Working from home has been a bad deal for companies and workers
New research concludes that working from home during the COVID-19 crisis lowered productivity, as people put in longer hours just to keep up with past performance
For most corporate leaders, one of the most important debates of the pandemic has been its impact on worker productivity. After all, no one had ever seen anything like the great Work From Office (WFO) to Work From Home (WFH) migration that took place in 2020. As the pandemic comes to an end, a growing divide pits those who want to maintain or even expand WFH models against those who believe that workers should return to the office.
Various articles during 2020 suggested that "white-collar" worker productivity had increased during the crisis, as commutes disappeared and workers had more time and flexibility in shaping their workday. However, new research from Michael Gibbs (Chicago), Friederike Mengel (Essex), and Christoph Siemroth (Essex) challenges these early optimistic assessments. With WFH looking to continue, and with many companies worrying about its implications on their businesses, this research provides leaders the first extensive empirical foundation for understanding the pandemic's effects on the productivity of highly skilled workers.
A novel aspect of this research is that rather than being based on surveys or secondary metrics, the authors gained access to the direct work activities of approximately 10,000 employees (out of a total workforce of 150,000) from the R&D team one of the world's largest IT services companies. Virtually all "have at least a bachelor's degree, often in a technology field such as computer engineering or electronics," and most "work at the company's large, modern corporate campuses," which "look and feel very similar to what one sees at Microsoft, Apple, or Amazon."
Fortunately for the researchers, the company mandates that all employees consent to the tracking of their daily work activities through a platform called Sapience. The company, note the researchers, has designed the system carefully. For example, Sapience "records the effective work hours, not the nominal work hours" which means that "even if an employee sits for 8 hours a day at their desk, Sapience might only record 5 hours, because breaks using the phone or surfing the web is not effective work time." Furthermore, because the Sapience data drives performance measurement and compensation, employees have strong incentives to continue to hit their goals after their transition to WFH.
Conveniently, Sapience tracks not just input but also employee output. The company uses a "normalized measure of output to make different jobs and roles comparable." For example, "for a programmer, the output measure might be programming tasks completed divided by tasks assigned, times 100." For other roles, "output might be the number of reviews (e.g., of code) completed relative to the monthly target, or the number of reports delivered relative to the target."
With input and output set, the authors calculated productivity "by dividing output by total time worked in a given month, measured in normalized output per hour worked."
To complement the Sapience data, the company acquired 914 licenses of Microsoft’s Viva Workplace Analytics [WPA] tool for a subset of the study's workers. Because WPA works with data from Outlook and other corporate systems, note the authors, "the company was able to extract retrospective measures for the pre-WFH period as well as for the WFH period."
From the Sapience and WPA data, the researchers calculated three metric sets. The first was working hours, which was simply overall time worked by an employee. The second was meeting characteristics: focus hours ("time spent working alone"), collaboration hours, meetings management, 1:1 meetings, coaching time, and Microsoft Teams calls. The third set contained information on other work tasks such as internal and external networking and time spent reading emails.
From their analysis, the authors concluded that during the pandemic, working hours per employee increased per day by roughly "1.9 hours (based on a seasonal time trend) or by 1.6 hours (with a linear time trend)," and worker output changed by "+0.25 percentage points or by -0.1 percentage points." Thus, WFH had no impact —positive or negative — on worker output. Productivity, however, was a different story:
Productivity decreased by -0.3 output percentage points per hour worked, or by 0.14 output percentage points per hour worked, depending on the time controls. Both are economically significant: if employees worked a maxed 40 hours per week, this would imply a drop in output of 12 or 5.6 output percentage points in a week. In other words, if employees had not increased the time worked during WFH, on average they would have completed only 88-94 of 100 tasks they were assigned.
Put simply, during the pandemic, employees worked longer but were less productive [italics mine]. The authors speculate that when working from home, most employees focus on achieving the same amount of work but must work longer hours to do it.
Interestingly, the authors consider the possibility that the firm's workers were somehow "gaming" or manipulating the time worked numbers, but they reject this possibility for several reasons. For example, in Sapience "merely keeping the computer on for longer or watching videos instead of working does not increase input.” Employees would have to put in “significant effort to fool the system — time that could be spent actually working.” Moreover, “gaming time measurement in Sapience would not translate into increases in the other time measurement in WPA,” because the "WPA time recording is from activity in MS Outlook, MS Teams, etc., rather than programming tools or similar, and is not dependent on mouse/keyboard activity."
With the main conclusion established, the authors go on to note other findings from their research. For example, they found that "employees with children work almost a third of an hour more per working day during WFH than employees without children, who themselves still work 1.4 hours more during WFH." Moreover, having children at home leads to "a 60% larger productivity drop compared to employees without children at home." Consequently, "the patterns we have seen for the average employee are exacerbated for employees with children."
Another finding is that men without children increased working time more than women without children. To the authors, this finding suggests a “gender difference in the WFH effect that is unrelated to childcare responsibilities.” Moreover, while men with children worked more than men without children, the numbers did not change for women. From these and other gender-based observations, the authors reach an interesting conclusion:
Our analysis therefore shows that female employees are more adversely affected by WFH, but this is not due to childcare responsibilities. The latter finding contrasts with much of the narrative in western countries, where childcare responsibilities are given as a main reason why women are more adversely affected by WFH. This does not seem to be the dominant effect in this country. Instead, we conjecture it is the greater expectations placed on women by parents and parents-in-law in the domestic setting that generates the gender difference.
One clearly negative finding was the detrimental effect that WFH had on internal networking and mentoring, both of which decreased during the pandemic. As the authors note, the number of both 1:1 supervisor meetings and coaching meetings decrease during WFH," and employees "seem to receive less mentoring and coaching." These lost opportunities to network, they suggest, "may help explain why WFH lowers productivity." Indeed, the risk of losing valuable mentoring time was identified early in the pandemic, and this study suggests those concerns were well-placed.
In the classical economic model, note the authors, "when inputs (labour and capital) are fixed, productivity and output go hand in hand." During the pandemic, however, output remained about the same while inputs increased, implying a decrease in productivity. This result is a negative outcome for both workers (who must work longer hours) and companies (who get a less productive workforce). In addition, all the time wasted in extra coordination and management activities, the lost mentoring effects, as well as the higher mental stress levels to workers and their families, and the overall conclusion is that pandemic-related WFH has been bad for most employees and their companies. Moreover, the oft-mentioned commuting time savings did not make up for these negative outcomes, note the authors, because "if we compute the sum of work and commute time during WFO, and compare it to work times during WFH, then employees still spent significantly more time at work during WFH than during WFO."
In concluding their analysis of what happened in the past year, the researchers accept that WFH models that were differently designed might have better outcomes. "Employees whose role allows for effective working from home might do so,” they note, “while others that rely more on interpersonal interactions might return to the office, at least for a few days a week.” Moreover, given how reticent some employees are to return to the office, it is very likely that companies will continue to experiment with WFH models, perhaps with different outcomes.
In the meantime, this important study opens a valuable window into what happened in 2020, suggesting that the stress and burnout so many workers reported (and continue to report) were real. The study also challenges the many advocates of expanding WFH options post-pandemic. They will either have to present empirical data with different results or suggest new models that will not leave both workers and companies on the losing end of the WFO-WFH tradeoff.
Gibbs, Michael and Mengel, Friederike and Siemroth, Christoph, Work from Home & Productivity: Evidence from Personnel & Analytics Data on IT Professionals (May 6, 2021). University of Chicago, Becker Friedman Institute for Economics Working Paper No. 2021-56, Available at SSRN: https://ssrn.com/abstract=3843197.