COVID-19 and Remote Work:

An Early Look at US Data∗

Erik Brynjolfsson John Horton
MIT, Stanford & NBER MIT & NBER
Adam Ozimek Daniel Rock Garima Sharma
Upwork MIT MIT
Hong Yi Tu Ye
MIT
April 8, 2020

Abstract
We report the results of a nationally-representative sample of the US
population on how they are adapting to the COVID-19 pandemic. The
survey ran from April 1-5, 2020. Of those employed four weeks earlier,
34.1% report they were commuting and are now working from home.
In addition, 11.8% report being laid-off or furloughed in the last 4
weeks. There is a strong negative relationship between the fraction in
a state still commuting to work and the fraction working from home
which suggests that many workers currently commuting could be con-
verted to remote workers. We find that the share of people switching
to remote work can be predicted by the incidence of COVID-19 and
that younger people were more likely to switch to remote work. Fur-
thermore, using data on state unemployment insurance (UI) claims,
∗
MIT’s COUHES ruled this project exempt (project number E-2075). Code & Data:
https://github.com/johnjosephhorton/remote work/. Thanks to Sam Lord for helpful
comments. Thanks to Todd Smith for reminding us that Alaska and Hawaii are US
states.

1
 we find that states with higher fractions of remote workers have higher-
than-expected UI claims.

1 Introduction
The on-going COVID-19 pandemic has confined large numbers of people
to their homes via quarantines and shelter-in-place orders. Large numbers
of businesses are closed and many people are not working remotely There
have already been enormous and unprecedented increases in workers filing
unemployment insurance claims (Goldsmith-Pinkham and Sojourner, 2020).
To get a real-time sense of how firms and workers are responding, we
conducted a survey using Google Consumer Surveys (GCS).1 We asked a
single question: ““Have you started to work from home in the last 4 weeks?”
with the following response options:

1. “I continue to commute to work”

2. “I have recently been furloughed or laid-off”

3. “Used to commute, now work from home”

4. “Used to work from home and still do”

5. “Used to work from home, but now I commute”

6. “None of the above / Not working for pay”

We launched our survey on April 1, 2020 and collected responses until

April 5, collecting a total of 25,000 responses. We find that over one third of
workers have responded to the pandemic by shifting to remote work, while
1
GCS is a relatively low-cost tool for rapidly collecting responses to simple questions
Stephens-Davidowitz and Varian (2014), and response representativeness is often compa-
rable to similar alternatives (Santoso et al., 2016; Brynjolfsson et al., 2019).

2
another 11% have been laid-off or furloughed. There is a great deal of varia-
tion across states in the share of people switching to remote work as well the
share of people who continue to commute. We find that these can each be
predicted by incidence of COVID-19. We also find that younger people were
more likely than older people to switch from commuting to remote work.

2 Results
Of the respondents, 14,173 reported something other than “None of the
above...” This gives an implied employment rate of 57%, which is slightly
lower than the BLS estimate of about 60%.2 For the rest our analysis, we
restrict our sample to those reporting being employed four weeks prior.
The distribution of answers pooled over all respondents is shown in Fig-
ure 1. We can see that the most common response from workers was that
they continue to commute, at 37.6% (95% CI is [36.3,38.9]). But the next
most common was that they have switched from commuting to working from
home.
The fraction of workers who switched to working from home is about
34.1%. In addition, 14.6% reporting they were already working from home
pre-COVID-19. This suggests nearly half the workforce is now working from
home, significantly more than the Dingel and Neiman (2020) estimate of 34%
of people working at home.3
2
https://fred.stlouisfed.org/series/EMRATIO
3
Our estimates are broadly consistent with the literature, however there is a relatively
wide range of estimates. Krantz-Kentkrantz (2019) uses 2013-2017 American Time Use
Survey (ATUS) data to show 20.5% of workers working from home in some way on an
average day. However, our question implies working from home all the time. The remote
worker fraction in the ATUS is 11.4%. Our 14.2% estimate is also broadly consistent
with the “Freelancing in America Survey” that reported 16.8% of workers report doing
most or all of their work remotely, though this includes people working from co-working
spaces, coffee shops, homes, etc (Ozimek, 2020). At the lowest end, the 2019 Census
reports 5.3% of workers as “working from home.” The wide range in answers suggests
respondent uncertainty about the precise meaning of questions. Nevertheless, our results
lie well within the existing estimates.

3
Figure 1: Answers to the question “Have you started to work from home
in the last 4 weeks?”, conditional upon being in the labor force from a US
sample

I continue to commute to work ●

Used to commute, now work from home ●

Used to work from home and still do ●

I have recently been furloughed or laid−off ●

Used to work from home, but now I commute ●

0% 10% 20% 30% 40%

% of working respondents

We find that 11.8% of workers report recently being laid off or furloughed.
If we take the US labor force at about 165 million, this implies that about 16
million Americans are recently out of work. The total UI filings for the last
two weeks adds up to 9.9 million.4 However, not all unemployed have yet filed
for unemployment. Given the decline in hiring, Wolfers (2020) estimates we
are dealing with about 16 million unemployed, matching our point estimate.5
GCS also infers respondent gender. We analyzed responses by gender but
did not find any notable differences.

2.1 Geographic variation

COVID-19 has affected various parts of the US differently, with the main
epicenter in New York City. In Figure 2, we plot the fraction of respondents
choosing each answer by region. GCS captures a respondent’s city and state,
4
https://www.dol.gov/ui/data.pdf
5
https://www.nytimes.com/2020/04/03/upshot/coronavirus-jobless-rate-great-depression.
html

4
which are then mapped to the regions “Northeast”, “Midwest”, “West”, and
“South.”
In the first facet from the left, we can see that the South has the highest
fraction still commuting to work and the Northeast has the lowest. In the
second facet from the right, we can see that the Northeast has the highest
fraction of respondents switching to working from home, and the South has
the fewest. The Northeast started from the lowest fraction working from
home, though these fractions are imprecisely estimated and are all fairly
similar to each other. The Northeast fraction now working from home is
over 40%.
For a finer-grained look, we plot responses by state. In Figure 5 we plot
the fraction of respondents that switched to working remotely. As we saw in
Figure 2, the highest fractions now working from home are in the Northeast.
The South and parts of the Midwest show substantially less remote work. It
is important to keep in mind that some of these point estimates are fairly
imprecise.

2.2 By gender and age

In Figure 3 we report responses by inferred gender. Fractions are computed
separately for males and females, and then a slope graph is used to show
differences. Within all questions at a 95% confidence interval, the differences
between gender are not statistically significant. Within our sample, however,
it appears that men were modestly more likely to continue to commute to
work, and likewise women more were more likely to report switching from
commuter to work from home status. Men were also slightly more likely to
have been recently furloughed or laid-off. Consistently working from home
workers show little different in gender composition.
In Figure 4 we report responses by inferred age. A similar proportion of
workers continue to commute to work across all age groups, as is also the case
for the recently furloughed or laid-off worker contingent. On the other hand,

5
 Figure 2: Responses by US region

I have recently
I continue to
been furloughed
commute to work
or laid−off

18%
●
40% 15%
●
●
●
36% 12% ●

● ●
32%
● 10%
% of respondents

8%

Used to commute,
Used to work from
now work from
home and still do
home

21%
40%
●
18% ●

36% 15% ●
●
●
● ●
32% ● 12%

9%
T

TH

TH
S

ES

ES

ES

ES
EA

EA
U

U
W

W
SO

SO
H

H
ID

ID
RT

RT
M

M
O

O
N

6
 Figure 3: Responses by gender

I continue to
40% ●
commute to work
●
● Used to
● commute, now

work from home

30%
% of respondents

Used to work
20% from home and
still do
●

● I have recently
●

●
been furloughed
10%
or laid−off

Used to work
●
from home, but
●

0% now I commute

Female Male

7
the proportion of respondents that has recently converted from commuting
to work to remote work steadily declines from the 25-34 age group to the 65
and older category. The differences between the 25-34 age group and the 65
and older group are statistically significant, and, as Figure 4 shows, younger
workers (above age 25) are more likely to have been converted to work from
home from commuting.
Survey respondents in older age groups also reported remaining working
remotely with greater propensities. These results are directionally consistent
with the 2019 Census Bureau (2018), though our estimates are larger. The
differences may arise from a difference in the question asked. The Census
asks about how workers get to work. The 2019 Upwork “Freelancing in
America” study found younger workers were modestly more likely to work
mostly or entirely from home Upwork (2019). It is possible that our survey
is somewhere in between, grouping people who do some work at home with
those who are fully committed remote labor. We will investigate further in
future work.
The fractions laid-off or furloughed by US State are shown in Figure 6.
In Figure 7 we plot the fraction of respondents working from home versus
the fraction still commuting by US state. There is a clear negative rela-
tionship, suggesting a fraction of current commuters will likely—or could—
transition to work-from-home status. Each 10 percentage point increase in
the fraction still commuting is associated with about a 6 percentage point
decline in the fraction of workers now working from home.
Table 1 documents how heterogeneity in COVID infection rates (mea-
sured as the log of cases per 100,000 individuals6 ) affects switching to remote
work or continuing to work from home. Column (1) shows that a 100% rise
in COVID-19 cases per 100k individuals is associated with a 5% rise in the
fraction of workers who switch to working from home. Column (2) shows a
6
Data accessed on April 5, 2020 from The New York Times website:
https://www.nytimes.com/interactive/2020/us/coronavirus-us-cases.html#states

8
 Figure 4: Responses by inferred age

I continue to
40% ●
●
commute to work
● ●
● ● ●
●

● ● Used to
● commute, now
% of respondents

30% ●
work from home

Used to work
20% ●
from home and
still do
●
●
●

● ●
●
● I have recently
● ● ●
10% ●
been furloughed
or laid−off

18−24 25−34 35−44 45−54 55−64 65+

Age of respondents

9
 Figure 5: Fraction now working remotely, by US State

0e+00
frac

0.5
lat

0.4
−1e+06
0.3

0.2

−2e+06

−2e+06 −1e+06 0e+00 1e+06 2e+06

long

Figure 6: Fraction Laid-off/furloughed, by US State

0e+00
frac
0.25

0.20
lat

−1e+06 0.15

0.10

0.05

−2e+06

−2e+06 −1e+06 0e+00 1e+06 2e+06

long

10
 Figure 7: Still commuting versus work from home fractions by US State

60% DC

50%
MA NM
NH DE MO NY
40% NJ CT ME IN IL
Now WFH

MD KY
MI KS TX VAWI VT
CA OK
WA PA IA
CO LA TN
30% UT ID AZ OH SC MT NC ND
RI NE
AK OR FL WV GA MN
NV MS AL
20%
HI AR WY

10% SD
20% 30% 40% 50%
Still commuting to work

similar rise in COVID-19 incidence as predicting a 5.4% fall in the fraction of

those continuing to commute. We would expect these relationships if higher
spread is associated with higher responsiveness of government or individuals.
Surprisingly, we do not find a strong relationship between our measure of
incidence and survey reports of being furloughed or laid-off. These associ-
ations are not to be interpreted as causal and future work will explore the
causal effect of the pandemic on switches into remote work.
A natural question is how these various measures are affecting UI claims
by state. In Table 2, we combine our data with that on UI claims from
Goldsmith-Pinkham and Sojourner (2020).
We regress the log of two weeks of UI claims in a state on the state’s popu-
lation and the log state-specific fraction for each of the response possibilities.
Unsurprisingly, across all specifications, the state population explains a great
deal of the variation in UI claims. We are interested in asking whether our
survey measures account for some of the residual variation.

11
 Table 1: Predicting remote work by state incidence of COVID-19

Dependent variable:
(1) (2) (3)
Work from home Continue to commute Furloughed or laid-off
(1) (2) (3)
Log cases per 100k pop 0.0501∗∗∗ -0.0545∗∗∗ -0.000305
(0.0116) (0.0147) (0.00700)
Constant 0.136∗∗∗ 0.590∗∗∗ 0.122∗∗∗
(0.0472) (0.0590) (0.0275)

Observations 51 51 51
R2 0.230 0.188 0.000

Notes: Significance indicators: p ≤ 0.05 : ∗, p ≤ 0.01 : ∗∗ and p ≤ .001 : ∗ ∗ ∗.

To interpret these results, consider a state i that has population P opi ,

an employment rate of Ei and a COVID-induced fraction of the working
population laid-off of LOi . If UI claims perfectly measured lay-offs, then the
state’s UI claims would be P opi Ei LOi and so a regression of log UI claims
on the log of each of the terms should give each component a coefficient on
1. In the regressions in Table 2, we use a state-specific employment rate
pre-COVID calculated from our own survey and the state population.
In Column (1), we append the state-specific fraction reporting that they
were still commuting to work. The higher the fraction reporting still com-
muting, the lower the UI claims for that state. States that still have large
numbers commuting to work should have fewer lay-offs, and so the coefficient
on the fraction still commuting should be negative.
In Column (2), the greater the fraction that reports working from home,
the higher the UI claims. The fraction working from home should, on the
one hand, have a protective effect, keeping workers from filing for UI, but on
other hand, a state with a high WFH fraction has likely had a particularly
severe labor market shock, with many additional workers laid-off. What

12
 Table 2: Predicting UI claims by state

Dependent variable:
Log state two week UI claims
(1) (2) (3) (4)
Log state population 0.967∗∗∗ 0.936∗∗∗ 0.996∗∗∗ 0.992∗∗∗
(0.063) (0.066) (0.066) (0.068)
Still commuting frac. (log) −0.744∗∗∗
(0.254)
Switch to WFH frac. (log) 0.695∗∗∗
(0.248)
Laid-off frac. (log) 0.255
(0.204)
Still WFH (log) 0.101
(0.218)
log(lfpr) −0.393 −0.598 −0.189 −0.110
(0.808) (0.825) (0.862) (0.878)
Constant −4.185∗∗∗ −2.283∗ −3.209∗∗ −3.456∗∗
(1.097) (1.211) (1.238) (1.326)
Observations 50 50 50 50
R2 0.854 0.852 0.833 0.828
Adjusted R2 0.845 0.843 0.822 0.817
Notes: Significance indicators: p ≤ 0.05 : ∗, p ≤ 0.01 : ∗∗ and p ≤ .001 : ∗ ∗ ∗.

13
seems likely is that workers who would otherwise be continuing to commute
to work are splitting into (a) work-from-home or (b) filing for UI. As states
impose further restrictions on mobility, we should be able to get a sense
of the resulting increases in UI filings based on how many workers can be
successfully transitioned to remote work.
Surprisingly, Column (3), which includes a direct measure of reported
lay-offs has the “right” sign, but is small in magnitude. Furthermore, our
expectation is that the elasticity of UI claims to the number laid-off should
be 1. However, not all states have the same ability to process claims, and
so there is an omitted variable, ri , that is the probability that a truly laid-
off worker is able to file an UI claim successfully. If ri falls the larger the
magnitude of LOi , then the coefficient on LOi would be biased downward.

2.3 Implications and suggestions for future work

These are a set of preliminary analyses of an evolving crisis. We have docu-
mented some early shifts in the economy, and it remains to be seen if some
of these changes are last beyond the end of the pandemic. For instance, once
busineess and individuals invest in the fixed costs of remote work, including
technology but perhaps more importantly in developing the necessary human
capital and organizational processes, then they may decide to stay with the
new methods. Furthermore, the crisis has forced people to try out new ap-
proaches, some of which may be unexpectedly efficient or effective. In either
case, lasting changes from the crisis would be expected.
Additional work to understand these changes is needed. Future research
will address how state-specific occupational distributions will affect the share
of labor performed by remote workers, as well as the impact of the occupa-
tional distribution on the UI system. Long term changes may involve not only
remote work, but also the structure of industries and international trade. For
example, tasks that can be done by remote workers may be more likely to be
off-shored, as distance becomes less relevant. The tasks the comprise many

14
occupations may be unbundled and re-bundled to separate those that require
in-person presence at a business from those that can be done remotely.
Critical to those decisions made by employers are the productivity effects
of remote work. More evidence is needed to evaluate the productivity changes
induced by allowing work from home. Relatedly, what percentage of tasks
can be done remotely and how does it vary across professions and industries?
Can we also observe (and explain) heterogeneity across states? These task-
specific questions will be the focus of our next round of surveys.
For the COVID-19 pandemic in particular, we are also interested in how
the disease spreads differentially across types of jobs. Remote work is one
way in which employers can protect both the health and job security of their
employees.

3 Conclusion
We document some early facts about how the US labor force is responding
to COVID-19 pandemic. In particular, we find that in the past four weeks
over one third of the labor force has switched to remote work. The state-
level COVID-19 infection rates predict these switches. If there is hysteresis
as people learn new ways to work remotely and businesses reorganize, the
pandemic-driven changes may portend more lasting effects on the organiza-
tion of work. We will continue to track changes to the nature of remote work,
asking how pandemic-induced changes transform workplaces in the short and
long-term.

The code and data for this project are here: https://github.com/johnjosephhorton/remote work/.

15
References
Brynjolfsson, Erik, Avinash Collis, and Felix Eggers, “Using massive
online choice experiments to measure changes in well-being,” Proceedings
of the National Academy of Sciences, 2019, 116 (15), 7250–7255.

Bureau, US Census, “American Community Survey One-Year Public Use

Microdata Sample, Table B08101,” 2018.

Dingel, Jonathan and Brent Neiman, “How Many Jobs Can be Done
at Home?,” Working paper, 2020.

Goldsmith-Pinkham, Paul and Aaron Sojourner, “Predicting Initial

Unemployment Insurance Claims Using Google Trends,” Working paper,
2020.

Krantz-Kentkrantz, Rachel M, “Where did workers perform their jobs

in the early 21st century?,” Monthly Labor Review, 2019, pp. 1–10.

Ozimek, Adam, “Report: Overboard on Offshore Fears,” 2020.

Santoso, Lie Philip, Robert Stein, and Randy Stevenson, “Survey

experiments with Google consumer surveys: Promise and pitfalls for aca-
demic research in social science,” Political Analysis, 2016, 24 (3), 356–373.

Stephens-Davidowitz, Seth and Hal Varian, “A hands-on guide to

Google data,” Mountain View, CA, 2014.

Upwork, “Freelancing in America - 2019,” Report, 2019.

16