In March 2020, commuting halted for much of the U.S. workforce. Downtowns went dark; the lunch rush went silent. People reorganized their apartments and houses to accommodate home offices. Zoom became a ubiquitous brand.
By now, despite the slow and uneven resuscitation of the workplace, remote work has become integral for many Americans. This carries profound implications for the economy, social organization, and the composition and structure of cities around the world.
But estimates of how many people are working remotely vary. The Bureau of Labor Statistics found that around 27% of the U.S. workforce was working remotely at least part time as of August and September 2022, while a handful of academic surveys have suggested that the number is closer to 50%.
“Unfortunately, we have a measurement problem,” said Hong-Yi TuYe, a PhD student at MIT Sloan. “When we’re thinking about designing public policy, what’s the right metric to use to make informed decisions?”
In a new working paper, TuYe and six coauthors discuss their own survey of remote workers and use its results to demonstrate that government statistics often undercount how many people are working remotely. Understanding a few basic differences in how the various surveys were conducted, from the wording of questions to the occupations included, can help synchronize the results and offer a foundation for policymakers moving forward.
TuYe’s coauthors on the paper are Stanford University professor Erik Brynjolfsson, PhD ’91; MIT Sloan professor Arizona State University research professor Christos Makridis; Princeton University professor Alexandre Mas; Adam Ozimek, who is now the chief economist at Economic Innovation Group; and University of Pennsylvania professor Daniel Rock, SM ’16, PhD ’19.
A new survey of remote workers
In a survey, nearly 50% of respondents reported working remotely at least once per week as of December 2020.
The researchers started by running a survey of their own, the Remote Life Survey, which they compared with results from other remote-work surveys. (The researchers plan to conduct future waves of this survey, which will allow them to track changes in remote work over time.)
Nearly 50% of respondents to the Remote Life Survey reported working remotely at least once a week as of December 2020. That percentage is far higher than federal data and sits at the upper end of outcomes from academic surveys. But as the researchers go on to explain, differences in survey methodology and wording go a long way toward explaining the discrepancies.
Key differences across survey designs
The researchers determined that there are four main reasons remote work statistics can vary:
1. Reaching people both online and through the mail versus online alone. The Remote Life Survey, unlike most other surveys, reached respondents both online and through the mail. “We found that it’s valuable to include some folks that you can only reach by mail,” TuYe said. People who are not regularly online are also less likely to work remotely, and so excluding them from surveys slightly inflates the measure of people who are working from home.
2. The distribution of occupations within a survey. Given that some jobs can shift to remote work arrangements relatively easily, such as finance or marketing jobs, whereas others, like nursing or restaurant work, cannot, the composition of survey respondents can influence the overall measure of remote work. If a survey happens to recruit more heavily from professions amenable to working from home, the work-from-home figures will be inflated. If the recruitment pulls more from “essential” industries, in which workers have to be on-site, the figures will be depressed.
“This offers a kind of warning to surveyors,” TuYe said. “If you have one set of occupations in your surveys and the government has another, then people need to take this into account when they’re trying to estimate how many people overall are working from home.” To help resolve this problem, TuYe and his colleagues developed a way for researchers to normalize their results based on the professions involved in the survey.
3. Including people who are self-employed. The researchers found that the government excluded from its sample people who are self-employed, a group that often works from home. This proved to be an important variable when tallying up how many people work from home, and it became more important as the pandemic got underway and levels of self-employment rose.
4. The wording of survey questions. This might be the most influential and, perhaps, straightforward factor, the researchers found. For example, the survey that the Bureau of Labor Statistics conducted explicitly asked people whether they were working from home “because of the coronavirus pandemic.” That wording excluded people who were working remotely before the pandemic, leading to a substantial undercount of the total number.
“If you prime people to think about the adoption of remote work since COVID-19, then you get a much lower number as a result,” TuYe said. “In this sense, the government’s measure is more like a flow — a description of change over time — where other surveys are measuring a stock.”
Why it’s important to get the numbers right
As TuYe and his colleagues note in their paper, remote work represents a massive and fast-moving shift in the world of labor.
Getting these numbers right is a crucial issue for economists and policymakers, given that it affects everything from real estate and public transportation to individual productivity and the job market.
Learning about common mistakes when interpreting remote-work numbers can help them avoid common pitfalls and create more accurate estimates.
“You can think of these findings as an alert to all different groups, from people in government to private employers or investors in commercial real estate,” TuYe said. “It’s often said that you cannot manage what you do not measure, but measurement alone is just a first step. We have to understand how that measurement works if we expect to put it to good use.”