w30292pdf
w30292pdf
Page 1 of 520
nick Bloom shared this file. Want to do more with it?
  1. NBER WORKING PAPER SERIESHOW HYBRID WORKING FROM HOME WORKS OUTNicholas BloomRuobing HanJames LiangWorking Paper 30292http://www.nber.org/papers/w30292NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts AvenueCambridge, MA 02138July 2022, Revised January 2023We thank our formal discussant Ed Glaeser and audiences at Berkeley, Stanford, USC, UCLA, SAIF-FISF, AMES 2022, ESAM 2022, AEA, CESI, NBER, Scancor, AEI, and Luohan for comments. We thank the Smith Richardson Foundation and Toulouse Network for Information Technology for co-funding. Finally, we thank Jennifer Cao, Tracy Zhang, Sherry Ye, Fangyuan Chen, Xing Zhang, Ying He, Junyi Li, Byron Ye from Trip.com and Mert Akan and Shelby Buckman from Stanford for data, advice, and logistical support, and Lindiandian Yi for research assistantship. Conflict of interest, AEA, and IRB statement: No funding was received from Trip.com. James Liang is the co-founder, former CEO, and current Chairman of Trip.com. No other co-author has any financial relationship with Trip.com. Neither the results nor the paper was pre-screened by anyone. The experiment was registered with the American Economic Association after the experiment had begun (when Bloom and Han were invited to analyze the data) but before any data was analyzed. The experiment was IRB-exempt as the experiment was initiated by Trip.com before Bloom and Han joined the project, and only anonymous data was shared with the Stanford team.The views expressed herein are thoseof the authors and do not necessarily reflect the views of the National Bureau of Economic Research.NBER working papers are circulated for discussion and comment purposes. They have not been peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies officialNBER publications.© 2022 by Nicholas Bloom, Ruobing Han, and James Liang. All rights reserved. Short sections oftext, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit,including © notice, is given to the source.
  2. How Hybrid Working From Home Works OutNicholas Bloom, Ruobing Han, and James LiangNBER Working Paper No. 30292July 2022, Revised January 2023JEL No. J0ABSTRACTHybrid working from home (hybrid), whereby employees work a mix of days at home and at work each week, has become common for graduate employees. This paper evaluates a randomized control trial of hybrid on 1612 graduate engineers, marketing and finance employees of a large technology firm. There are four key results. First, hybrid was highly valued by employees on average, reducing attrition by 33% and improving job-satisfaction measures. Second, hybrid reduced working hours on home days and increased them on office days and the weekend, altering the structure of the working week. Third, hybrid increased messaging and video calls, even when all employees were in the office, reflecting a move towards more electronic communication. Finally, there were large differences in the valuations of hybrid between managers and non-managers. Non-managers were more likely to volunteer into the hybrid experiment, to work from home on eligible days, to predict positive impacts on productivity, and to reduce their attrition under hybrid. In contrast, managers were less likely to volunteer, less likely to work from home on eligible days, predicted a negative average impact of hybrid on productivity, and saw increased attrition rates under hybrid.Nicholas BloomStanford UniversityDepartment of Economics579 Jane Stanford WayStanford, CA 94305-6072and NBERnbloom@stanford.eduRuobing Han579 Jane Stanford Way, Stanford, 94305.California, USA.robinhan@stanford.eduJames LiangTrip.com 968 Jin Zhong Road, Shanghai, 200335Chinaand Peking Universityliangjz@trip.com
  3. 21.IntroductionWorking from home (WFH) has been increasing for several decades in the United States but surged after the COVID-19 pandemic. By early 2023 about 30% of full-paid daysareworked from home, with hybrid WFH being the most common approach to this (Barrero et al., 2023). Hybrid WFH – hereafter called hybrid – involves typically 2 to 3 days a week at home and the other days in the office.Hybrid has ispromoted as combining the best of working in the office and working from home. The idea is to break an employee’s working week into tasks, distinguishing between tasks that are typically best in person, like meetings, training events, or mentoringon office days, and those that are best individually, like reading, writing, or coding on home days. This hybrid WFH approach has four benefits. First, on home days, employees avoid having to commute and prepare for work, which for the average US employee savesabout 70 minutes a day.1Second, home working is often better for individual-focused activities like coding or writing as it is usually quieter.2Third, WFH also allows greater time flexibility. For example, employees can break to exercise, to go to the doctor or pick their children up from school. Finally, hybrid WFH can also reduce space costs iffirms rotate the days in which teams work from home. Detractors argue, however, that hybrid WFH is complex, suffers from transition costs from switching between fully in-person and fully remote, and is detrimental to employee performance and innovation. Not surprisingly, there is a vigorous debate about the efficacy of hybrid WFH among managers, policymakers, and the media. In this paper, we describe a randomized controltrialthat took place in 2021 and2022 when Trip.com, a NASDAQ-listed global travel agent with 35,000 employees headquartered in Shanghai, decided to evaluate hybrid WFH. Their motivation was to improve employee job satisfaction to reduce attrition and ease hiring. The obstacle to implementing hybrid WFH was managers’ concernsthat employees would underperform on their days at home. So Trip.com decided to formally evaluate a hybrid WFH system in two divisions over six months before making a decision over a full firm roll-out. 1See, for example, Barrero et al. (2020)and Teodorovicz et al. (2022). 2For example, Bloom, Liang, Roberts and Ying (2015) found a 4% increase in per-minute productivity for home-working on individual tasks, which was primarily attributed to a quieter home working environment.
  4. 3They took the 1612 engineers, marketing, and finance employees in the Airfare and IT divisions and randomized them into providing the option to WFHon Wednesday and Friday. Importantly, this contained 1219 non-managerial employees and 393 managersas the firm was keen to evaluate the impact of hybrid-WFH on both managers and non-mangers. Employees with odd birthdays (those born on the 1st, 3rd, 5th,etc., day of the month) were randomized into the treatment group allowing hybrid-WFH, while those with even birthdays were the control group who continued to come into the office full-time as before.The experiment revealed four results. First, WFH reduced average attrition rates by 33% and improved self-reported work satisfaction scores, highlighting how employees place a sizeable average value on this amenity. This is consistent with the prior results from Mas and Pallais (2017 and 2018), Maestas (2018) and Barrero et al. (2023) that employees value working from home at the equivalent of about a 4% to 8% wage increase. Second, WFH reduced hours worked by around two hours on home days, but increased it on in office workdays and on the weekend, and reducedsick days and holidays. Employees reported working from home afforded them the flexibility to attend a dentist appointment, pick their children up from school, exercise,or travel to their hometown early on a Friday. This matches the survey evidence from the US that the second-largest benefit of working from home is flexibility (the largest is avoiding commuting). Third, WFH employees increased individual messaging and group videocall communication, even when all employees were in the office, reflecting the impact of remote work on modes of communication. This suggests that home working leads to persistent changes in employees’ behavior even in their days in the office. Interviewing employees, we heard that they became accustomedto a more electronic communication, carrying this over to their days working in the office. Fourth, we found striking differences in impact and opinions on hybrid between non-managers and managers. Non-managers were significantly more likely to volunteer to be in the first wave of the experiment, to take-up the option to work at home on eligible Wednesday and Fridays, to report a positive impact of hybrid on productivity, and to display lower quit rates in the hybrid treatment group. This was
  5. 4particularly true for employees with longer-commutes. In contrast, managers reported significantly negative impacts of hybrid-WFH on their productivity, were less likely to volunteer and work at home on eligible days and displayed higher quit rates of those randomized into the hybrid treatment group. There was also one important null result, which is we found no significant impact of hybrid treatment on employees’performance reviewsor promotion rates. Lines of code written, another measure of employee productivity for IT engineers, rose by 4.4% for hybrid employees, but this was insignificant. Given the large 1612 employee size of this experimental sample, this rules out substantial positive or negative impacts of hybrid-WFH on productivity. This is important as the main criticism of hybrid-WFH has been its negative impact on productivity, specifically in collaborative environments, which executives like Elon Musk at Teslaand Twitter, Jamie Dimon at JP Morgan, and David Solomon at Goldman Sachs have noted when pushing for a full return to the office. The prior randomized control trial evidence on WFH has focused on call center employees, who, while easy to measure and evaluate, do not have much of a team or creative component intheir job. In contrast, the Trip.com employees in this experiment are graduate employees working in teams and creating new products and services, in both managerial and non-managerial roles. So, this larger sample null result – which also holds in all major sub-samples – is notable. Once the 6-month experiment ended, the firm evaluated the data on performance, attrition, employee satisfaction, and communications and decided to roll the WFH policy out to the entire company at an executive meeting. Their view was the impact on reduced attrition was extremely positive with no negative impacts on performance and some potential longer-run costs savings from reduced office space. This was announced on February 14th, 2022, and was immediately picked up by the Chinese media, with coverage in Reuters, South China Morning Post, US News, Yahoo and the Standard since working from home was exceedingly rare in China. Since then, several other Chinese tech firms following this experiment have started adopting similar policies.Thispaper connects to threestrands of literature.
  6. 5First, there is literature on the adoption and impact of working from home on firms and their employees. The closest papers are Bloom et al. (2014), who run a field experiment at cTrip.com3 on 250 call-center workers, and Emanuel and Harrington (2021), who examine the call center of a large US firm. Both find positive productivity effects of working from home of around 10%, reduced attrition but negative promotion effects. The challenge with these papers is that call center employees are not really team workers or involved in creative tasks, so it is hard to extrapolate these findings to professional or managerial employees. Choudhury et al. (2019) study US Patent examiners who were allowed to work from anywhere, finding productivity rose by 4.4%, although they note that patent examiners, like call center workers, are in roles thatdo not require significant collaboration. Choudhury et al. (2022) study103 employees in an 8-week experiment in an NGO in Bangladesh, finding in an RCT that hybrid WFH is associated with higher levels of emails and increased productivity. Kunn et al. (2021) exploits natural variation in remote vs. in-person competitions in professional chess during the pandemic and find a negative impact of the remote activity, but again chess playing is not a team activity. Finally, Gibbs et al. (2022) examine graduate team employees in an Asian IT firm who shifted to working from home during the pandemic. They find large negative effects on productivity, although their result examines a shift from fullyin-person tofullyremote executed at speed at the onset of the pandemic withouta control group, making these results hard to assess. Our study is, as far as we know, the only randomized evaluation of the impact of hybrid working-from-home, and importantly the only study to include hundreds of managerial and non-managerialgraduate employees. A second piece of literature tries to evaluate the impact of working from home through self-assessed performance surveys. Etheridge et al. (2020) find that employees who work from home state they are about as productive as in the office. Barreroet al. (2023) report US employees to perceive about a 3% to 5% increase in productivity, while Aksoy et al. (2022) report slightly lower (but still positive) figures from 20 countries globally. Finally, recent research examinesthe extent and incidence of WFH during the COVID pandemicand the outcomes associated with WFH.See, for example, Adams-Prassle et al. (2020), Althoff et al. (2020), Bartik et al. (2020), Bick et al. (2020), Brynjolfsson et al. (2020), Mongey et al. (2020) and Ozimek 3 cTrip.compurchased Trip.com, a smaller Singaporean travel agent, and adopted the name. So cTrip.com and Trip.com are ostensibly the same firm.
  7. 6(2020). Bai et al (2021), Davis et al. (2021), Favilukis et al. (2020), Pagano et al. (2020), and Papanikolaou and Schmidt (2020) study the relationship between firm-level stock returns during the COVID pandemic and the capacity of their employees to work from home. Behrens et al. (2021) and Davis et al. (2022) offer general equilibrium analyses of WFH and its consequences, while Gupta et al. (2021) look at the impact of WFH on property markets, Liu and Su (2021) the impact of density via WFH on valuationsand Delventhal et al. (2021) impacts on city structure. Like us, they stress that the effects of a shift to WFH are highly uneven across people and locations. Section II discusses the experimental design, section III the impact of attrition and employee satisfaction, section IV the impact of messaging and communication, and section V the impact on performance evaluations, promotions, and productivity. Finally, section VI concludes.II The ExperimentOur experiment took place at Trip.com, the third-largest global travel agent by sales in 2019 (after Expedia.com and Booking.com). Trip.com was established in 1999, was quoted on NASDAQ in 2003, and was worth about $20bn at the time of the experiment. It is headquartered in Shanghai with offices across China and internationally, accommodating its roughly 35,000 employees.In the summer of 2021, Trip.com decided to evaluate hybrid WFHafter seeing itspopularity amongstUS tech firms. Thefirm believed this could improve employee jobsatisfaction, reducing the costs of employee attrition, and potentially improving productivity. The key obstacle to implementing hybrid WFH was the concern of many managers that employees would underperform on their days at home. In addition, in 2021, no major Chinese firm was offering hybrid WFH, withtotal attendance at the office the norm. So Trip.com decided to formally evaluate a hybrid WFH system in two divisions over six months before making a decision over a full firm roll-out.
  8. 7The first step took place on July 27th, 2021, when the firm surveyed 1612 eligible4engineers, marketing and finance employees in the Airfare and IT divisions about the option of hybrid WFH. They chose these two divisions as representative of the firm, with a mix of employee types to assess any potentially heterogeneous impacts. Just over 70% of the employees in these divisions are technical employees, writing software code for the website, and front-end or back-end operating systems. The remainder work in business development, so talking to airlines, travel agents, or vendors to develop new services and products, in market planning and executing advertising and marketing campaigns, and in business services, dealing with a range of financial, regulatory, and strategy issues. Across these groups 393 were managers and 1219 non-managers, providing sufficient samples of both groups to evaluate their response to hybrid working from home. The employees were sent an email (see Appendix A1) outlining how the 6-month experiment offeredthem the option (but not the obligation) to WFH on Wednesday and Friday. After the initial email and two follow-up reminders, a group of 518 employees volunteered. The firm randomized odd employees, those born on the 1st, 3rd, 5th, 7th, etc., to become eligible for the hybrid-WFH schemestarting on the week of the 9thof August. The top management at the firmwas surprised at the low volunteer rate of the optional hybrid-WFH scheme. They suspectedthat many employees were hesitating because of concerns that volunteering would be seen as a negative signal of ambition and productivity.5 So, on 6thSeptember, all the remaining 1094 non-volunteer employeeswere told they were also included in the program. The odd birthday employees were again randomized into the hybrid WFH treatment andbegan the experiment on the week of 13thSeptember.Figure 1 shows some pictures of employees working in the office (left side) and employees working from home from October 2021(right side). A few points are worth noting. First, in 2021 COVID incidence rates in Shanghai were extremely low, so employees were neither masked nor socially distancedat the office. So, this office control baseline is comparable to a pre (or post) COVID situation 4This excludes interns and rookies that are in probation periods, as onsite learning and mentoring are very important for them.5This is not unreasonable. For example, Harrington and Emmanuel (2021) found in the US firm they evaluated that work-from-home employees were negatively selected on productivity.
  9. 8in the US. Second, employees all workedin modern open-plan offices in desk groupings of four or six colleagues from the same team. In contrast, when working from home, they usually worked alone in their apartments, typically in the living room or kitchen.Figure 2 shows the overall office (on the left), highlighting thisis a large modern building, similar to many large US and European corporate offices. The figure also shows an example floorplan for part of the second floor, highlighting how entire teams tend to have their desks clustered together. The experimental sample are typically in their mid-30s, about two-thirds are male, with all of them having a university undergraduate degree and almost one-third having a graduate degree (typically a master’s degree). In addition, nearly half of the employees have children (details in Appendix Table A2). In Table 1 we examine the decision to volunteer for the work from home experiment. We see that non-managers and employees with a long commute were significantly more likely to volunteer. The magnitudes of these differences are also large, with managers almost 38% (-0.38=-0.134/0.353) less likely than non-managers to volunteer. We see female employees are less likely (-17%) to volunteer than male employees. Interestingly, there is no relationship between volunteering and prior performance scores, highlighting, at least in this case, the lack of any negative selection effects around WFH. Figure 3 plots the take-up rates of WFH on Wednesday and Fridayby volunteer and non-volunteer groups. We see a fewnotable facts. First, take-up overall was about 55% for volunteers and 40% for non-volunteers, indicating that both groups tended to WFH only oneday, typically Friday, each week. At Trip.com, large meetings and product launches often happen mid-week, so Fridays areseen as a better day towork from home. Second, the take-up rate even for non-volunteers was 40%, indicating that Trip.com's suspicion that many employees did not volunteer out of fear of negative signaling was well-founded, and highlights how amenities like work-from-home, holiday, maternity, or paternity leave may need to be mandatory to ensure reasonable take-up rates. Third, take-up surged on Fridays before major holidays. Many employees returned to their hometowns,6 using their WFH day to travel home on the quieter Thursdayevening or Friday morning. Finally, take-up rates jumped for both treatment and control 6Most Shanghai employees who arenot local are from neighboring provinces of Jiangsu and Zhejiang. And the Shanghai headquarters is right next to the bullet-train station. So, some employees stopped renting expensive Shanghai apartments and just bookedhotels nearby for three nights per week.
  10. 9employees in late January 2022 after a COVID case in the Shanghai building. Trip.com allowed all employees at that point to WFH, so the experiment effectively ended early on Friday, 21stJanuary. Looking at take-up rates in Table 2 we see that, non-managers, employees with a longer commute and employees with children under 18 are more likely to opt to WFH oneligible days. In contrast,take-up is, perhaps surprisingly, flat across gender, ageand prior performance. Table 3 shows striking evidence of heavy coordination across employees within teams on work-from-home days. In column (1), we regress whether a treated individual works remotely on a Wednesday or Friday conditional on the share of their teammembers working remotelyon that day, finding a large, highly significant coefficient of 0.551. This indicates if an individual's whole team works from home on a particular day, they are 55% more like to do the same than if the entire team comes to work. In columns (2) and (3) we control for date and desk fixedeffects to check if this correlation is being driven by individual events (like holidayswhen everyone works from home) or persistent differences across teams, finding very similar results. So, this reveals remarkably strong coordination by teams towards working from home on the same days. This matches the survey results from the US that employees'primary motivation for coming to work is to spend time with colleagues and co-workers (e.g., Appendix figure A6), suggesting hybrid works best when office and home days are coordinated. Table 3 column (4) adds an indicator for whether the team-manager is WFH on the same day and finds this is significant, although about 20% of the magnitude of the team coefficient (and adding it has little impact on the team coefficient). This suggests managerial presence is secondary in importance to team-mate presence. In column (5) we see that coordination is strongest in teams with a high share of connected individuals, proxied by having exchanged messages on at 5+ days in the 3-months prior to the experiment.7In columns (6) and (7) we include proxies forfriends within the team – the share of team-members in the same intake cohort or same age8 – and find no effect. Together with column (5) this suggests employees are potentiallycoordinating to be in the office together with team members when they are working closely with them, rather than mostly for social reasons. 7As explained in greater detail below this counts messages in a firm work messaging system which is not typically used for personal or social messaging.8In Trip.com it is typical for employees to have friends from their intake cohort since these groups underwent early training sessions together. Similar age is a weaker proxy for possible friendship.
  11. 10III Employee Attrition and Job SatisfactionPerhaps the most important result for Trip.com was the substantial reduction in attrition rates seen in the treatment employees. As seen in Table 4, the attrition rate over the 6-month experimental period for the treatment group was 4.7% versus 7.2% for the control group, a 33% reduction in attrition. This was matched by a significant increase in survey measures of job recommendation, work satisfaction, life satisfaction and work-life balance in treatment employees (Appendix Table A10). Interestingly, this impact of hybrid on attrition rates varied heavily across employees as Table 4 and Figure 4 explores.The most striking difference in Figure 4 is between non-managers whose treatment quit rate was 5.3% compared to 8.6% for the control group (a 38% reduction) and managers whose treatment quit was 3.1% compared to 2%for the control group (a 55% increase). This result matches a sentiment often reported in the media that managing employees remotely is hard, so managers are less enthusiastic about remote work than employees. We see in the other three panels of Figure 4 that female employees, employees with longer commutes, and less tenured employees (those with no more than 36 months) are significantly more likely to have lower attrition rates in the WFH treatment group.The commuting result is perhaps the most intuitive as the most frequency cited benefit of working from home is avoiding the commute9. The link with tenure arises because more recently hired employees have weaker social ties with co-workers so have a relatively lower value from coming to the office.Finally, the negative impact on female quit rates from hybrid is interesting and echoes papers like Emanuel et al. (2022) and Goldin (2022) emphasizing a potentially greater valuation on remote work by women. Women are less likely, however, to volunteer toWFH as noted in Table 1 and show no difference in take-up rates in Table 2, possibly suggesting some gender differences in concerns over the costs of actively signaling a preference for remote work. 9For example,Appendix Figure A7
  12. 11IV Working PatternsIV.A Working Hours The experiment also changed working hours and patterns. US surveys find the second largest benefit of remote work is increasedworkscheduleflexibility (with reduced commuting the first benefit). Consistent with this, treatment employees from Trip.comreported an improved ability to flex their hours when working from home, such as visiting the dentist, spending time with their children, taking a home delivery, or performing chores. This greater flexibility when working from home shows up in Table 5 in terms of time-shifting from WFH days to other days. In the top panel, we see that treatment employees have a significant increase in their Virtual Private Network (VPN) time, which is required by employees to access company servers from home. This occursboth on their WFH days of Wednesday and Friday and on all other weekdays and the weekend. This suggests treatment employees increased working time from home even on office days and weekends. On the lower panel, we look at time in the office and notice this falls on Wednesdays and Fridays for treatment employees since they frequently work from home. This also falls somewhat on Thursdays when they are slightly more likely to take the day off, but it is unchanged on all other days. Figure 5 left-panel shows this increase in VPN use on office days and weekends, while the right-panel shows the increase in messaging outside of core working hours. These suggest that WFH spreads work outside of the core 9am-5pm Monday to Friday period into evenings and weekends, similar to results from McDermott and Hansen (2021) showing an increase in GitHub event activity in afterwork hours and weekends during the pandemic. In quantitative terms we see that on Wednesday and Friday, treatment employees reduced their office time bya total of 6.4 hours (3.071+3.296) but increased their VPN time by a total of 3.4 hours (1.656+1.745). Applying the 0.75 ratio of VPNtime/working timeimplies a total increase at the home of 4.5 hours from this additional VPN time.10 So WFH employees work about 1.9 hours (6.4-4.5) less on Wednesday and Friday. In contrast, on all other days, they have a net reduction inofficetime of 0.256 hours and an increase in VPN time of 0.379 hours, which applying the VPN/working time ratio implies a netincrease of 0.25 hours of total working time. Given treatment employees typically WFH one day a 10We surveyed 107 employees about VPN use and found this is used on average 75% of the time when working from home. So,1 hour of VPN time at home implies 1/0.75 hours (1 hour and 20 minutes) of additional working time.
We use cookies to provide, improve, protect and promote our services. Visit our Privacy Policy and Privacy Policy FAQs to learn more. You can manage your personal preferences, including your ‘Do not sell or share my personal data to third parties’ setting using the “Customize cookies” button below.