financial
crisis, Russian crisis, exchange rate.
A.4.2 Composite index
The composite Economic Policy Uncertainty (EPU) index consists of three components. The
first component is the news-based EPU. The second relies on tax code expiration data retrieved
from reports of the Congressional Budget Office (CBO) that compile lists of temporary federal
tax code provisions. Temporary tax measures can capture the uncertainty facing firms and
households as they can be extended with very short or no prior notice. The third component
relies on data on disagreements about economic forecasts that are retrieved from the Federal
Reserve Bank of Philadelphia’s Survey of Professional Forecasters. The level of disagreement
is measured empirically as the dispersion in the quarterly forecast of individuals for one year
in the future about three variables that are directly influenced by government policy: CPI,
40
purchases of goods and services by state and local governments, and purchases of goods and
services by the federal government.
For the construction of the composite (three-component) EPU measure, Baker et al. (2016)
take the following steps. First, they normalise each component by its own standard deviation
prior to January 2012. Then, they calculate the weighted average of the three components
using weights of 1/2 on the first component (i.e., news-based EPU), 1/6 on the second com-
ponent (i.e., based on temporary tax measures), and 1/6 on the two sub-components of the
third component (i.e., based on disagreement about the CPI forecast and disagreement about
the federal/state/local purchases forecast).
41
B
Ja online job postings per capita (%) Ja online job postings per capita (%)
n .3 .4 .5 .6 .7 n .2 .25 .3 .35 .4
20 20
20 20
Fe Fe
b b
20 20
20 20
M M
ar ar
20 20
20 20
A A
pr pr
20 2 02
20 0
M M
ay ay
20 20
20 20
Ju Ju
n n
20 20
20 20
Ju Ju
l2 l 20
02
online job postings per capita (%) 0 20
Ja A A
n .4 .6 .8 1 1.2 ug ug
20 20 20
2
(a) Australia
0 20 20
Fe
(c) New Zealand
b Se
p
Se
p
20 20 20
2 0 20 20
M O O
ar ct ct
20
20 2 02 20
A 0 20
pr N N
ov ov
20 20 20
20 20 20
M D D
ay ec ec
20 20 20
20 20 20
Ju
n
20
20
42 Ju
l2
02
0 online job postings per capita (%) online job postings per capita (%)
A Ja Ja
ug n .8 1 1.2 1.4 1.6 1.8 n .15 .2 .25 .3 .35
20 20 20
20 20 20
Additional descriptive statistics
Fe Fe
Se
p b b
20 20 20
(e) United Kingdom
2 0
20 20
O M M
ct ar ar
20 20 20
2 0
20 20
N A A
ov pr pr
20 20 2 02
20 20 0
D M M
ec ay ay
20 20 20
20 20 20
Ju Ju
n n
20 20
20 20
Ju Ju
l2 l 20
02
0 20
A A
ug ug
(b) Canada
20 20
(d) Singapore
20 20
Se Se
p p
20 20
Figure B1: Online job postings per capita in other countries
20 20
O O
ct ct
2 02 20
0 20
N N
ov ov
20 20
20 20
D D
ec ec
20 20
20 20
Source: Authors’ calculations based on online job postings data of Burning Glass Technologies (BGT).
Singapore and the United Kingdom to the respective nationwide population in January 2020 – December 2020. Notes: Monthly variation in ratios of number of online job postings in Australia, Canada, New Zealand,
Ja US EPU Ja US EPU Ja US EPU
n 0 500 1000 n 50 100 150 200 250 n 200 400 600 800 1000
20 20 20
20 20 2 0
Fe Fe Fe
b b b
20 20 20
20 20 20
M M M
ar ar ar
20 20 2 02
20 20 0
A A A
pr pr pr
20 20 20
20 20 20
M M M
ay ay ay
20 20 20
20 20 20
Ju Ju Ju
n n n
20 20 20
20 20 20
Ju Ju Ju
l 20 l 20 l2
20 20 02
0
US EPU A A A
Ja ug ug ug
n 0 200 400 600 20 20 20
20 20 20 20
20
Se Se Se
Fe
b p p p
20 20 20 20
2 20 20 20
0 O O O
M ct ct ct
ar 20 20 20
20 20 20 20
2
(a) Health policy
0 N N N
A ov ov ov
pr 20 20 20
20 20 20 20
20
M D D D
(d) Government policy
ay ec ec ec
20 20 20 20
20 20 20 20
Ju
n
(g) Entitlement programmes
20
2 0
Ju
l2
02
0
A US EPU US EPU US EPU
ug Ja Ja Ja
20 n 100 150 200 250 300 350 n 100 150 200 250 300 n 100 200 300 400 500
2 0
20
20
20
20
20
2
Se 0
p Fe
b
Fe
b
Fe
b
20
2 0
20
20
20
20
20
20
O M M M
ct ar ar ar
(j) Trade policy
20 20 20 2
20 20 20 02
N 0
ov A A A
20 pr pr pr
2 0
20
20
20
20
20
20
D M M M
ec ay ay ay
2 02 20 20 20
0 20 20 20
Ju Ju Ju
n n n
20 20 20
20 20 20
Ju Ju Ju
43 l 20 l 20 l2
20 20 02
0
US EPU A A A
Ja ug ug ug
n 0 50 100 150 20 20 20
20 20 20 20
20
Se Se Se
Fe
b p p p
20 20 20 20
2 20 20 20
0 O O O
M
(h) Regulation
ar ct ct ct
20 20 20
(b) Fiscal policy
20 20 20 20
2 0 N N N
A ov ov ov
pr
(e) Monetary policy
20 20 20 20
20 20 20 20
M D D D
ay ec ec ec
20 20 20 20
20 20 20 20
Ju
n
20
2 0
Ju
l2
02
0
A US EPU US EPU US EPU
ug Ja Ja Ja
20 n 0 100 200 300 400 n 50 100 150 200 250 300 n 100 200 300 400 500
2 0
20
20
20
20
20
2
Se 0
p Fe
b
Fe
b
Fe
b
Figure B2: US EPU by policy area
20
2 0
20
20
20
20
20
20
O M M M
ct ar ar ar
20 20 20 2
20 20 20 02
N 0
ov A A A
20 pr pr pr
2 0
20
20
20
20
20
20
D M M M
ec ay ay ay
2
Source: https://www.policyuncertainty.com/; see particularly Baker et al. (2016).
02 20 20 20
0 20 20 20
Ju Ju Ju
n n n
(k) Debt and currency crisis
20 20 20
20 20 20
Ju Ju Ju
l 20 l 20 l2
20 20 02
0
A A A
ug ug ug
20 20 20
20 20 20
Se Se Se
p p p
20 20 20
20 20 20
O O O
(c) Tax policy
ct ct ct
20 20 20
20 20 20
N N N
(f) Security policy
ov ov ov
20 20 20
20 20 20
D D D
ec ec ec
(i) Financial regulation
20 20 20
20 20 20
Notes: Monthly variation in US (news-based) EPU measure by policy area in January 2020 – December 2020.
Ja Global news−based EPU
Fen 20 200 250 300 350 400 450
M b 2 19
a 0
A r 2019
M pr 2 19
ay 01
Ju 20 9
n 1
Ju 20 9
A l 2 19
ug 01
Se 20 9
p
O 2019
N ct 2 19
ov 01
D 20 9
ec 1
Ja 20 9
Fen 2019
M b 2 20
a 0
A r 2020
M pr 2 20
ay 02
Ju 20 0
n 2
(a) Global
Ju 20 0
A l 2 20
ug 02
Se 20 0
p
O 2020
N ct 2 20
o 0
D v 2020
ec 2
Ja 20 0
Fen 2020
M b 2 21
a 0
A r 2021
M pr 2 21
ay 02
20 1
21
44
Ja Mean news−based EPU across countries (excl. US)
Fen 20 150 200 250 300 350
M b 2 19
ar 01
A 20 9
M pr 2 19
ay 01
Ju 20 9
n 1
Ju 20 9
A l 2 19
ug 01
Se 20 9
p
O 2019
N ct 2 19
o 0
D v 2019
ec 1
Ja 20 9
Fen 2019
M b 2 20
a 0
Source: https://www.policyuncertainty.com/; see particularly Baker et al. (2016).
A r 2020
M pr 2 20
ay 02
Ju 20 0
n 2
Ju 20 0
A l 2 20
u 0
Seg 2020
p
O 2020
Figure B3: Global EPU and EPU in countries other than the US
N ct 2 20
ov 02
Ireland, Italy, Japan, Korea, Netherlands, Russia, Spain, Singapore, UK, Sweden, and Mexico.
D 20 0
ec 2
Ja 20 0
Fen 2020
(b) Cross-country mean, excl. US
M b 2 21
a 0
A r 2021
M pr 2 21
ay 02
20 1
21
of countries comprises: Australia, Brazil, Canada, Chile, China, Colombia, France, Germany, Greece, India, based) EPU measures for a group of countries other than the US in January 2019 – May 2021. The group Notes: Monthly variation in the (news-based) Global EPU measure and the unweighted average of (news-
Ja US sales revenue growth expectations
Fen 20 −2 0 2 4 6
M b 2 19
a 0
A r 2 19
M pr 2019
ay 01
Ju 20 9
n 1
J 20 9
A ul 2 19
u 0
Seg 2019
Altig et al. (2020).
O p 2019
N ct 2 19
o 0
D v 2 19
ec 0
Ja 2019
Fen 2019
M b 2 20
a 0
January 2019 – June 2021.
A r 2 20
M pr 2020
ay 02
Ju 20 0
n 1
J 20 9
A ul 2 20
ug 02
Se 20 0
O p 2020
N ct 2 20
o 0
D v 2 20
ec 0
Ja 2020
Fen 2020
M b 2 21
a 0
A r 2 21
(a) Sales revenue growth expectations
M pr 2021
ay 02
Ju 20 1
n 11
20
21
45
Ja US employment growth expectations
Fen 20 0 1 2 3 4
M b 2 19
a 0
A r 2 19
M pr 2019
ay 01
Ju 20 9
n 1
J 20 9
A ul 2 19
u 0
Seg 2019
O p 2019
N ct 2 19
o 0
D v 2 19
ec 0
Ja 2019
Fen 2019
M b 2 20
Figure B4: Business expectations in the US
a 0
A r 2 20
M pr 2020
ay 02
Ju 20 0
n 1
J 20 9
A ul 2 20
u 0
Seg 2020
O p 2020
N ct 2 20
o 0
D v 2 20
ec 0
Ja 2020
Fen 2020
M b 2 21
a 0
A r 2 21
(b) Employment growth expectations
M pr 2021
ay 02
Ju 20 1
n 11
20
21
Source: Atlanta FED, University of Chicago Booth School of Business, Stanford University; see particularly Notes: Monthly variation in the sales revenue and employment growth expectation measures in the US in
Table B1: Autor et al. (2003) task content scores by (two-digit SOC) occupation
Two-digit (SOC) occupation code & title Non-routine Routine
cognitive analytical cognitive interactive manual manual cognitive
11 Management 5.40 6.71 0.49 2.78 2.24
13 Business and Financial Operations 4.76 2.73 0.02 3.06 1.05
15 Computer and Mathematical 7.90 4.72 0.71 2.58 4.90
17 Architecture and Engineering 8.49 6.75 1.02 4.68 7.76
19 Life, Physical, and Social Science 7.94 4.80 0.71 3.88 5.29
21 Community and Social Service 5.61 4.30 0.16 2.54 0.23
23 Legal 4.47 3.18 0.06 2.89 3.65
25 Education, Training, and Library 4.78 5.28 1.33 3.49 0.56
27 Arts, Design, Entertainment, Sports, and Media 3.83 2.82 1.92 4.43 3.28
29 Healthcare Practitioners and Technical 5.92 2.97 1.09 5.60 4.37
31 Healthcare Support 3.69 0.38 0.79 4.77 6.26
33 Protective Service 1.66 1.24 2.30 2.48 0.87
35 Food Preparation and Service 1.73 0.37 1.10 2.84 1.51
37 Building and Grounds Cleaning and Maintenance 1.81 1.27 2.21 2.97 4.22
39 Personal Care and Service 2.20 1.34 0.93 3.72 1.65
41 Sales and Related 3.91 2.18 0.67 3.74 2.18
43 Office and Administrative Support 2.87 0.79 0.20 4.07 5.82
45 Farming, Fishing, and Forestry 2.44 2.98 2.44 3.17 2.88
47 Construction and Extraction 2.64 0.36 3.15 4.29 8.41
49 Installation, Maintenance, and Repair 3.78 0.43 1.63 4.91 8.82
51 Production 3.53 0.79 0.99 4.54 8.76
53 Transportation and Material Moving 1.62 1.47 3.70 2.82 3.10
Notes: For the calculation of the scores by two-digit 2010 SOC occupation, we rely on Autor et al. (2003)’s 1980 measures of the non-routine cognitive analytical,
non-routine cognitive interactive, non-routine manual, routine manual, and routine cognitive task content of 330 occupations (identified by their occ1990dd codes). First,
we map occ1990dd codes to two-digit 2010 SOC codes according to the US IPUMS CPS (see Appendix Section A.2) and then, we calculate the simple average of the
measures across occ1990dd codes by the corresponding two-digit SOC code.
Source: Authors’ calculations based on data of Autor et al. (2003).
46
Table B2: Deming (2017) task content scores by (two-digit SOC) occupation
Two-digit (SOC) occupation code & title Social Coordination Interaction Customer- Service-
oriented oriented
11 Management 6.18 4.81 7.12 6.02 3.55
13 Business and Financial Operations 6.43 2.15 6.43 6.94 3.65
15 Computer and Mathematical 3.79 1.81 6.20 2.08 1.86
17 Architecture and Engineering 5.32 6.30 7.81 3.56 2.44
19 Life, Physical, and Social Science 4.37 3.20 7.37 3.35 2.86
21 Community and Social Service 6.71 4.72 7.59 7.09 8.66
23 Legal 5.22 2.54 6.56 5.20 3.06
25 Education, Training, and Library 4.93 4.09 6.95 5.12 6.27
27 Arts, Design, Entertainment, Sports, and Media 3.51 2.79 5.10 3.77 1.89
29 Healthcare Practitioners and Technical 4.52 2.94 6.34 6.74 6.97
31 Healthcare Support 3.26 1.26 4.80 5.49 6.79
33 Protective Service 4.37 2.15 5.08 7.15 5.61
35 Food Preparation and Service 1.55 1.35 3.75 7.17 4.21
37 Building and Grounds Cleaning and Maintenance 3.05 2.67 3.65 4.68 2.97
39 Personal Care and Service 2.42 1.56 4.20 8.38 5.39
41 Sales and Related 4.45 2.58 6.03 8.71 4.19
43 Office and Administrative Support 2.20 1.43 4.16 5.40 3.38
45 Farming, Fishing, and Forestry 1.94 1.86 3.32 2.47 1.44
47 Construction and Extraction 1.15 1.03 1.72 1.62 1.02
49 Installation, Maintenance, and Repair 1.40 1.18 2.58 2.94 1.71
51 Production 1.11 1.14 1.80 1.24 0.99
53 Transportation and Material Moving 2.07 1.77 3.16 4.00 2.20
Notes: For the calculation of the scores by two-digit 2010 SOC occupation, we rely on Deming (2017)’s task content measures. The author creates
such measures for different O*NET 1998 occupations by considering tasks that pertain to the DOT category of non-routine cognitive interactive tasks
(social, coordination, interaction, customer-oriented), and the DOT category of non-routine manual tasks (service-oriented). The precise O*NET
definition of each type of task are shown in Appendix Section A.3. The author re-scales all measures so that they range between 0 and 10. In order
to produce scores at the two-digit SOC level, we first map O*NET occupation codes to two-digit 2010 SOC codes. We do so by using sequentially the
crosswalk between O*NET codes and occ1990 codes, the crosswalk between occ1990 codes and occ1990dd codes, and the crosswalk between occ1990dd
codes and two-digit SOC codes. We then calculate simple averages of the measures across occ1990dd codes by the corresponding two-digit SOC code.
Source: Authors’ calculations based on data of Deming (2017).
47
Table B3: Additional occupational characteristics by (two-digit SOC) occupation
Two-digit (SOC) occupation code & title Essential Contact intensity Work-from-home
Main Alternative Main Alternative O*NET Manual
11 Management 0 0 47 47 0.66 0.66
13 Business and Financial Operations 0 0 50 49 0.80 0.86
15 Computer and Mathematical 0 0 43 43 1.00 1.00
17 Architecture and Engineering 0 0 50 49 0.44 0.71
19 Life, Physical, and Social Science 0 0 48 48 0.62 0.42
21 Community and Social Service 1 1 58 58 0.50 0.58
23 Legal 0 0 49 49 0.88 0.88
25 Education, Training, and Library 0 0 65 67 0.93 0.76
27 Arts, Design, Entertainment, Sports, and Media 0 0 58 56 0.63 0.33
29 Healthcare Practitioners and Technical 1 1 82 84 0.09 0.06
31 Healthcare Support 1 1 87 90 0.11 0.00
33 Protective Service 1 1 71 70 0.14 0.00
35 Food Preparation and Service 0 0 75 75 0.06 0.00
37 Building and Grounds Cleaning and Maintenance 0 0 54 55 0.00 0.00
39 Personal Care and Service 0 0 74 73 0.25 0.00
41 Sales and Related 1 0 58 55 0.54 0.43
43 Office and Administrative Support 0 0 56 57 0.60 0.52
45 Farming, Fishing, and Forestry 1 1 43 45 0.06 0.00
47 Construction and Extraction 0 0 69 69 0.02 0.00
49 Installation, Maintenance, and Repair 0 0 63 64 0.02 0.00
51 Production 1 1 57 56 0.05 0.00
53 Transportation and Material Moving 1 1 64 62 0.06 0.00
Notes: For the identification of (non-)essential two-digit 2010 SOC occupations in columns (1) and (2), we rely on the data of Mongey et al. (2021). We
classify an occupation as (non-)essential if its employment share in essential industries is at least (less than) 60%. For the calculation of the measures of
physical contact intensity by two-digit 2010 SOC occupation in columns (3) and (4), we rely on the data of Leibovici et al. (2020). We calculate the simple
average of the physical contact intensity measures across four-digit 2010 SOC occupations by the corresponding two-digit SOC code. For the calculation
of the scores by two-digit 2010 SOC occupation in columns (5) and (6), we rely on Dingel & Neiman (2020)’s work-from-home measures for 8-digit 2010
SOC and 6-digit 2010 SOC occupations, respectively. Using O*NET information on tele-work, the authors calculate the first measure as the fraction of
tasks of occupations that can be performed from distance. Hence, value 0 indicates that zero fraction of tasks of an occupation can be performed from
distance, while value 1 indicates that all tasks of an occupation can be performed from distance. Alternatively, the authors assign manually value 0, 0.5
or 1 to each 6-digit 2010 SOC code. We produce the scores of columns (5) and (6) by computing simple averages of these measures across 8-digit and
6-digit SOC codes, respectively, by the corresponding two-digit SOC code.
Source: Authors’ assignments of unit or zero values based on the data and methodology of Mongey et al. (2021) in columns (1) and (2). Authors’
calculations based on Leibovici et al. (2020) in columns (3) and (4) and on Dingel & Neiman (2020) in columns (5) and (6).
48
Table B4: Autor & Dorn (2013) task content scores by (two-digit SOC) occupation
Two-digit (SOC) occupation code & title Abstract Non-routine manual Routine
11 Management 6.09 0.49 2.54
13 Business and Financial Operations 3.75 0.02 2.05
15 Computer and Mathematical 6.01 0.68 3.57
17 Architecture and Engineering 7.72 1.04 6.13
19 Life, Physical, and Social Science 6.23 0.75 4.57
21 Community and Social Service 5.05 0.09 1.36
23 Legal 3.15 0.08 2.55
25 Education, Training, and Library 4.92 1.34 2.04
27 Arts, Design, Entertainment, Sports, and Media 3.23 2.00 3.87
29 Healthcare Practitioners and Technical 4.44 1.09 5.01
31 Healthcare Support 1.76 1.09 5.26
33 Protective Service 1.41 2.40 1.58
35 Food Preparation and Service 1.04 1.12 2.41
37 Building and Grounds Cleaning and Maintenance 1.32 2.26 3.48
39 Personal Care and Service 1.71 0.91 2.67
41 Sales and Related 3.15 0.56 2.91
43 Office and Administrative Support 1.80 0.22 4.87
45 Farming, Fishing, and Forestry 2.85 2.42 2.64
47 Construction and Extraction 1.60 3.04 6.17
49 Installation, Maintenance, and Repair 2.12 1.61 6.92
51 Production 2.07 0.99 6.67
53 Transportation and Material Moving 1.97 3.80 3.16
Notes: For the calculation of the scores by two-digit 2010 SOC occupation, we rely on Autor & Dorn (2013)’s 1980 measures of
the abstract, non-routine manual, and routine task content of 330 occupations (identified by their occ1990dd codes). The authors
calculate the scores of the abstract, non-routine manual, and routine task content of occupations by aggregating the original
five task scores of Autor et al. (2003) who rely on the 1997 DOT: non-routine cognitive analytical and non-routine cognitive
interactive; non-routine manual; routine manual and routine cognitive. After mapping occ1990dd codes to two-digit 2010 SOC
codes according to the US IPUMS CPS (see Appendix Section A.2), we calculate the simple average of the measures across
occ1990dd codes by the corresponding two-digit SOC code.
Source: Authors’ calculations based on data of Autor & Dorn (2013).
49
C Additional econometric results
Figure C1: Covid-induced economic uncertainty by area and occupational demand
.002 .0005
0
0
−.002 −.0015 −.001 −.0005
−.002
−.004
−.006
EPU (main) EPU (alt) EPU (main) EPU (alt)
EPU (health) EPU (fiscal) EPU (health) EPU (fiscal)
EPU (tax) EPU (gov’t) EPU (tax) EPU (gov’t)
EPU (monetary) EPU (security) EPU (monetary) EPU (security)
EPU (entitl’t prog.) EPU (regulation) EPU (entitl’t prog.) EPU (regulation)
EPU (fin. regulation) EPU (trade policy) EPU (fin. regulation) EPU (trade policy)
EPU (debt/currency) EPU (debt/currency)
(a) Job postings per capita, EPU in 2020 (b) Share of job postings, EPU in 2020
.008 .003
.006 .002
.004 .001
.002 0
0 −.001
EPU 2019 (main) EPU 2019 (alt) EPU 2019 (main) EPU 2019 (alt)
EPU 2019 (health) EPU 2019 (fiscal) EPU 2019 (health) EPU 2019 (fiscal)
EPU 2019 (tax) EPU 2019 (gov’t) EPU 2019 (tax) EPU 2019 (gov’t)
EPU 2019 (monetary) EPU 2019 (security) EPU 2019 (monetary) EPU 2019 (security)
EPU 2019 (entitl’t prog.) EPU 2019 (regulation) EPU 2019 (entitl’t prog.) EPU 2019 (regulation)
EPU 2019 (fin. regulation) EPU 2019 (trade policy) EPU 2019 (fin. regulation) EPU 2019 (trade policy)
EPU 2019 (debt/currency) EPU 2019 (debt/currency)
(c) Job postings per capita, EPU in 2019 (d) Share of job postings, EPU in 2019
Notes: OLS estimations with robust standard errors and two-digit (SOC) occupation-US state and month fixed
effects.
50
Table C1: Alternative measures of Covid-induced economic uncertainty
Panel A: Mean and median EPU across countries, excl. US
(1) (2) (3) (4) (5) (6)
Dep. var.: log # of OJP OJP p.c. OJP share log # of OJP OJP p.c. OJP share
Mean of EP Uost -0.025** -0.0088*** -0.0019***
[0.010] [0.0008] [0.0006]
Median of EP Uost -0.034*** -0.0095*** -0.0023***
[0.009] [0.0007] [0.0006]
Observations 13,464 13,464 13,464 13,464 13,464 13,464
R2 0.987 0.929 0.953 0.987 0.930 0.953
Panel B: US sales revenue and employment growth expectations
(1) (2) (3) (4) (5) (6)
Dep. var.: log # of OJP OJP p.c. OJP share log # of OJP OJP p.c. OJP share
SBUost (sales gr. exp.) 0.0056** 0.0017*** 0.00074***
[0.002] [0.0002] [0.0001]
SBUost (emp. gr. exp.) 0.011*** 0.0014*** 0.00064***
[0.003] [0.0002] [0.0002]
Observations 13,464 13,464 13,464 13,464 13,464 13,464
R2 0.987 0.929 0.953 0.987 0.929 0.953
Panel C: Global EPU in 2019
(1) (2) (3) (4) (5) (6)
Dep. var.: log # of OJP OJP p.c. OJP share log # of OJP OJP p.c. OJP share
Global EP Uost 2019 0.97*** 0.10*** 0.059***
[0.2] [0.01] [0.01]
Global EP Uost PPP 2019 0.86*** 0.081*** 0.056***
[0.2] [0.01] [0.01]
Observations 13,464 13,464 13,464 13,464 13,464 13,464
R2 0.987 0.929 0.953 0.987 0.929 0.953
Panel D: Mean and median EPU across countries, excl. US, in 2019
(1) (2) (3) (4) (5) (6)
Dep. var.: log # of OJP OJP p.c. OJP share log # of OJP OJP p.c. OJP share
Mean of EP Uost 2019 0.94*** 0.10*** 0.086***
[0.2] [0.02] [0.01]
Median of EP Uost 2019 0.23 0.032** 0.070***
[0.2] [0.02] [0.01]
Observations 13,464 13,464 13,464 13,464 13,464 13,464
R2 0.987 0.929 0.954 0.987 0.929 0.953
Notes: OLS estimations with robust standard errors in all columns of all panels. Two-digit (SOC) occupation-US state and month fixed effects
are included in the equations. Asterisks denote significance at 1% (***), 5% (**), and 10% (*).
51
Table C2: Alternative measures of occupational characteristics
Panel A: Cognitive analytical/interactive, non-routine manual, routine cognitive/manual
(1) (2) (3) (4) (5)
Dep. var.: OJP share OJP share OJP share OJP share OJP share
EP Uost -0.0022* -0.0013*** -0.0021*** -0.00083 -0.00020
[0.001] [0.0004] [0.0004] [0.002] [0.0005]
EP Ut * cogn. analytical (alt) -0.00026
[0.0004]
EP Uost * cogn. analytical (alt) 0.0012
[0.001]
EP Ut * cogn. interactive (alt) 0.00024
[0.0003]
EP Uost * cogn. interactive (alt) 0.00032
[0.0003]
EP Ut * non-routine manual (alt) 0.0022***
[0.0005]
EP Uost * non-routine manual (alt) 0.0017**
[0.0007]
EP Ut * routine manual (alt) 0.00041
[0.0005]
EP Uost * routine manual (alt) -0.00035
[0.002]
EP Ut * routine cognitive (alt) 0.0010***
[0.0003]
EP Uost * routine cognitive (alt) -0.0014***
[0.0004]
Observations 13,464 13,464 13,464 13,464 13,464
R2 0.953 0.953 0.954 0.953 0.953
Panel B: Customer/Service-orientedness, physical contact intensity, tele-work, essential
(1) (2) (3) (4) (5)
Dep. var.: OJP share OJP share OJP share OJP share OJP share
EP Uost 0.0014 -0.0021** -0.0056*** 0.000054 -0.0021***
[0.0010] [0.0010] [0.002] [0.0006] [0.0003]
EP Ut * customer (alt) -0.0020***
[0.0004]
EP Uost * customer (alt) -0.0025***
[0.0009]
EP Ut * service (alt) -0.00064
[0.0004]
EP Uost * service (alt) 0.0011
[0.0010]
EP Ut * contact (alt) 0.00048
[0.0006]
EP Uost * contact (alt) 0.0045***
[0.002]
EP Ut * tele-work (alt) -0.00036
[0.0004]
EP Uost * tele-work (alt) -0.0017***
[0.0006]
EP Ut * essential (alt) 0.00013
[0.0002]
EP Uost * essential (alt) 0.052***
[0.01]
Observations 13,464 13,464 13,464 13,464 13,464
R2 0.954 0.953 0.953 0.953 0.954
Notes: OLS estimations with robust standard errors in all columns of both panels. Two-digit (SOC) occupation-US state and month
fixed effects are included in the equations. Asterisks denote significance at 1% (***), 5% (**), and 10% (*).
52
Table C3: Accounting for Autor & Dorn (2013)’s abstract, manual, routine task content
(1) (2) (3) (4) (5) (6)
Dep. var.: OJP share OJP share OJP share OJP share OJP share OJP share
EP Uost -0.0023*** -0.0022*** -0.00079 -0.0018*** -0.0020*** -0.00017
[0.0006] [0.0004] [0.0007] [0.0006] [0.0004] [0.0006]
EP Ut * abstract -0.00013
[0.0003]
EP Uost * abstract 0.0014**
[0.0005]
EP Ut * non-routine 0.0021***
[0.0004]
EP Uost * non-routine 0.0019**
[0.0008]
EP Ut * routine 0.00084***
[0.0003]
EP Uost * routine -0.00042
[0.0007]
EP Ut * abstract (alt) -0.000022
[0.0003]
EP Uost * abstract (alt) 0.00079
[0.0005]
EP Ut * non-routine (alt) 0.0022***
[0.0005]
EP Uost * non-routine (alt) 0.0016**
[0.0008]
EP Ut * routine (alt) 0.0010***
[0.0003]
EP Uost * routine (alt) -0.0012**
[0.0006]
Observations 13,464 13,464 13,464 13,464 13,464 13,464
R2 0.953 0.954 0.953 0.953 0.954 0.953
Notes: OLS estimations with robust standard errors in all columns. Two-digit (SOC) occupation-US state and month fixed effects are
included in the equations. Asterisks denote significance at 1% (***), 5% (**), and 10% (*).
53
Table C4: Controlling for US state-month fixed effects
Panel A: Cognitive analytical/interactive, non-routine manual, routine cognitive/manual
(1) (2) (3) (4) (5)
Dep. var.: OJP share OJP share OJP share OJP share OJP share
EP Uost -0.0032*** -0.0029*** -0.0029*** -0.0018 -0.0016***
[0.001] [0.0005] [0.0004] [0.002] [0.0005]
EP Ut * cogn. analytical -0.00032
[0.0004]
EP Uost * cogn. analytical 0.0012
[0.001]
EP Ut * cogn. interactive -0.00016
[0.0003]
EP Uost * cogn. interactive 0.00098***
[0.0004]
EP Ut * non-routine manual 0.0021***
[0.0004]
EP Uost * non-routine manual 0.0017**
[0.0008]
EP Ut * routine manual 0.00030
[0.0006]
EP Uost * routine manual -0.00037
[0.002]
EP Ut * routine cognitive 0.00097***
[0.0003]
EP Uost * routine cognitive -0.00073*
[0.0004]
N 13,464 13,464 13,464 13,464 13,464
R2 0.953 0.953 0.954 0.953 0.953
Panel B: Customer/Service-orientedness, physical contact intensity, tele-work, essential
(1) (2) (3) (4) (5)
Dep. var.: OJP share OJP share OJP share OJP share OJP share
EP Uost 0.000031 -0.0024** -0.0058*** -0.00073 -0.0030***
[0.001] [0.001] [0.002] [0.0008] [0.0004]
EP Ut * customer -0.0021***
[0.0004]
EP Uost * customer -0.0018*
[0.001]
EP Ut * service -0.00034
[0.0004]
EP Uost * service 0.00026
[0.001]
EP Ut * contact 0.00071
[0.0006]
EP Uost * contact 0.0037**
[0.002]
EP Ut * WFH -0.00039
[0.0004]
EP Uost * WFH -0.0017**
[0.0007]
EP Ut * essential -0.00024
[0.0004]
EP Uost * essential 0.0018***
[0.0005]
N 13,464 13,464 13,464 13,464 13,464
R2 0.954 0.953 0.953 0.953 0.954
Notes: OLS estimations with robust standard errors in all columns of both panels. Two-digit (SOC) occupation-US state and US
state-month fixed effects are included in the equations. Asterisks denote significance at 1% (***), 5% (**), and 10% (*).
54
INSTITUT DE RECHERCHE
ÉCONOMIQUES ET SOCIALES
Place Montesquieu 3
1348 Louvain-la-Neuve
ISSN 1379-244X D/2022/3082/02
File and source
- File
- covid-induced-economic-uncertainty-tasks-and-occupational-demand-8554a244.pdf
- Size
- 1,281,851 bytes
- SHA-256
- 8554a244e14a65d4fedd15c9d543d21ed7a4e48fec67da3b92dddb47563e4935
- Original
- sites.uclouvain.be