Work-from-home productivity during the COVID-19 pandemic: Evidence from Japan.

This study examines the productivity of working from home (WFH) practices during the COVID-19 pandemic. The results reveal that the mean WFH productivity relative to working at the usual workplace was about 60%-70%, and it was lower for employees and firms that started WFH practice only after the spread of the COVID-19 pandemic. However, there was a large dispersion of WFH productivity, both by individual and firm characteristics. Highly educated and high-wage employees tended to exhibit a small reduction in WFH productivity. The results obtained from the employee and employer surveys were generally consistent with each other.

Basic Survey of Japanese Business Structure and Activities

Ministry of Economy, Trade and Industry

Occupational Information Network

Research Institute of Economy, Trade, and Industry

Survey of Corporate Management and Economic Policy

total factor productivity

working from home

INTRODUCTION

Following the spread of the COVID‐19 pandemic, the practice of working from home (WFH) has been increasingly adopted in major advanced countries. During normal times, the percentage of workers participating in WFH practice was approximately 10% or less in major advanced countries, but the number of workers who frequently or occasionally conduct their jobs at home has increased suddenly since March 2020 (e.g., Adams et al., 2020; Bartik et al., 2020; Bick et al., 2020; Brynjolfsson et al., 2020; Buchheim et al., 2020; Okubo, 2020). In Japan, although teleworking, including WFH, has been promoted by the government as part of the “Work‐Style Reform” in recent years, the share of WFH workers was only about 5% in 2017 (Morikawa, 2018). However, many firms introduced WFH practices to prevent COVID‐19 infection. In particular, the number of workers engaged in WFH increased further following the first declaration of a state of emergency by the Japanese government in April 2020.

Since the onset of the COVID‐19 pandemic, epidemiological models that augment economic behavior have been developed, and simulation analyses of the effects of social distancing measures, such as a shelter‐in‐place order and mandatory shutdown of service industries and schools to suppress COVID‐19 infections, have been conducted in many countries (e.g., Atkeson, 2020; Eichenbaum et al., 2021; Jones et al., 2021). These studies generally indicate that stringent social distancing policies are effective in mitigating the spread of the pandemic (i.e., “flattening the curve”), but they have large negative effects on economic activity, that is, there is a trade‐off, at least in the short run, between public health and the severity of the recession. Some of the simulation models explicitly consider WFH practices (e.g., Akbarpour et al., 2020; Aum et al., 2021; Bodenstein et al., 2020; Brotherhood et al., 2020; Fujii & Nakata, 2021; Jones et al., 2021) because the feasibility of WFH practice can mitigate the trade‐off between health and economic activity arising from social distancing policies.

Along with the accumulation of actual data on the number of COVID‐19 infections and deaths, empirical evaluations of the effects of WFH have also been conducted (e.g., Adams‐Prassl et al., 2020; Alipour et al., 2021; Béland et al., 2020a, 2020b; Fadinger & Schymik, 2020; Lin & Meissner, 2020; Mongey et al., 2021). These ex‐post analyses generally confirm that WFH suppresses the spread of the pandemic and/or lessens the negative effect of the pandemic on production and employment.

However, not only the feasibility of WFH, but also its effect on productivity relative to working at the usual workplace affects the efficacy of WFH in mitigating the negative influence of social distancing policies on the economy. In simulation studies, the percentage of jobs that can be performed at home is often taken from task‐based estimates, such as in Dingel and Neiman (2020). In contrast, because estimates of WFH productivity are scarce, simulation studies have assumed arbitrary figures of WFH productivity (e.g., 50% or 70% relative to working at the workplace). To supplement the paucity of studies on WFH practices brought about by COVID‐19, this study presents quantitative evidence on the prevalence, intensity, and productivity of WFH based on original surveys of employees and firms in Japan during the COVID‐19 pandemic.

Our analysis of the survey results revealed that, for a large majority of employees and firms in Japan, productivity at home was lower than that at the business premises. The mean WFH productivity relative to working at the usual workplace was about 60%–70%, and it was lower for employees who started WFH only after the spread of the COVID‐19 pandemic.

The remainder of this paper is organized as follows. Section 2 presents a brief review of the related literature and describes the contributions of this study. Section 3 explains the design of employee survey and reports on the prevalence, intensity, and productivity of WFH practice during the COVID‐19 pandemic, with a focus on the differences in individual characteristics. Section 4 explains the design and results of the firm survey on WFH and how it relates to firm characteristics. Section 5 provides the conclusions and discusses implications.

LITERATURE REVIEW

Following the spread of the COVID‐19 pandemic, estimations have been presented on how many jobs can potentially be performed at home (e.g., Adams et al., 2020; Boeri et al., 2020; Brussevich et al., 2020; Dingel & Neiman, 2020). Using data on task contents of occupations taken from the Occupational Information Network (O*NET), Dingel and Neiman (2020), an early representative study, estimated that 34% of jobs in the United States can plausibly be performed at home. Boeri et al. (2020) indicate that between 23% and 32% of jobs can potentially be carried out at home in major European countries. Adams et al. (2020), using unique surveys from the United States and the United Kingdom, document the percentage of tasks (on a scale of 0%–100%) that workers can do from home, which differs from estimates on the percentage of jobs achievable at home. Although the share of tasks that can be done from home varies considerably across, as well as within, occupations and industries, the mean figures are around 40% in both countries. The results suggest that some tasks must be performed at the business premises, even for workers whose jobs can mostly be conducted at home.

More recently, using individual‐level survey data, several studies have reported results on the percentage of workers who engage in WFH practices during the COVID‐19 pandemic (e.g., Barrero et al., 2021; Bick et al., 2020; Brynjolfsson et al., 2020; Buchheim et al., 2020). These studies show that between 35% and 50% of workers engage in WFH in the United States and some major European countries. Certain studies report results from firm surveys that about half of the firms introduced WFH practice in April 2020 (e.g., Bartik et al., 2020; Buchheim et al., 2020).

Overall, quantitative evidence on the potential and actual percentages of WFH practices has been accumulating rapidly. In contrast, evidence on the productivity of employees who practice WFH during the COVID‐19 pandemic has been limited. In this respect, Dingel and Neiman (2020) note that it is not straightforward to use the percentage of jobs plausibly performed at home to estimate the share of output that would be produced under stringent social distancing policies, because an individual worker's productivity may differ considerably when working at home against working at the usual workplace.

At a normal time before the COVID‐19 pandemic, Bloom et al. (2015) present evidence from a field experiment with call center employees in China. They show that WFH practices enhanced the total factor productivity (TFP) of the organization. The positive effect on productivity arises from both improvements in individual workers' performance and from reductions in office space. In contrast, Battiston et al. (2021), exploiting a natural experiment with a public sector organization in the United Kingdom in charge of answering emergency calls, find that productivity is higher when teammates are in the same room, and that the effect is stronger for urgent and complex tasks. They suggested that teleworking is unsuitable for tasks requiring face‐to‐face communication. Dutcher (2012), based on a laboratory experimental approach, indicates that telecommuting may have a positive effect on employee productivity for creative tasks but a negative impact on dull tasks.

These studies indicate that employee productivity under WFH practices depends on the characteristics of the occupations and the specific tasks undertaken. The recent increase in WFH practices has been widespread, involving a variety of white‐collar workers, but causal evidence of the productivity of ordinary office workers under WFH has been scant. As the recent surge in WFH practices brought about by the COVID‐19 pandemic can be considered a natural experiment, we can observe useful evidence of employee productivity under WFH practices through appropriately designed surveys.

After the onset of the COVID‐19 pandemic, a small number of studies have presented evidence on WFH productivity based on individual or firm surveys. Etheridge et al. (2020), whose study is based on a survey of individuals in the United Kingdom, show that, on average, WFH productivity is not significantly different from that of workplace productivity, but it varies depending on individuals' socio‐economic status, industry, and occupation. Barrero et al. (2021), based on a survey of individuals in the United States, indicate that the majority of respondents who have adopted WFH practice report higher WFH productivity than their expectations before the start of the pandemic. Empirical studies investigating the productivity of WFH under COVID‐19 from the employer side have been rare. An exception is Bartik et al. (2020), who use data from a survey of small‐ and medium‐sized firms in the United States from March to April 2020, reported a decrease in productivity of about 20% on average. However, as the authors stated, the results reflect the self‐selection of workers into WFH.

To supplement the paucity of evidence on WFH productivity, based on an originally designed survey for individuals conducted in June 2020, as well as a firm survey from August to September 2020, this study presents novel observations about the prevalence, intensity, and productivity of the WFH practice in Japan. As the quantitative evidence on WFH productivity has been limited, this study contributes to the literature and policymaking for tackling the negative effects of the COVID‐19 pandemic on the economy. Importantly, this study presents complementary evidence from both the employee and employer sides. It is challenging to measure the productivity of individual workers, particularly white‐collar workers. This study measures WFH productivity as a percentage of an employee's productivity at home relative to his/her own productivity at the usual workplace, thus avoiding a comparison of productivity against other workers.

EMPLOYEE SURVEY

Survey design

The employee survey data are retrieved from the “Follow‐up Survey of Life and Consumption under the Changing Economic Structure” designed by the author of this paper and conducted by the Research Institute of Economy, Trade, and Industry (RIETI) in late June 2020. The online survey questionnaire was sent via e‐mail to 10,041 individuals who responded to a survey conducted in 2017. In the 2017 survey, the sample individuals were randomly chosen from the 2.3 million registered monitors of Rakuten Insight, Inc., stratified by gender, age (from 20 to 79 years), and region (prefecture), in proportion to the population composition of the 2015 Population Census (Statistics Bureau, Ministry of Internal Affairs and Communications).

There were 5105 respondents (50.8% response rate) to the 2020 survey. This study mainly used a sample of 3324 individuals who were working at the time of the survey. The analyses in this study are based on cross‐sectional information obtained from the 2020 survey, but data from the 2017 survey are also used when necessary. For example, the educational attainment of individuals is taken from the 2017 survey.

The major questions regarding WFH included (1) whether an employee participated in WFH practice and the timing when the WFH started, (2) frequency of WFH, and (3) subjective productivity under WFH conditions. In addition, the survey collected information about various individual characteristics, such as gender, age, and prefecture of residence. Those who were working also provided information on the type of employment (nine categories), occupation (13 categories), industry (14 categories), firm size (13 categories), weekly working hours (eight categories), annual earnings (tax inclusive; 18 categories), prefecture of usual workplace, and commuting hours (round trip; 10 categories). These items were in the form of multiple‐choice questions and were generally consistent with those in the Employment Status Survey (Ministry of Internal Affairs and Communications). In the analysis presented later, some categories are integrated into smaller number of classifications. For example, the type of employment is integrated into regular (permanent full‐time) and non‐regular (part‐time, contractual, and temporary) employees.

The specific question regarding WFH practice was “Did you practice WFH after the onset of the COVID‐19 pandemic and the stay‐at‐home request from the government?” The choices were: (1) “I have been practicing WFH since before the COVID‐19 pandemic,” (2) “I have started practicing WFH after the onset of the COVID‐19 pandemic,” and (3) “I have not practiced WFH.” Next, for those who chose (1) or (2), the survey asked the frequency of WFH: “How many workdays did you spend WFH when the frequency of your WFH days was the highest?” This question requires a specific share of the WFH. For example, for a worker who spent three days in a week WFH (assuming a 5‐day work week), the response is 0.6.

Regarding WFH productivity, which is the focus of this study, the question was “Suppose your productivity at the usual workplace is 100, how do you evaluate your work productivity at home? Please answer this question by considering all of your tasks.” For this question, it was noted that “If your productivity at home is higher than that at the usual workplace, please answer with a figure higher than 100.” In fact, some respondents reported figures higher than 100. Because this productivity measure is subjective, some measurement errors of true productivity are unavoidable. However, it should be stressed that an employee's productivity under WFH conditions was asked as a relative measure against his/her own productivity at the business premises, not as a comparison with his/her colleagues.

In the following, we present the cross‐tabulation and simple regression results of the answers to the questions explained above.

Prevalence and frequency of WFH

Among the 5105 survey respondents, 3324 were working at the time of the survey. This subsection describes the prevalence and frequency of WFH practices in this sample. Table 1 shows the tabulated results on the prevalence of WFH during the COVID‐19 pandemic. About 35.8% (column (1)) of all workers participated in the WFH arrangement, of which 10.6% had been under such an arrangement before the COVID‐19 pandemic (hereinafter “early WFH adopters”), and 25.3% started the practice after the onset of the COVID‐19 pandemic (hereinafter “new WFH adopters”). When the “state of emergency” was declared, the Japanese government repeatedly requested firms and workers to implement WFH as much as possible. We conjecture that majority of the new WFH adopters were suddenly obliged to start WFH due to the infection risk of the pandemic and the request from the government. However, we cannot eliminate the possibility that some of them would have adopted WFH even without the COVID‐19 incidence.

TABLE 1

Prevalence of WFH practice: Tabulation from the employee survey

	(1) All workers	(2) Employees
Doing WFH	35.8%	32.2%
Early WFH adopters	10.6%	4.3%
New WFH adopters	25.3%	27.9%
Not doing WFH	64.2%	67.8%
Observations	3324	2718

Note: The percentages in column (2) are calculated after excluding “company executive,” “self‐employed,” and “family worker” from all workers.

The figures presented above include self‐employed and family workers who usually conduct businesses at home. When limiting the sample to employees (2718 people), the corresponding percentages are 32.2%, 4.3%, and 27.9%, respectively (column (2)). It is obvious that a large number of employees started WFH after the onset of the COVID‐19 pandemic. The percentage of WFH adopters is somewhat smaller than the comparable figures for the United States and major European countries, as referred to in the previous section.

We estimated a linear probability model to investigate the employee characteristics correlated with the WFH adoption. The binary dependent variable is whether WFH is adopted, and the explanatory variables are gender (female dummy), age category dummies, education dummies, annual earnings (expressed in logarithm), commuting hours (expressed in logarithm), employment type (non‐regular employee dummy), industry dummies, occupation dummies, and firm size dummies. The reference categories for the dummy variables are male, age 40–49, high school education, regular employee, manufacturing industry, clerical job, and firm size of 100–299 employees. The central values of the earnings categories were applied as a logarithmic transformation to construct the variable of annual earnings. In this calculation, “less than 500 thousand yen” and “20 million yen or more” were treated as 250 thousand yen and 21.25 million yen, respectively. A similar logarithmic transformation was applied to the variable of commuting hours. In this calculation, “four hours or longer” was treated as 4.25 h.

The results are presented in Table 2. The coefficients for the age categories of 20–29 and 30–39 years, university and postgraduate education, annual earnings, commuting hours, information and communications industry, administrative and trade‐related occupations, and firm size of 1000 or more employees are positive and significant, meaning that these characteristics are associated with a higher probability of participating in WFH practice after controlling for other observable characteristics. Meanwhile, the coefficients for the transport industry, healthcare and welfare industry, sales occupation, and production‐related occupation are negative and significant. Interestingly, the coefficients for female and non‐regular employees are insignificant, which differs from the observations through simple cross‐tabulation. The results suggest that female and non‐regular employees tend to work in industries and occupations where WFH is difficult.

TABLE 2

Probability of participating in WFH practice: Estimation result of the employee survey

Variables	Coef.	Robust std. err.
Female	−0.005	(0.020)
20–29	0.107	(0.038)	***
30–39	0.054	(0.024)	**
50–59	0.028	(0.022)
60–69	0.036	(0.023)
70–79	0.072	(0.044)	*
Vocational school	0.018	(0.026)
Junior (2‐year) college	0.037	(0.025)
4‐year university	0.093	(0.020)	***
Graduate school	0.233	(0.041)	***
Ln earnings	0.071	(0.012)	***
Ln commuting hours	0.095	(0.010)	***
Non‐regular employee	0.020	(0.023)
Agriculture	−0.036	(0.078)
Construction	0.020	(0.041)
Information & communications	0.277	(0.044)	***
Transport	−0.145	(0.035)	***
Wholesale & retail	−0.031	(0.037)
Finance & insurance	0.078	(0.049)
Real estate	0.036	(0.070)
Accommodations & restaurants	−0.082	(0.048)	*
Health care & welfare	−0.213	(0.028)	***
Education	0.071	(0.042)	*
Other services	−0.019	(0.030)
Public services	0.043	(0.047)
Other industries	0.081	(0.033)	**
Administrative & managerial	0.075	(0.036)	**
Professional & engineering	0.006	(0.029)
Sales	−0.133	(0.041)	***
Trade‐related	0.130	(0.041)	***
Service	−0.058	(0.030)	*
Production & other	−0.130	(0.025)	***
99 or smaller	−0.009	(0.023)
300–499	−0.008	(0.036)
500–999	0.057	(0.035)
1000 or larger	0.087	(0.026)	***
Government	0.001	(0.051)
Observations	2656
R‐squared	.2815

Note: Linear probability model estimation with robust standard errors are in parentheses. ***p < .01, **p < .05, *p < .1. The dependent variable is the adoption of WFH. The categories used as references are male, age 40–49, junior/senior high school, regular employee, manufacturing industry, clerical job, and firm size of 100–299 employees.

Table 3 shows the tabulated results of the frequency of WFH among employees who participated in WFH practice. The figures in the first column are the ratio of WFH days to weekly workdays when the WFH frequency was the highest. About 20.4% of these employees did their jobs completely at home. The mean and median frequency of WFH were 0.557 and 0.5, respectively. In other words, in the case of a 5‐day workweek, typical WFH workers spend 2 to 3 days a week at home, but the frequency of WFH is highly dispersed. Overall, “hybrid mode”—a combination of WFH and office work—was prevalent in Japan during the COVID‐19 pandemic. Traditionally, Japanese firms tend to adopt a mechanism of horizontal coordination through informal information sharing (see, e.g., Aoki, 1986, 1990). In relation to this practice, job descriptions of employees are often not explicitly documented. Although the internal organizations and operations of Japanese firms have been changing gradually, the prevalence of hybrid mode of WFH may reflects the characteristics of Japanese firms to maintain the role of face‐to‐face communication at the workplace even under the COVID‐19 pandemic.

TABLE 3

Distribution of the frequency of WFH: Tabulation from the employee survey

WFH frequency	%
0.1	13.8
0.2	11.1
0.3	8.7
0.4	7.2
0.5	14.3
0.6	4.0
0.7	4.8
0.8	8.9
0.9	6.8
1.0	20.4
Observations	876

Note: Frequency of WFH among employees who participated in WFH practice.

While not reported in the table, the difference of WFH frequency by the timing of WFH initiation is small. The mean frequency of WFH for those who engaged in WFH practice before the COVID‐19 pandemic (i.e., early WFH adopters) was 0.592, and that for those who started WFH after the onset of COVID‐19 (new WFH adopters) was 0.551 (median figures are 0.55 and 0.5, respectively). According to a survey conducted in 2017, the majority of teleworkers spend only one day or less per week (Morikawa, 2018), meaning that the frequency of WFH increased substantially after the COVID‐19 pandemic, even for early WFH adopters.

We can calculate the individual‐level WFH hours by multiplying the usual weekly working hours by the frequency of WFH. The aggregated share of WFH hours (“WFH intensity”) can be calculated as the sum of the WFH hours divided by the sum of the weekly working hours of all employees. The resulting aggregate share of WFH is 19.4%, which is slightly less than one‐fifth of work conducted at home by the all employees in our sample. The remaining 80.6% of the work is conducted in the usual workplace. Although the number of workers engaged in WFH dramatically increased after the COVID‐19 pandemic, the macroeconomic contribution of WFH labor input was not large, because many jobs cannot be done at home and the number of full‐time WFH workers was limited.

It is expected that WFH will contribute to mitigating the congestion of public transport. Using data on commuting hours, we can calculate the reduction in aggregate commuting hours attributable to WFH, which is estimated to be 24.5%. Because both the probability and frequency of WFH are higher among long commuters, the contribution of WFH to the saving of commuting hours is larger than its share of total working hours. This calculation suggests that WFH has contributed to reducing the risk of infection arising from physical contact among commuters.

The ratios of saved commuting hours to total working hours are approximately 9% for those engaged in WFH and 3% for all employees. This calculation indicates that even if some portion of saved commuting hours is allocated to additional work, the contribution to labor supply is not quantitatively large at the aggregate level.

Productivity of WFH

The distribution of WFH adopters' subjective productivity at home relative to their usual workplace (= 100%) is summarized in Table 4. The mean and median of this measure of WFH productivity were 60.6% and 70%, respectively. However, this WFH productivity measure is very heterogeneous: the standard deviation is 35.1% and the gap between the 75th and 25th percentiles is 56.5%. The percentages of WFH adopters whose productivity at home is higher than, equal to, and lower than the productivity at the workplace are 3.9%, 14.2%, and 82.0%, respectively. For a large majority of employees, productivity at home is lower than productivity at the business premises.

TABLE 4

WFH productivity: Tabulation from the employee survey

	Mean	Std. dev.	p25	p50	p75	N	Home < office
All WFH employees	60.6	35.1	30	70	86.5	876	82.0%
Early WFH adopters	76.8	35.5	70	85	100	118	62.7%
New WFH adopters	58.1	34.4	30	60	80	758	85.0%

Note: WFH adopters' subjective productivity at home relative to their usual workplace (=100). Early WFH adopters are those who started WFH before the COVID‐19 pandemic. New WFH adopters are those who started WFH after the onset of the COVID‐19 pandemic. The last column indicates the percentage of employees working from home whose productivity at home is less than 100.

However, the WFH productivity is very different for the subsamples of early and new adopters. The mean of early adopters is 76.8%, which is 18.7 percentage points higher than that of new adopters (58.1%), and the difference is statistically significant at the 1% level. We conjecture that the higher WFH productivity of early adopters reflects both the selection mechanism and the learning effect. It is conceivable that early WFH adopters voluntarily self‐selected into a WFH arrangement because their jobs are easy to do at home and their working environment at home is not inferior to the workplace. In addition, the accumulation of WFH experience may have improved productivity at home. However, it should be noted that even for early adopters, their subjective productivity at home is, on average, lower than their productivity at the business premises. The percentage of those exhibiting higher or equal WFH productivity relative to the workplace is only about a third, even for the subsample of early adopters. The results suggest that the WFH productivity of new adopters improves through the effect of learning‐by‐experience, but we conjecture that their long‐run WFH productivity will converge to about 70%–80% of their productivity at the workplace.

As stated before, the WFH intensity, that is, the share of WFH hours to total labor input, is about 19.4%, and the contribution of WFH to total earnings is 24.5%. It is possible to make a rough estimate of the loss of aggregate labor productivity arising from the WFH as follows:

In this calculation, the productivity loss is assumed to be zero for workers not engaged in WFH. According to this mechanical calculation, the productivity loss is 7.6% at the aggregate level. If we assume that the WFH productivity of new adopters converges with that of early adopters through the learning effect, the loss will be reduced by 1.2 percentage points to 6.4%.

Table 5 reports ordinary least square (OLS) regression results regarding WFH productivity. The explanatory variables are the same as those in the WFH probability estimation (Table 2): gender, age, education, annual earnings (expressed in logarithm), commuting hours (expressed in logarithm), employment type, industry, occupation, and firm size.

TABLE 5

WFH productivity: Estimation results of the employee survey

Variables	(1)			(2)			(3)
	Coef.	Robust std. err.		Coef.	Robust std. err.		Coef.	Robust std. err.
Female	−2.424	(3.462)		−2.701	(3.455)		−4.239	(3.387)
Vocational school	5.899	(5.460)		5.880	(5.367)		5.328	(5.255)
Junior (2‐year) college	13.610	(5.651)	**	14.087	(5.605)	**	13.480	(5.537)	**
4‐year university	13.095	(3.726)	***	12.761	(3.679)	***	12.247	(3.619)	***
Graduate school	18.484	(4.620)	***	18.138	(4.622)	***	16.941	(4.543)	***
Ln earnings	5.557	(2.146)	**	5.534	(2.133)	**	5.326	(2.059)	**
Ln commuting hours	3.001	(1.529)	*	2.873	(1.512)	*	1.945	(1.509)
Non‐regular employee	8.464	(4.284)	**	8.057	(4.283)	*	7.580	(4.229)	*
99 or smaller	−1.132	(3.926)		−1.036	(3.941)		−1.054	(3.811)
300–499	7.241	(5.774)		7.019	(5.812)		7.070	(5.788)
500–999	−2.675	(4.885)		−1.902	(4.918)		−2.392	(4.829)
1000 or larger	−1.393	(3.834)		−1.371	(3.845)		−2.109	(3.734)
Government	−5.039	(6.775)		−5.036	(6.684)		−5.577	(6.716)
New WFH adopter				−14.051	(4.341)	***
WFH frequency							0.174	(0.038)	***
Cons.	18.533	(14.939)		32.782	(15.501)	**	11.630	(14.516)
Age dummies	Yes			Yes			Yes
Industry dummies	Yes			Yes			Yes
Occupation dummies	Yes			Yes			Yes
Observations	828			828			828
R‐squared	.1819			.1963			.2037

Note: OLS estimations with robust standard errors are in parentheses. ***p < .01, **p < .05, *p < .1. The dependent variable is the subjective productivity at home relative to their usual workplace. The categories used as references were male, age 40–49, junior/senior high school, regular employee, manufacturing industry, clerical job, and firm size of 100–299 employees.

According to the baseline regression (column (1)), the coefficients for high education, annual earnings, and commuting hours are positive and significant. Higher education is associated with a higher probability of WFH adoption and higher WFH productivity. Quantitatively, the WFH productivities of those who graduated from a four‐year university and graduate school are approximately 14% and 20% higher than that of high school graduates, respectively. The positive coefficient for annual earnings suggests that human capital other than schooling is also associated with the relatively high productivity of WFH.

Although the significance level is marginal, the positive coefficient for commuting hours suggests long commuters can avoid fatigue from commuting by using WFH. Another possible interpretation is that long commuters may allocate saved commuting hours to additional work. In this case, the measured gain from WFH cannot be interpreted as a pure productivity effect but as an increase in labor supply. We cannot deny the possibility that the respondents may interpret the increased working hours as a productivity gain of WFH. However, as calculated in the previous subsection, the contribution of saved commuting hours to labor supply is not quantitatively large (9% for those engaged in WFH and 3% for all employees), even for those engaged in WFH. Suppose, for example, half of the saved commuting hours are allocated to additional work. In this case, increase in the labor supply of WFH workers is approximately 4.5%. Since the productivity loss of WFH workers is 17.3%, the additional labor supply only partially offset the reduced productivity. The possible bias of reported productivity of WFH arising from saved commuting hours will be positive (if respondents answered worker‐day productivity) or negative (if respondents answered worker‐hour productivity). In any case, the main conclusion―the mean WFH productivity relative to working at the business premises is low―is unaltered.

Unexpectedly, the coefficient for non‐regular employees is positive and significant at the 5% level. The size of the estimated coefficient means that among those who practice WFH, the productivity relative to the workplace of non‐regular employees is about 8% higher than that of regular employees. Our interpretation is that the job descriptions of non‐regular employees, such as part‐time, contract, and temporary workers, is clear, and they are less likely to bear the burden of sudden unexpected tasks and coordinating roles in the workplace.

Column (2) of Table 5 shows the results of using the new WFH adopter dummy as an additional explanatory variable. As expected, the coefficient for this dummy is negative, large, and highly significant. After controlling for the other observable individual characteristics, the WFH productivity of new adopters is 14.1% lower than that of early adopters, although the gap is smaller than the raw comparison (18.7%). Column (3) of the table shows the estimation results when the frequency of WFH is added as an explanatory variable. The estimated coefficient for this variable is positive and highly significant. Quantitatively, the relative WFH productivity of employees with one more day of WFH per week is approximately 3.5 percentage points higher. This result suggests that employees with relatively high WFH productivity tend to practice WFH frequently.

The survey asked about factors that adversely affect WFH productivity. There were eight choices for this question. The results are summarized in Table 6. The major reasons for reduced productivity at home are: (1) loss of quick communication that is only possible through face‐to‐face interactions with their colleagues at the workplace, (2) poor telecommunication environment at home relative to the workplace, (3) rules and regulations that require some tasks to be conducted in the office, and (4) some tasks cannot be conducted at home, even though these are not required by rules and regulations.

TABLE 6

Factors affecting WFH productivity: Tabulation from the employee survey

	Factors reducing productivity at home	%
1	Loss of quick communication that is only possible through face‐to‐face interactions with their colleagues at the workplace	38.5
2	Poor telecommunication environment at home relative to the workplace	34.9
3	The requirements by rules and regulations that some tasks must be conducted in the office	33.1
4	Some tasks cannot be conducted at home even though these are not required by rules and regulations	32.5
5	It is difficult to concentrate on job because of the presence of family members	19.9
6	Lack of pressure from the boss, colleagues, and subordinates	19.3
7	Lack of a private room specifically designed for work	15.1
8	Other reasons	10.2

Note: Multiple answers were allowed for this question.

Among these obstacles, the telecommunication environment at home can be improved through investments in hardware and software, whereas inappropriate rules and regulations can be amended to some extent. However, the loss of face‐to‐face interactions is an inherent constraint on WFH productivity. Some studies in the United States report higher WFH productivity (e.g., Barrero et al., 2021), which is different from our findings for Japan. Relatively small living space in Japan and differences in rules and regulations between the two countries may explain the gap to some extent. However, we conjecture that the importance of face‐to‐face communication at the workplace in Japanese firms is a major contributor to the gap of WFH productivity, because, as discussed in the previous subsection, the importance of coordination through informal information sharing and the lack of explicit job description have been the characteristics of traditional Japanese firms. However, further study is needed to uncover the exact reasons.

FIRM SURVEY

The firm survey data are retrieved from the Survey of Corporate Management and Economic Policy (SCMEP), which was designed by the author of this paper and conducted by the RIETI from August to September 2020. The survey questionnaire was sent to 2498 Japanese firms that responded to the previous SCMEP in early 2019. The number of firms that responded to the current SCMEP was 1579 (a response rate of about 63%). The SCMEP in 2019 was sent to 15,000 firms that operate in the manufacturing and service industries, which were randomly selected from the registered list of the Basic Survey of Japanese Business Structure and Activities (BSJBSA), an annual statistical survey conducted by the Ministry of Economy, Trade and Industry (METI). The firms registered in the BSJBSA have at least 50 employees and capital of at least 30 million yen among firms with business establishments that belong to manufacturing, wholesale, retail, and service industries. In other words, the SCMEP does not include small firms with fewer than 50 employees or capital of less than 30 million yen.

The distribution of the firms that responded to this survey by industry is as follows: manufacturing (53.5%), information and communications (5.3%), wholesale (17.8%), retail (10.2%), service (9.0%), and other industries (4.2%). According to the latest BSJBSA, the industry composition of 29,780 firms is as follows: manufacturing (43.1%), information and communications (9.1%), wholesale (18.6%), retail (12.1%), services (9.1%), and other industries (3.4%). Therefore, manufacturing firms are somewhat overrepresented, and information and communications firms are somewhat underrepresented in the SCMEP. Regarding firm size (classified by capital over 100 million yen or less), 34.8% are large firms, whereas 65.2% are small‐ and medium‐sized firms. The corresponding figures of the BSJBSA firms are 28.2% and 71.8%.

The following are the main questions regarding WFH: (1) whether the WFH practice has been implemented, (2) the percentage of employees who have used this workstyle (at the maximum time); (3) the mean WFH frequency of teleworkers (at the maximum time); and (4) the mean productivity of WFH relative to the workplace. The specific wording of the questions and choices, as well as the results, are explained in the following subsections. Further, the SCMEP also asked about the industries (six categories) of the firm's main business, number of regular and non‐regular employees, and number of employees by gender.

As mentioned previously, given that the current survey targets firms that responded to the 2019 SCMEP, information on the firm characteristics collected in the 2019 survey is available. In this study, we use the percentage of employees who have a university degree or higher in the analysis. By linking with BSJBSA's data for the fiscal year 2018, information available in the BSJBSA can be used as well. This study uses capital, mean wages (total cash salary divided by the number of employees), and location of firm headquarters (prefectures) from the BSJBSA. The major variables and summary statistics are presented in Table 7.

TABLE 7

Summary statistics of the firm survey

Summary statistics	Nobs.	Mean	Std. dev.
WFH adoption	1579	0.495	0.500
WFH intensity	741	0.299	0.286
WFH intensity (adjusted)	1579	0.140	0.246
WFH productivity	762	68.281	23.440
WFH productivity (adjusted)	1579	32.951	37.814
Number of employees	1561	4.973	0.879
Headquartered in Tokyo	1579	0.168	0.374
Population density (log)	1579	6.573	1.497
Female ratio	1561	0.311	0.196
Non‐regular ratio	1552	0.234	0.240
Ratio of university or higher	1364	0.315	0.246
Mean wages (log)	1514	1.411	0.394

Note: WFH intensity is calculated as the ratio of WFH employees multiplied by WFH frequency per week (expressed as a percentage). “Adjusted” figures are calculated by including WFH non‐adopters, whose WFH intensity and productivity are zero.

Adoption and intensity of WFH

The question regarding whether the WFH practice has been adopted is as follows: “Did your firm implement the WFH practice during the spread of the COVID‐19?” The three options are as follows: (1) “The WFH practice has been adopted before the onset of the COVID‐19 pandemic,” (2) “The WFH practice were introduced after the spread of the COVID‐19,” and (3) “The WFH practice has not been adopted.”

Table 8 presents the tabulation results. The percentage of firms that adopted the WFH system prior to the onset of the COVID‐19 pandemic (“early WFH adopters”) was 4.1%, 45.5% introduced it after the pandemic (“new WFH adopters”), and 50.4% did not. The WFH adoption rate is high in large firms. By industry, the information and communications industry is the highest. A significant difference is observed between firms headquartered in Tokyo and other prefectures.

TABLE 8

Adoption of WFH: Tabulation from the firm survey

	(1) Early WFH adopters	(2) New WFH adopters	(3) Non‐adopters
Total	4.1%	45.5%	50.4%
Large firms	5.3%	57.8%	36.9%
Small & medium firms	3.5%	38.9%	57.5%
Manufacturing	3.0%	42.6%	54.4%
Information & communications	20.5%	75.9%	3.6%
Wholesale	2.1%	57.1%	40.7%
Retail	1.2%	28.6%	70.2%
Services	5.6%	38.0%	56.3%
Other industries	10.6%	51.5%	37.9%
Headquartered in Tokyo	11.1%	73.7%	15.3%
Other prefectures	2.7%	39.9%	57.4%

Note: Early WFH adopters are firms adopting the WFH system before the COVID‐19 pandemic. New WFH adopters are the firms adopted the WFH system after the COVID‐19 pandemic (N = 1574).

Table 9 shows the estimation results of the linear probability model to explain WFH adoption. The larger the firm size (the logarithm of the number of regular employees), the higher is the probability of adopting the WFH system. By industry, firms in the information and communications industry have a high probability, whereas those in retail industry have a low probability of adopting the WFH system. Firms headquartered in Tokyo, those with a large share of female and highly educated employees, and those paying higher wages have a higher probability of adopting the WFH system (column (1)). When the population density of the head office location is used as the explanatory variable, rather than the Tokyo dummy, the coefficient of population density is positive and significant at the 1% level (column (2)). Quantitatively, doubling the population density, the probability of adopting the WFH system is approximately 7% higher. In this specification, no essential difference is observed in the coefficients of the other explanatory variables.

TABLE 9

Determinants of WFH adoption: Estimation results of the firm survey

	(1)			(2)
	Coef.	Robust std. err.		Coef.	Robust std. err.
Number of employees (log)	0.097	(0.016)	***	0.103	(0.016)	***
Information & communications	0.161	(0.051)	***	0.218	(0.048)	***
Wholesale	−0.038	(0.037)		−0.016	(0.037)
Retail	−0.210	(0.043)	***	−0.183	(0.042)	***
Services	−0.040	(0.045)		−0.045	(0.045)
Other industries	0.006	(0.061)		0.016	(0.061)
Tokyo	0.257	(0.032)	***
Population density (log)				0.075	(0.009)	***
Female ratio	0.207	(0.080)	**	0.216	(0.080)	***
Non‐regular ratio	−0.073	(0.063)		−0.114	(0.061)	*
University or higher	0.546	(0.061)	***	0.469	(0.062)	***
Mean wages (log)	0.214	(0.042)	***	0.196	(0.042)	***
Observations	1292			1292
R‐squared	.2657			.2749

Note: Linear probability model estimations with robust standard errors are in parentheses. ***p < .01, **p < .05, *p < .1. The dependent variable is the adoption of WFH practice. Manufacturing is the reference category for industries.

The positive coefficients of Tokyo and population density are consistent with the positive coefficient of commuting hours found in the employee survey (Table 2). We conjecture higher adoption rates for firms located in Tokyo and other populated regions to the higher COVID‐19 infection rates and longer commuting hours in large metropolitan areas.

Even for firms that have adopted the WFH system, not all employees used this workstyle. The share of employees covered by the WFH system may differ by firm. In this regard, the survey asks, “What percentage of your employees used the WFH practice after the spread of COVID‐19?” The tabulation results are presented in column (1) of Table 10. The mean percentage of WFH‐adopting firms was 30.7%. The percentages are 49.1% for early WFH adopters and 29.0% for new WFH adopters. This result shows that firms that are relatively easy to introduce the WFH practice owing to the nature of their businesses adopted it earlier, and a large percentage of employees tend to telework during the COVID‐19 pandemic. By industry, information and communications industry (59.6%) is the highest, and manufacturing industry (18.8%) is the lowest. Large firms have higher percentages than small firms, and firms headquartered in Tokyo have a high share of teleworkers.

TABLE 10

Share, frequency, and intensity of WFH: Tabulation from the firm survey

	(1) Share of employees using WFH	(2) Frequency of WFH per week	(3) WFH intensity (unweighted)	(4) WFH intensity (weighted)
Total	30.7%	3.67	23.3%	10.9%
Early WFH adopters	49.1%	4.54	41.4%	37.6%
New WFH adopters	29.0%	3.59	21.7%	20.7%
Large firms	34.6%	3.61	25.5%	14.4%
Small & medium firms	27.7%	3.71	21.7%	9.1%
Manufacturing	18.8%	3.64	13.6%	6.0%
Information & communications	59.6%	4.28	51.3%	44.6%
Wholesale	38.2%	3.41	27.4%	15.3%
Retail	21.8%	3.39	15.2%	3.9%
Services	41.8%	3.77	32.1%	13.3%
Other industries	49.6%	3.93	42.1%	24.2%
Headquartered in Tokyo	48.4%	3.82	38.2%	30.5%
Other prefectures	23.7%	3.61	17.3%	7.0%
Observations	778	771	741	1579

Note: WFH intensity—contribution of WFH hours to total hours—is calculated as the ratio of WFH employees multiplied by the frequency of WFH per week (expressed as a percentage). Column (4) includes firms that did not adopt WFH, of which the WFH intensity was regarded as zero. Early WFH adopters are the firms adopting the WFH system before the COVID‐19 pandemic. New WFH adopters are the firms adopted the WFH system after the COVID‐19 pandemic.

Even for employees who adopted this workstyle, they were not necessarily full‐time teleworkers (i.e., working at home on all working days) as many of them went to the workplace several days per week. Thus, the survey asks, “What was the average number of WFH days per week for employees who implemented the WFH practice?” The tabulation results are summarized in column (2) of Table 10. The mean frequency of WFH implementation was 3.67 days per week. Assuming that the normal working week is 5 days, teleworkers spend more than 70% of their work hours at home.

Similar to the share of teleworkers, employees of firms that have adopted the WFH system prior to the COVID‐19 pandemic implemented this workstyle frequently, with an average of 4.54 days. Teleworkers of early WFH adopters performed about 90% of their work at home during the peak when the “state of emergency” was declared. Conversely, the average number of firms that introduced it after COVID‐19 was 3.59 days. By industry, information and communications (4.28 days) is the highest, whereas retail (3.39 days) has the lowest frequency. The WFH practice was somewhat more frequently conducted in firms headquartered in Tokyo, but the quantitative difference with firms headquartered in other prefectures is small.

Next, we calculate the WFH intensity at the firm level as the ratio of WFH employees multiplied by the frequency of WFH per week (converted into percentage), which indicates the contribution of WFH hours to the total labor input. As we do not have information on the working hours of individual employees, this calculation assumes that the working hours of all employees are equal. In addition, even on the WFH day, it is possible that he or she may go to the workplace for several hours. Therefore, we should emphasize that the WFH intensity calculated here is only an approximation.

The results are presented in column (3) of Table 10. The mean WFH intensity of firms adopting the WFH system was 23.3%. Even if a firm adopts this workstyle, many employees did not exploit it, and even those who used this work arrangement did not necessarily work at home throughout the week; hence, the mean contribution of the WFH hours to labor input was less than a quarter. Firms that have adopted the WFH system before the COVID‐19 pandemic, large firms, and those headquartered in Tokyo show high WFH intensity. By industry, information and communications industry is the largest at 51.3%, meaning that at the time when the WFH intensity was at its peak, about half of the total labor input was performed at home. Conversely, the manufacturing industry is the smallest at 13.6%, reflecting the difficulty of implementing the WFH system at factory production sites.

The weighted average of the WFH intensity (column (4) of Table 10) is calculated using the number of firms' employees as weight and including the WFH non‐adopters (838 firms), whose WFH intensity is regarded as zero. In other words, it shows the contribution of the WFH system to the total labor input at the aggregate level. The weighted average is 10.9% for all firms that responded to the SCMEP: the contribution of WFH to labor input is surprisingly small, even during the period when WFH peaks. By industry, information and communications is the highest at 44.6%, whereas retail is extremely small at 3.9%. This is an unsurprising result, as more than 70% of firms in retail industry did not implement the WFH system (Table 8).

Table 11 shows the results of the OLS estimations that explain the WFH intensity based on various firm characteristics. In columns (1) and (2), only firms that adopted the WFH system during the COVID‐19 pandemic are used as samples. Information from the 2019 SCMEP is used to calculate the ratio of university graduates or higher, and that from the BSJBSA for the fiscal year 2018 was used to calculate the mean wages. Since the reference category of the industry dummy is manufacturing, the coefficient of each industry indicates a difference from manufacturing.

TABLE 11

Determinants of WFH intensity: Estimation results of the firm survey

	(1)			(2)			(3)			(4)
	Coef.	Robust std. err.		Coef.	Robust std. err.		Coef.	Robust std. err.		Coef.	Robust std. err.
Number of employees (log)	0.000	(0.010)		0.005	(0.010)		0.009	(0.007)		0.013	(0.007)	*
Information & communications	0.243	(0.042)	***	0.278	(0.040)	***	0.262	(0.040)	***	0.295	(0.039)	***
Wholesale	0.034	(0.022)		0.048	(0.022)	**	0.010	(0.013)		0.024	(0.014)	*
Retail	−0.024	(0.035)		−0.027	(0.034)		−0.025	(0.012)	**	−0.013	(0.012)
Services	0.159	(0.039)	***	0.162	(0.037)	***	0.073	(0.022)	***	0.075	(0.021)	***
Other industries	0.194	(0.044)	***	0.206	(0.046)	***	0.133	(0.034)	***	0.144	(0.036)	***
Tokyo	0.150	(0.021)	***				0.164	(0.018)	***
Population density (log)				0.043	(0.006)	***				0.036	(0.004)	***
Female ratio	0.117	(0.060)	*	0.151	(0.059)	**	0.120	(0.032)	***	0.129	(0.032)	***
Non‐regular ratio	−0.114	(0.045)	**	−0.145	(0.044)	***	−0.062	(0.025)	**	−0.086	(0.024)	***
University or higher	0.249	(0.042)	***	0.206	(0.043)	***	0.234	(0.029)	***	0.207	(0.029)	***
Mean wages (log)	0.061	(0.034)	*	0.059	(0.034)	*	0.072	(0.020)	***	0.068	(0.020)	***
Cons.	−0.072	(0.074)		−0.355	(0.080)	***	−0.182	(0.043)	***	−0.395	(0.045)	***
Observations	623			623			1292			1292
R‐squared	.4300			.4232			.4577			.4357

Note: OLS estimations with robust standard errors are in parentheses. ***p < .01, **p < .05, *p < .1. The dependent variable is the WFH intensity (contribution of WFH hours to total hours). Columns (3) and (4) include firms that did not adopt WFH, of which the WFH intensity was regarded as zero. Manufacturing is the reference category for industries.

The estimation results show that information and communications, and service industries have high WFH intensity. The coefficient for Tokyo has a relatively large positive value: the WFH intensity is about 15% higher than that of firms headquartered in other prefectures. The coefficient of female ratio is positive despite the low significance level, and the coefficient of non‐regular ratio is negative and significant at the 5% level. The coefficient of the ratio of employees with university degrees or higher is positive and significant at the 1% level. These results indicate that firms with high female, regular, and educated employees have high WFH intensity after accounting for other firm characteristics. Despite the low significance level, the coefficient of mean wages is positive, suggesting that firms paying higher wages tend to have higher WFH intensity. Column (2) is the result when the prefectural population density is used instead of the Tokyo dummy. The WFH intensity is higher for firms headquartered in densely populated prefectures. Even in this specification, the coefficients of the other variables are generally the same as the result in column (1), except that the coefficient of wholesale industry becomes statistically significant.

In these estimations, firms that did not adopt the WFH system were excluded from the sample. The results of the same specification, including firms that did not adopt WFH, of which the WFH intensity is regarded as zero, are shown in columns (3) and (4) of the table. The results are essentially the same as those in columns (1) and (2).

The question about WFH productivity evaluated from the employer's viewpoint, which is the main interest of this study, is “Suppose that employees' productivity at the workplace is 100, how do you evaluate their productivity at home? Please answer the mean productivity by considering all tasks covered by the WFH system.” The questionnaire noted that “if WFH is more productive than the workplace, please answer with a figure over 100.” Thus, respondents can answer the possibility that WFH is more productive than workplace. In fact, nine firms answered that the productivity of WFH exceeds 100 (the maximum was 120), and 50 firms answered exactly 100.

Table 12 presents the tabulation results. The simple average of firms adopting WFH was 68.3%. According to a survey of employees, the mean subjective productivity of WFH was 60.6%, as reported in the previous section. As the firms surveyed in this study are those with at least 50 employees, we should be careful in simply comparing the figures. However, it is interesting that the figures are not much different from the subjective WFH productivity of employees. If disaggregated by industry, the mean WFH productivity from employees and firms is as follows: manufacturing (employee 70.1, firm 68.0), information and communications industry (73.5, 80.3), wholesale and retail (57.0, 64.5). These results suggest that reporting bias in the employee survey is unlikely to be serious.

TABLE 12

Mean productivity of WFH relative to the productivity at the workplace: Tabulation from the firm survey

	Productivity
Total	68.3
Early WFH adopters	81.8
New WFH adopters	67.0
Large firms	69.4
Small & medium firms	67.4
Manufacturing	68.0
Information & communications	80.3
Wholesale	65.0
Retail	62.6
Services	66.5
Other industries	69.5
Headquartered in Tokyo	72.0
Other prefectures	66.8
Observations	762

Note: Productivity of WFH is relative to productivity at the workplace (=100). Early WFH adopters are the firms adopting the WFH system before the COVID‐19 pandemic. New WFH adopters are the firms adopted the WFH system after the COVID‐19 pandemic.

The mean WFH productivity was 81.8% and 67.0% for the early and new WFH adopters, respectively. The 14.7 percentage points difference is statistically significant at the 1% level. In the employee survey, a large difference (18.7 percentage points) in productivity was observed between those who had adopted the WFH system before the COVID‐19 pandemic and those who adopted it after the spread of COVID‐19. The results of the firm survey are consistent with those of the employee survey. However, even firms that had adopted the WFH system before COVID‐19 expanded their coverage to employees who had not used this workstyle until the spread of the COVID‐19 infection. Therefore, teleworkers in the firm survey did not necessarily experience WFH prior to the onset of the COVID‐19 pandemic.

Almost no difference was found by firm size. By industry, the mean WFH productivity of information and communications industry is the highest at 80.3%, and the figures for the rest of the industries are between 62.6% and 69.5%. The mean figure of firms headquartered in Tokyo is 72% and that in other prefectures is 66.8%, a difference of 5.2 percentage points. As this is only a simple comparison wherein other factors are not controlled for, differences in the composition of industries and employees are reflected. However, as will be described later, even if a regression analysis is conducted, the coefficient of Tokyo is positive and statistically significant. This result is similar to that of the employee survey.

We conjecture that, as a large number of employees working with firms located in Tokyo commute for a long distance, it may be related to the reduction in fatigue caused by long commuting. Another possible reason is that firms' headquarters are concentrated in Tokyo, and jobs and tasks conducted in headquarters are relatively compatible with the WFH practice. Although the survey question asked about the labor productivity of the WFH workers, scaling down expensive office space may potentially contribute to improving TFP (Bloom et al., 2015). However, according to the SCMEP, the percentage of firms that reduced the size of the office was 0.5%. The percentages were higher for firms headquartered in Tokyo (0.9%) than firms in other prefectures (0.4%). This difference may contribute to higher firm‐level TFP in Tokyo, but it is quantitatively limited.

Table 13 shows the OLS estimation results to explain WFH productivity using the same explanatory variables as in the previous regressions. The coefficient of firm size is insignificant, confirming the simple tabulation results reported in Table 12. By industry, the coefficient for the information and communications is positive and that for retail is negative, and they are both highly significant, reflecting the nature of businesses. The coefficient of Tokyo is positive and significant at the 5% level, and the interpretation is the same as stated above. The coefficient of the female ratio is positive and significant at the 5% level. Firms' evaluation of WFH productivity is higher in firms with a large share of female employees after accounting for the other firm characteristics. The coefficient of non‐regular worker ratio is insignificant. Although firms with a large share of non‐regular workers have low WFH intensity, as mentioned previously, no significant difference is observed in terms of productivity. The coefficient of mean wages is positive and significant at the 5% level, indicating that firms paying lower wages tend to be less productive in the WFH system. This is similar to the results of the employee survey.

TABLE 13

Determinants of WFH productivity: Estimation results of the firm survey

	(1)			(2)
	Coef.	Robost std. err.		Coef.	Robust std. err.
Number of employees (log)	1.162	(0.949)		1.314	(0.940)
Information & communications	9.936	(3.490)	***	10.951	(3.437)	***
Wholesale	−5.299	(2.874)	*	−4.949	(2.857)	*
Retail	−14.640	(5.083)	***	−14.843	(5.107)	***
Services	−4.909	(3.617)		−4.994	(3.620)
Other industries	−0.729	(3.880)		−0.503	(3.959)
Tokyo	4.217	(1.876)	**
Density (log)				1.458	(0.626)	**
Female ratio	14.298	(6.518)	**	15.274	(6.491)	**
Non‐regular ratio	1.153	(5.369)		0.397	(5.358)
University or higher	2.107	(4.298)		0.598	(4.356)
Mean wages (log)	7.948	(3.244)	**	7.810	(3.230)	**
Cons.	44.203	(7.764)	***	34.829	(8.517)	***
Observations	615			615
R‐squared	.0714			.0732

Note: OLS estimations with robust standard errors are in parentheses. ***p < .01, **p < .05, *p < .1. The dependent variable is the Productivity of WFH is relative to productivity at the workplace (=100). Manufacturing is the reference category for industries.

The firm survey asked multiple‐choice questions about factors that adversely affect the adoption and productivity of WFH. The choices are generally the same with the employee survey. Table 14 presents the tabulation results. Many firms chose, in descending order, (1) some tasks cannot be conducted at home although these are not required by the rules and regulations, (2) poor telecommunication environment at home relative to the workplace, (3) rules and regulations that require some tasks to be conducted in the office, and (4) loss of immediate communication that is only possible through face‐to‐face interactions with colleagues at the workplace.

TABLE 14

Factors affecting adoption and productivity of WFH: Tabulation from the firm survey

	Negative factors	%
1	Some tasks cannot be conducted at home even though these are not required by rules and regulations	76.1
2	Poor telecommunication environment at home relative to the workplace	60.8
3	The requirements by rules and regulations that some tasks must be conducted in the office	57.7
4	Loss of quick communication that is only possible through face‐to‐face interactions with their colleagues at the workplace	46.0
5	Lack of a private room specifically designed for work	36.9
6	Lack of pressure from the boss, colleagues, and subordinates	36.4
7	Much of the work requires direct interaction with customers	34.3
8	It is difficult to concentrate on job because of the presence of family members	33.0
9	Other reasons	4.1

Note: Multiple answers were allowed for this question.

Although the percentages are different, these four reasons occupy the top positions in both the employee and firm surveys. As written in Section 3, the importance of face‐to‐face communication at the workplace may be an important reason behind the low WFH productivity in Japan. However, differences in rules and regulations, quality of Internet service, and living space may also be related to the gap in WFH productivity with other countries.

In summary, the WFH productivity pattern based on the firm survey is generally consistent with the results of the employee survey presented in the previous section.

CONCLUSION

This study, using unique data from original surveys of employees and employers, presents evidence on the prevalence, intensity, and productivity of WFH practices during the COVID‐19 pandemic in Japan. The findings from the employee and employer surveys are generally consistent with each other. The main results for the period covered in the survey are summarized below.

First, the percentage of employees who practiced WFH was about 32%. The WFH intensity (i.e., the contribution of WFH to total labor input) was about 19% in the employee survey and about 11% in the firm survey.

Second, highly educated, high‐wage, white‐collar employees who work in large firms in metropolitan areas tended to practice WFH, which suggests that infection risk and social distancing policies may exacerbate economic disparity among employees.

Third, the mean WFH productivity was about 60%–70% of the productivity at business premises. Although the WFH practice is attracting attention owing to COVID‐19, employees' productivity was lower than that in the usual workplace, at least on average. WFH productivity was particularly lower for employees and firms that started WFH after the onset of the COVID‐19 pandemic. We conjecture that the likely mechanisms behind the WFH productivity gap between early and new adopters are the selection mechanism and the learning effect.

Fourth, highly educated, high‐wage employees, long‐distance commuters, and those who work in information and communications industry tended to exhibit a relatively small reduction in productivity. Those who worked at home before COVID‐19 tended to practice WFH frequently, suggesting a natural selection of work location based on productivity.

Fifth, the lack of face‐to‐face interactions, poor telecommunication environment at home, and tasks that must be conducted in the office were the major impediments to improving productivity at home. Working practices in Japanese organizations requiring frequent face‐to‐face communication may be an important reason for lower WFH productivity.

Considering the possibility that the WFH system becomes a new workstyle even after this pandemic, finding out the optimal combination of WFH and office work―hybrid mode―may be an important agenda for firms continuing with WFH practices. In this regard, human resource management plays an important role in the selective application of WFH practice to employees who are engaged in tasks suitable for this workstyle. In addition, new communication tools, such as teleconference systems, should be selectively used, depending on their advantages and disadvantages relative to face‐to‐face communications.

It is extremely difficult to accurately measure the productivity of individual workers, particularly that of white‐collar workers. In this study, WFH productivity is expressed as a relative figure to the teleworkers' productivity in the usual workplace, not as a comparison with other workers. However, it should be noted that the productivity measure used in this study depends on subjective reporting; measurement errors are unavoidable.

Although the COVID‐19 pandemic is a natural experiment that exogenously increased the adoption of WFH practice among a wide range of white‐collar workers, it should be noted that, for many service jobs that require physical contact with customers, such as doctors, nurses, and hairdressers, the productivity of teleworking is prohibitively low.

Supporting Information S1

Click here for additional data file.