Potential Jobs Impacted by Covid-19
In this blog, we conduct an exercise to determine the potential consequences of the Covid-19 pandemic on near-term labor market outcomes. This is not a forecast, but an attempt to provide some discipline around potential bounds of the number of jobs impacted by the crisis. We estimate that between nine and 26 million jobs are potentially affected,1 with a best guess of around 15 million. If these jobs are lost, the June unemployment rate could reach between 14% and 18%, with a best guess of around 15%.
This exercise is limited in a number of ways that we discuss a little later. In particular, some of the key parameters used for our calculations require significant guesswork given the unusual environment we are in. We also ignore a few issues in our calculations that may have an impact on our projections, including most importantly the recently passed $2 trillion fiscal package (the CARES Act),2 which provides significant incentives to keep employees on payrolls.
Our calculation has two basic parts. First, we estimate the share of workers in each occupation that may be working from home. We then use occupation-by-industry shares prior to the outbreak to infer the share of workers in an industry that can work from home. Second, we use the list of essential services from the Massachusetts shelter-at-home declaration to infer which industries might be considered essential, partially essential, or not essential. We chose Massachusetts because it was among the first to publish its list, and it is unlikely that differences across states are substantial enough to affect the estimates. We use these categorizations to guess what share of the not-working-from-home workforce in each industry may be able to work at still-operating establishments. This last piece requires significant guesswork, and therefore we tweak estimates for plausible low and high ranges.
Work from home
We infer telecommuting from the 2016 National Longitudinal Survey of Youth (NLSY79), which asks how many hours per week respondents typically work from home in their primary job. The NLSY79 is a nationally representative sample of individuals 14 to 22 when they were first interviewed in 1979; therefore, in 2016, this cohort is 51 to 59. We include individuals who worked at least one hour per week, had a non-missing number of hours working from home, and had a valid occupation classification. We classify individuals who worked more than five hours per week as potential telecommuters.
We then aggregate these responses to the occupation level. This allows us to identify “high telecommute” occupations, which we operationalize as those where more than 13% of workers worked from home at least five hours per week.3 These are mainly high-skill occupations. We then consider two scenarios, one in which 50% of the workers in high telecommute occupations will actually work from home, and one in which 75% will telecommute. To provide a further range of estimates, we repeat this exercise classifying individuals who worked more than one hour a week from home as potential telecommuters. Our estimates of the aggregate share of workers who could telecommute are shown in the first column of table 1. We estimate between 25.4% and 39.0% of the workforce telecommutes. Our baseline estimate uses the greater-than five hour, 75% category, or 36.7%.
Our occupation-specific estimates have a 0.79 to 0.87 correlation with those made by Dingel and Nieman (2020), who use task data from the O*NET database to infer telecommuting. Moreover, our baseline aggregate estimate of 36.7% is similar to Dingel and Nieman’s aggregate estimate of 34%.4 We also show how our labor market projections vary if we use the Dingel and Nieman occupation-specific telecommute shares.
Lastly, we calculate telecommuting by three-digit NAICS industry as the employment-weighted average share of work from home across two-digit SOC occupations within each industry. Using each industry’s value added share in GDP produces an estimate of the share of GDP that would be produced by telecommuters under our alternatives (column 2 of table 1).
1. Telecommute shares, by share of workers and share of GDP
|Scenario||Share of workers5||Share of GDP6|
|>5 hours/week telecommute|
|Actual occupation shares||13.7||14.0|
|50% of high telecommute work from home||25.4||27.0|
|75% of high telecommute work from home||36.7||39.1|
|>1 hours/week telecommute|
|Actual occupation shares||23.8||24.1|
|50% of high telecommute work from home||27.7||29.2|
|75% of high telecommute work from home||39.0||41.4|
Work outside the home
With over three-quarters of the working age population living in states or localities under “stay-at-home” orders as of March 30, and much of the rest of the country under voluntary self-quarantine,7 we next attempt to estimate the share of workers that cannot telecommute but might be able to work at still-operating establishments.
We start with the list of essential services from the stay-at-home order made by the state of Massachusetts.8 We then classify each three-digit NAICS industry as either fully essential, partially essential, or non-essential.9 In the case of partially essential three-digit industries, we match industry descriptions in the Massachusetts order to four-, five-, or six-digit NAICS industries and then sum up employment in these detailed industries to get an “essential share” for the three-digit industry.
The last step is to try to guess the share of workers not able to telecommute that are still able to work in each industry. This is by far the least disciplined part of our exercise. We make some educated guesses based on the fraction of each three-digit NAICS industry that we identify as essential, and the degree to which we think particular industries likely require workers on site. To reflect its extreme uncertainty, we also report low and high scenarios that essentially vary industry work-away-from-home employment by 10 percentage points in either direction.10
To come up with an estimate of total workers potentially on the job, we then sum up the share of telecommuters and the share working away from home in each industry, capping each industry at between 20% and 100% of its February 2020 employment level.
Table 2 reports our estimates of the number of jobs in which people are potentially unable to work because of the virus. Our baseline estimate is that 15 million jobs will be impacted through April. If all those jobs are lost, the unemployment rate would rise to 11.6% in June.11
With regard to alternatives, using the Dingel-Nieman telecommute shares instead of ours results in somewhat less damage. Our “low” projection reduces those at work by 10 percentage points relative to the baseline, subject to the 20% of February employment floor for non-essential industries. The high projection assumes 10 percentage points more individuals are at work than what is guessed in the baseline, subject to a cap at 100%. These scenarios result in nine to 26 million jobs impacted and, if all are lost, a June unemployment rate between 14 and 18%.
That said, it is possible that many people included in our estimates will not be counted as unemployed if they are being paid while not working. This number could be large, especially in light of the CARES Act, which provides incentives to firms to keep employees on payrolls. Moreover, it is unclear how the Bureau of Labor Statistics’ employment questions will categorize those who are not getting paid if they expect to return to their jobs when the quarantines are lifted and therefore are not making any effort to find a new job.12
2. Jobs impacted and potential June unemployment rate
|Scenario||Jobs impacted||Possible June unemployment rate|
|Baseline using Dingel-Neiman telecommute||–13.1 million||14.8%|
There are a number of issues that we ignore.
- We do not account for the impact of the CARES Act. Including this impact will change the size of our June unemployment projection.
- We do not explicitly distinguish localities by whether they are under a stay-at-home order. We informally try to adjust the total industry-specific share working away from home to account for the lost activity due to quarantining.
- We cap three-digit industry-specific employment at February 2020 levels. There are a few selected industries—e.g., hospitals, ground transportation and distribution, retail grocers—that are presumably ramping up employment. On the other side, we put a floor on three-digit industry-specific employment of 20% of February 2020 levels; some industries, e.g., restaurants, could fall below that level.
- We ignore any productivity implications associated with new working arrangements.
- We calculated the unemployment rates associated with our projected job losses prior to the most recent announcement of 6.65 million new unemployment insurance claims, and so may underestimate the true values.
Appendix 1. Share of workers that work from home at least 5 hours, by occupation
|Occupation||Share work from home||Assume 50% of high telecommute work from home||Assume 75% of high telecommute work from home||Occupation share (Feb 2020)|
|Architecture and engineering*||0.136||0.500||0.750||0.021|
|Arts, design, entertainment, sports, media*||0.198||0.500||0.750||0.021|
|Building and grounds cleaning, maintenance||0.043||0.043||0.043||0.034|
|Business and financial operations*||0.196||0.500||0.750||0.058|
|Community and social service*||0.209||0.500||0.750||0.018|
|Computer and mathematical*||0.427||0.500||0.750||0.036|
|Construction and extraction||0.066||0.066||0.066||0.052|
|Education, training, and library*||0.311||0.500||0.750||0.061|
|Farming, fishing, and forestry||0.000||0.000||0.000||0.007|
|Food preparation and serving related||0.034||0.034||0.034||0.053|
|Healthcare practitioners and technical||0.069||0.069||0.069||0.063|
|Installation, maintenance, and repair||0.028||0.028||0.028||0.031|
|Life, physical, and social science*||0.210||0.500||0.750||0.010|
|Office and administrative support||0.066||0.066||0.066||0.103|
|Personal care and service||0.110||0.110||0.110||0.026|
|Sales and related*||0.192||0.500||0.750||0.095|
|Transportation and material moving||0.024||0.024||0.024||0.075|
* indicates a “high telecommute” occupation” (i.e. more than 13% of workers worked from home at least 5 hours per week).
1 We realize that nearly every job has been affected by the virus in one way or another. We are measuring those at risk of being lost.
3 A list of the occupations, and their telecommuting shares, is in Appendix Table 1. The 13% threshold was judgmentally determined by the degree to which the occupation seemed to rely on computer usage.
4 29% of workers reported being able to work at home and 25% actually worked from home in the 2017 American Time Use Survey (ATUS, available online). Roughly 28% of 45–64 year old respondents in that survey worked from home, pretty similar to the aggregate and suggesting the NLSY age cohort we rely on is reasonably representative of the labor force more broadly.
5 Occupation weights are derived from the February 2020 CPS.
6 We calculate industry*occupation shares of workers from the 2018 ACS. We then assign the same occupation-specific telecommute shares to each industry and finally aggregate the industries using weights from the BEA’s estimate of value-added output by industry for Q3 2019.
7 Most populated areas that are not under stay-at-home orders appear to be reducing residential movement somewhat, similar to the rest of the country, at least according to phone usage reported online. Some rural areas, especially in the mountain and upper Midwest region, appear to be restricting movement less than the country as a whole.
9 According to our reading of the Massachusetts order, of the 83 three-digit NAICS industries, 17 are non-essential, 24 are partially essential, and 42 are fully essential.
10 In general, we assume that all workers are at work in industries that we identify as fully essential. For those that are partially essential, we assume that the fraction essential plus some remaining share of non-essential workers are at work. This remaining share reflects a best guess of the required workers that need to be on site. For those that are non-essential, we make our best guess of workers that actually are on site given the nature of the industry and anecdotal reports of the employment situation in that industry, if available. An alternative set of such assumptions are produced in a March 27 report by Nomura Global Markets Research. If we use a version of their assumptions in our model, they imply job losses of 24 to 27 million jobs, close to our more pessimistic projection.
11 Briefly, the unemployment rate forecasts come from a model that estimates the relationship between unemployment insurance (UI) claims, the UI take-up rate, employment growth, and earnings growth and the change in the unemployment rate between 1990 and 2019. It then feeds in an assumed path for the explanatory variables into the estimated relationships to generate a forecasted unemployment rate path. The UI claims path includes the 3.28 million claims for the week ending March 21 and assumes 4.75 million claims for the week ending March 28.
12 This may include the 10 million workers over the age of 65 and many others under 65 who are at high risk and may choose to stay at home rather than work away from home in an essential industry. 22% of individuals over the age of 65 telecommute, according to the ATUS.