Working from home vs working from office in terms of job performance during the COVID‐19 pandemic crisis: evidence from China

Abstract

Despite being a worldwide disaster, the COVID‐19 pandemic has also provided an opportunity for renewed discussion about the way we work. By contextualizing in the early periods of China's ending of lockdown policy on COVID‐19, this paper offers evidence to respond to an essential discussion in the field of working from home (WFH): In terms of job performance, can WFH replace working from the office (WFO)? The present study compares job performance in terms of quality and productivity between WFH and WFO from 861 Chinese respondents using entropy balance matching, a quasi‐experimental methodology. Results reveal that WFH enhances job performance in terms of job quality but lowers it in terms of job productivity. In addition, the present study aims to capture and empirically measure the variations in fundamental job characteristics in terms of job control and job demand between WFH and WFO by applying the job demand control support model. More specifically, we find that job control items, such as ‘talking right’ and ‘work rate’, and job demand items, such as ‘a long time of intense concentration’ and ‘hecticness of the job’, are vital factors that contribute to how these differences exert influence on employees' performance in the context of the pandemic.

Key points.

1. WFH is positively related to job quality but negatively related to job productivity.

2. WFH affects job performance via job demand and job control.

3. Social support contributes to job productivity when working from home.

Introduction

The outbreak of the COVID‐19 pandemic boosted an unprecedentedly massive and rapid shift of people's work routines (Bartram and Cooke 2022; Yan et al. 2021). To a large extent, millions of employees around the world have been forced to resort to remote work (Bouziri et al. 2020; Hurley and Popescu 2021; Rogers 2021; Woods and Miklencicova 2021), which leads to the most significant social experiment of ‘working from home (WFH)’ emerging in human history (Zhang, Yu and Marin 2021). According to a report in LinkedIn, as Asia‐Pacific responded to the crisis, organizations in China, Australia, India and Singapore, quickly adapted to support a remote workforce. WFH differs considerably from working from office (WFO) in terms of job attributes and work environment. WFO is characterized by a relatively high degree of formalization and a fixed working routine, including place, time, and task arrangements (Palumbo 2020). Information and communications technology (ICT) was widely adopted with regard to work and organizational management (Balica 2019; Kassick 2019; Nemțeanu, Dabija and Stanca 2021; Olsen 2019). WFH is characterized by the freedom from constraints associated with working in a formal and fixed workplace due to progress in ICT (Nakrošienė, Bučiūnienė and Goštautaitė 2019).

Long before the COVID‐19 pandemic, WFH had already been suggested as a modern human resource policy for organizations, and it has resulted in a definite trend firmly entrenched in society (Illegems, Verbeke and S’Jegers 2001; Stanek and Mokhtarian 1998). It enables employees to be more productive by avoiding long commutes, skirting office politics, having fewer office distractions, and giving more chance to develop a better work–life balance (Hopkins and McKay 2019; Nakrošienė, Bučiūnienė and Goštautaitė 2019). Simultaneously, a stream of scholars have argued that WFH is not an alternative working routine and may even lead to poor employee performance (Fonner and Roloff 2010). Thus, a key question in the field has been raised: Can WFH replace WFO? Around this question, the debate has become fierce alongside the development of ICT and globalization. Nevertheless, past research has not yet reached a consensus, which constitutes a significant gap in the current knowledge.

Thus, drawing on the above research gaps, the present research is designed as a comparison study contextualized in the ongoing COVID‐19 pandemic. On the basis of the job demand–control–support (JDCS) model, a well‐documented theory that elucidates the effects of fundamental job characteristics (Johnson and Hall 1988), and combined with entropy balance matching (Watson and Elliot 2016), the present study investigates the difference between WFH group and other working cohorts in terms of job characteristics and its effects on job performance. More specifically, based on the JDCS model, we propose the mediation effect of job demand and job control and the moderation effect of employers' anti‐epidemic policy as the social support on the relationship between job demand/job control and employee job performance.

The contributions of this study are as follows. First, we shed new light on the mixed effects of WFH on job performance. We find that WFH can increase job quality but reduce job productivity. Second, underpinned by the JDCS framework, the present paper empirically tests the differences of job characteristics between WFH and other working routines regarding job demand, job control and social supports, and its direct and indirect effects on employees' satisfaction on performance. In this case, the present paper extends the JDCS model from the field of classical work routine to understand WFH. Furthermore, we employ the entropy balancing method to alleviate the methodological concerns with selection bias in the previous literature. Doing so allows for examining the causal effect of WFH on job characteristics and job performance to support the random hypothesis in comparison quasi‐experiment research.

The remainder of this paper is organized as follows. The next section presents the literature review, followed by a discussion of the hypothesis development. Further sections present the methods and results, respectively. The final section presents a discussion and implications, followed by future scope and conclusion.

Literature review

WFH is a working arrangement in which employees fulfill the essential responsibilities that their job entails while remaining at home using ICT (International Labor Organization 2020, 5). Although a slight difference exists among terms such as WFH, teleworking, telecommuting and remote working, these concepts are largely interchangeable. WFH is considered home‐based teleworking, because teleworking may include various locations away from the primary worksite or the employers' premises (such as mobile working). Telecommuting refers to substituting telecommunications for commuter travel. Some differences exist between the terms teleworking and telecommuting, mainly because teleworking is broader and may not always be a substitute for commuting, but they are relatively minor. The basic difference between telework and remote work is that a teleworker uses personal electronic devices in addition to working physically remotely from a place other than an office or company premises, whereas remote work does not require visits to the main workplace or the use of electronic personal devices; and compared with WFH, remote work has the flexibility to work anywhere rather than being limited to the home. In addition, WFH may imply a long‐term contract, and individuals may have an emotional relationship with the organization; however, in remote work, this is not easy to achieve (Tønnessena, Dhira and Flåten 2021).

This paper aims to illustrate whether WFH can replace the classical working routine. A comparison study between WFH and other working routines seems to be a promising way to solve this question. However, we should consider two significant challenges of conducting a comparison study on WFH and other working routines. First, a ubiquitous theoretical framework is critical for providing solid support to capture fundamental job characteristics of diverse working routines. Only by doing so can we compare the difference between WFH and the other cohorts at the datum line. Second, we need to conquer the self‐selection bias. Most employees considering the possibility of WFH as the alternative way are familiar with applying ICT applications (e.g. email and online meeting apps) and necessary equipment (e.g. laptop and smartphone). In addition, employees' meta‐cognitive knowledge – their understanding of their capacity to cope with various situations under WFH ways (e.g. interruption caused by children and communication with line manager) – may play a similar self‐selective role. On the basis of these self‐selective factors, individuals evaluate the advantages and disadvantages of WFH and make decisions (Williams, McDonald and Cathcart 2017). Not controlling for this nonrandom self‐selection implies that observed job performance may reflect individuals' superior knowledge, capacity, or equipment rather than the actual effect of WFH. However, it is difficult to isolate the effects of job characteristics of WFH and the influence of individual heterogeneity explicitly associated with WFH. Thus, this paper adopts the JDCS model to investigate the effect of WFH on employees' job performance.

In the last 20 years, inconsistent findings have been found on the effect of WFH on employees' performance, especially in terms of work efficiency, turnover intention, goal completion, work motivation and job satisfaction (Gajendran and Harrison 2007; Golden 2006). On the one hand, some studies have found that WFH leads to high job performance (Bloom et al. 2015; Campo, Avolio and Carlier 2021; Choukir et al. 2022; Ipsen et al. 2021; Liu, Wan and Fan 2021). On the other hand, studies have found that WFH may lead to employees' lack of supervision, miscommunication, and less organizational commitment (Madell 2021). These disadvantages can create uncertainty that affects job satisfaction and consequently lead to lowering performance among employees, as gauged by companies' key performance indicators (Pepitone 2013). Some scholars have argued that WFH is negatively related to employees' job performance (Mustajab et al. 2020; Van Der Lippe and Lippényi 2020). Raišienė et al. (2020) suggested an investigation of the influence of WFH on job performance based on a contingency view, which depends on employees' gender, age, education, work experience, and telework experience. Table 1 summarizes the related literature.

Table 1.

Summary of related literature

Author	Objective	Methodology	Results/Findings	Association between WFH and performance
Bloom et al. (2015)	To investigate whether WFH works	Experiment	WFH led to a 13% performance increase	Positive
Choukir et al. (2022)	To investigate the effects of WFH on job performance	Survey, SEM	WFH positively affects employees’ job performance	Positive
Liu, Wan and Fan (2021)	To investigate the relationship between WFH and job performance	Survey, regression	WFH can improve job performance through job crafting	Positive
Ipsen et al. (2021)	To investigate people’s experiences of WFH during the pandemic and to identify the main factors of advantages and disadvantages of WFH	Survey, descriptive statistics, exploratory factor analyses, t‐test, ANOVA	WFH can improve work efficiency	Positive
Campo, Avolio and Carlier (2021)	To investigate the relationship among telework, job performance, work–life balance and family supportive supervisor behavior in the context of COVID‐19	Survey, partial least squares structural equation modelling (PLS‐SEM)	WFH is positively correlated with job performance	Positive
Van Der Lippe and Lippényi (2020)	To investigate the influence of co‐workers WFH on individual and team performance	Survey, SEM	WFH negatively impacted employee performance. Moreover, team performance is worse when more co‐workers are working from home	Negative
Mustajab et al. (2020)	To investigate the impacts of working from home on employee productivity	Survey, qualitative method with an exploratory approach	WFH is responsible for the decline in employee productivity	Negative
Raišienė et al. (2020)	To investigate the efficiency of WFH	Survey, correlation analysis	There are differences in the evaluation of factors affecting work efficiency and qualities required from a remote worker, depending on gender, age, education, work experience, and experience of telework	Contingency

Hypothesis development

Which one is better? Influence of WFH on job performance

The JDCS model provides a sound theoretical basis for the influence of WFH on job performance. It originated from the job demand–control (JDC) model, which explains how job characteristics alter employees' stress, performance and satisfaction (Karasek and Theorell 1990). The JDC model posits two fundamental characteristics of an occupation: job demand and job control. Job demand is defined initially as ‘physical consumptions and psychological tensions involved in accomplishing the workload’, which negatively relate to workplace well‐being and relevant performance (Karasek and Theorell 1990, 291). Job control (originally decision latitude) is the extent to which an employee has the authority to decide and utilize skills concerning the job and exert a positive effect on workplace outcomes. The JDCS model compounds the prominence of environmental factors on the overall well‐being within the workplace (Baka 2020). Thus, social support was integrated into the JDC model (named JDCS model) as a further fundamental characteristic of the work environment, implicating its synergistic effect on reducing stress and promoting well‐being in the working environment (Johnson and Hall 1988).

Given the inconsistent findings on the relationship between WFH and job performance, we further investigate the effect of WFH on job performance based on the JDCS model. The COVID‐19 pandemic has made WFH a sudden reality, as the ILO defined WFH in the context of the COVID‐19 pandemic as a temporary and alternative home‐based teleworking arrangement (ILO 2020). Waizenegger et al. (2020) articulated the differences between remote e‐working before and during the COVID‐19 pandemic.

Given the two mechanisms of JDCS, we further investigate the effect of WFC on job performance separately from the perspective of job demand and job control. On the one hand, WFH may lead to high job control, which benefits job performance, because not all job functions and tasks can be done outside the employers' premises or the specified workplace (Waizenegger et al. 2020). WFH is not practical or feasible or cannot be deployed quickly in some jobs and tasks (Williams, McDonald and Cathcart 2017). Accordingly, employees can arrange their time and energy with adequate job autonomy when they are WFH. They can deal with tasks under the best working status and promote work productivity and quality. On the other hand, WFH may lead to high job demand, which decreases job performance. Job demands are typically operationalized in terms of quantitative aspects, such as workload and time pressure (Hopkins and McKay 2019; Karasek and Theorell 1990). The boundary between working and leisure times becomes ambiguous when employees are WFH. Employees are usually pushed to work for longer hours and face high job demand, which is harmful to work productivity and quality. Therefore, assessing the influence of WFH on employees' feeling of their work completion is vaguer and more complicated compared with WFO, which leads us to propose our first hypothesis as a set of two alternatives:

Hypothesis 1a:

Employees who are WFH are more satisfied with their job performance (i.e. job quality and job productivity).

Hypothesis 1b:

Employees who are WFH are less satisfied with their job performance (i.e. job quality and job productivity).

Mediating role of job demand between WFH and job performance

On the basis of the JDCS model (Karasek and Theorell 1990), we tend to examine the differences of job fundamental characteristics and the moderating effect of social support on job performance between WFH and other working routines. WFH may increase job demand due to its possibility of pushing individuals to work for longer hours and increase the intensity of individuals. It will lead to a high investment of personal resources and bring adverse effects afterward.

First, WFH acquires more personal energy and time to invest in dealing with ‘communication via technology’, and employees may need to learn and equip with knowledge accordingly, including terms of using WFH tools and methods of collaboration (Yang et al. 2021). Moreover, employees may face the risks of technology fatigue or crash, which may result in negative psychological effects of misinformation and putting off work accomplishments (Khan 2021). Second, when employees need to continue to work beyond the regular working hours, they will inevitably face continuous additional work pressure, which makes them unable to relax and recover physically and mentally. Accordingly, more personal time and resources are demanded to invest in the job (Xie et al. 2018). Ayyagari, Grover and Purvis (2011) believed that WFH forms in such a convenient manner where employees may be required to stay on call for quarantine for a long time. WFH may influence employees' everyday life and lead to a perception of higher expectations for their working hours and intensity by their company and work loading. Ter Hoeven, van Zoonen and Fonner (2016) also verified this and reported that WFH may cost extra job demands from employees, including financial assets, energy, time and psychological capital. If those demands are too high, they may further make a series of workplace deviation behaviors, such as time‐encroached behaviors, to alleviate their loss of personal resources (Vayre 2021), consequently reducing their job performance.

Hypothesis 2:

The relationship between WFH and job performance is mediated by job demand.

Mediating role of job control between WFH and job performance

We further reason that the relationship between WFH and job performance is mediated by job control. The most prominent advantage of WFH is regarded as flexibly anytime and anywhere, which can significantly enhance employees' sense of job control and autonomy (Richardson and Thompson 2012). Mazmanian, Orlikowski and Yates (2013) found that employees who complete work tasks through WFH would have increased perceived work control and work flexibility. WFH can also enhance job autonomy in respect of task arrangement, work manner and task order (Mazmanian, Orlikowski and Yates 2013). Studies have also verified that WFH will promote employees' benefits in the field of the family via a more flexible and adaptable arrangement (Dockery and Bawa 2018). As a result, it can balance their work and family duties concerning the different daily situations and perform well (Tønnessena, Dhira and Flåten 2021).

Hypothesis 3:

The relationship between WFH and job performance is mediated by job control.

Moderating role of employers' anti‐epidemic policy

Social support is characterized by helpful relations with supervisors and coworkers (Mayo et al. 2012). Previous evidence has argued that a lack of support from employers when applying WFH may lead to a series of problems and thus reduce job performance (Palumbo 2020). According to the JDCS model, social support often buffers the effects of job demands and job control on the work‐related outcomes of employees (Johnson and Hall 1988). We investigate the moderation effect of social support on the relationship between job demand/control and job performance.

First, WFH may lead to isolation among employees if they have fewer interactions with their coworkers, supervisors and managers. Second, employees may not get recognition and support when needed, which may lead to employees' dissatisfaction, as their social needs cannot be fulfilled by WFH (Marshall, Michaels and Mulki 2007). Another negative consequence is receiving less recognition for achievements because exhibiting their work achievements is more difficult when all communication is conducted electronically (Zhang 2016). The limitation exists because when employees are WFH, they usually submit their work when it is ready. However, their manager may not see the process involved in producing a deliverable; some employees may work overtime, but their work is only judged by the result, not by the difficulties they overcome. Thus, policies or strategies should be implemented to enhance employers' feeling of embeddedness, not only for the sake of job performance but also for their well‐being and sustainability of human resourcing of organizations.

Particularly, considering the context of the epidemic, support actions from employers aiming to be anti‐epidemic and protect employees will be essential to improve the positive consequences of WFH. Thus, the present paper takes employers' anti‐epidemic policy as prominent social support worthy of examining. Indeed, some Chinese companies coined proactive guidance and support for employees (Reeves et al. 2020). The support reportedly helped employees feel less stressed, experience more positive feelings toward their leader and their team, and created an atmosphere of trust and understanding that motivated them to apply themselves more fully to work (Xu and Thomas 2011). In this case, we suggest that a moderating effect of the employers' anti‐epidemic policy is significantly observed on the influence of WFH on job performance. Figure 1 shows the conceptual framework.

Hypothesis 4a:

Social support moderates the relationship between job demand and job performance, such that the relationship is weaker when social support is high rather than low.

Hypothesis 4b:

Social support moderates the relationship between job control and job performance, such that the relationship is stronger when social support is high rather than low.

Figure 1.

The conceptual framework

Methods

Setting

Our sample was collected from China. It is the first region where the government applied a lockdown policy, which encouraged employers to organize their employees to WFH to mitigate the massive health crisis. Nevertheless, in March 2020, due to the sound control of COVID‐19 spread, after only a few months' lockdowns, Chinese citizens were able to return gradually to their normal work–life routine. As a result, some employees were WFH, and some of them returned to their normal work routine. Different from the previous research conducted in a limited number of industries or focusing on a particular occupation group, such a situation provides us a unique opportunity to design comparison research to understand early, initial reactions of a wide range of occupational groups and industries toward WFH and its social effect in the epidemic context.

Data

Data were collected via an online survey, provided by a Chinese survey company called Wenjuanxing (www.wjx.cn), a platform providing functions equivalent to Amazon Mechanical Turk. Research on WFH confronts a widely noted difficulty in managing data face‐to‐face, especially during this particular epidemic term. Thus, we chose to issue and collect the questionnaire online.

We initially did a pilot survey on 1 March 2020, with 100 observations. Later, after adjustments to the questionnaire, we issued the formal study of 5 March 2020, a month after the earliest date for work resumption according to the Chinese government. Thus, some employees were returning to workplace (RTW), and some continued WFH after Chinese New Year. As mentioned before, this particular time allows us to do a comparison study that covers various types of occupation and organization to seek the differences between WFH and RTW when society is confronted with a significant public health emergency. After collecting data for two weeks, we gathered 1342 observations.

Furthermore, to alleviate the self‐selective bias caused by participants passively excluded from WFH due to lacking necessary conditions, we took the inclusion criteria that required the participants to be equipped with requirements of WFH, such as essential online tools and Internet access. We identified the qualified group by asking, ‘Do you think you have the qualified conditions to be working from home (e.g., possesses Internet access, laptop, smart phone, software, and apps)?’ Then, we selected those who answered yes. After cleansing invalid data, the final sample consisted of 861 individuals, among which 442 claimed that they were WFH, and 419 were RTW.

Participants

Our sample comprised participants who were portrayed as young and received a high‐level of education, who were aged around 31–35 on average. The participants were 44% male. The majority of the participants were qualified with undergraduate degree. Particularly, 9.98% of the participants were married without children, 58.65% were married with children, 30.89% were single without children, and 0.4% were single with children. Around half of the participants (50.41%) worked for private enterprises, 16.7% worked for state‐owned enterprises, 15.21% worked for foreign companies, and others worked in government or public institutions. The participants at management positions accounted for 41%. Those who had marketing duties accounted for 31%. Others had positions in R&D. The participants worked for 9.36 days on average after the Chinese New Year (also the deadline of the epidemic blockade), and 71% of them had experience of training or education while WFH. The participants were from 16 places in China, the largest portions were from Guangdong Province (13.43%), Shanghai (7.66%), Shandong (6.15%), and Jiangsu (6.15%).

Measures

Dependent variable

Job performance was measured by two items adopted from a structured measurement coined by Viswesvaran, Ones and Schmidt’s (1996) measurement of job performance (overall job performance, productivity, and quality). We applied the two dimensions of job performance, namely, ‘productivity’ and ‘quality’, which were examined by self‐evaluation questions: 1) In terms of productivity, how do you evaluate the quantity or volume of work produced today (e.g. number of transactions completed)? 2) In terms of quality, how do you feel about how well the job was done today (You can consider several aspects of the quality of tasks completed, including lack of errors, accuracy to specifications, thoroughness, and amount of wastage)? The answers were measured using a Likert scale, from 1 (poor) to 5 (excellent). As a key self‐evaluation measurement of job performance, Viswesvaran, Ones and Schmidt’s (1996) instrument has been widely applied by following scholars in the fields of organizational behavior, psychology, and human resource management (Judge et al. 2001; Lee, Berry and Gonzalez‐Mulé 2019; Murphy 2020).

Independent variable

WFH was used here to identify the work status of respondents, with 1 representing WFH, and 0 representing WFO.

Mediators

Job demand and job control were measured following Gonzalez‐Mulé and Cockburn (2017) work, which is a well‐documented instrument widely applied in research and referred to as the JDC model.

Job demand was measured by eight questions (e.g. ‘To what extent do you agree that your job requires working very hard?’ ‘To what extent do you agree that your job requires working very fast?’). The answer was measured using a Likert scale, from 1 (completely disagree) to 5 (completely agree; Cronbach's alpha = 0.83).

Job control was measured by seven questions (e.g. ‘To what extent do you agree that your job allows you to make a lot of decisions on your own?’ ‘To what extent do you agree that you have a lot to say about what happens on your job?’). The answer was measured using a Likert scale, from 1 (completely disagree) to 5 (completely agree; Cronbach's alpha = 0.75).

Moderator

Social support was measured by employees' satisfaction on employers' anti‐epidemic policy. The survey question was, ‘Overall, are you satisfied with your employers’ anti‐epidemic support (e.g. financial support, emotional support from line managers, anti‐epidemic knowledge guides, and clear guidelines of WFH)?’ The answer was a dummy one, 1 representing yes, and 0 indicating no.

Control variables

First, we controlled for effective communication as a key factor that affects the quality of job performance, given that the majority of the literature has argued that ineffective communication is one of the greatest challenges of interpersonal collaborations mediated by ICTs in WHF (Wang et al. 2021). We controlled a set of communication factors in terms of ‘accurately delivered job content’ and ‘fully expressed the information’, among others. The answers were designed as a Likert scale, from 1 (completely disagree) to 5 (completely agree).

Furthermore, consistent with earlier studies, we controlled for difference of working hours, namely, the difference between daily working hours and today’s working hours, working experiences, normal daily working hours, daily number of colleagues they worked with, daily number of leaders they worked with, daily number of departments they worked with, daily commuting time, positions, age, gender, education, marital status, nature of employers, position levels, and days of starting work after the Chinese New Year. The definitions of variables are provided in Table A1.

Table A1.

Definition of variables

Variables	Definition	Cronbach alpha
Condition qualified with WFH	Is measured by following question: ‘Do you think you own the qualified conditions to working from home? (e.g. able to access internet, have laptop, smart phone, necessary software and apps)’ Answer: Dummy, 1: yes; 0: no	n.a.
Job performance – quality	Is measured by following question: ‘How do you feel about how well the job was done today? (You can consider several aspects of the quality of tasks completed including lack of errors, accuracy to specifications, thoroughness, and amount of wastage).’ Answer: A Likert Scale, 1 poor to 5 excellent	n.a.
Job performance – productivity	Is measured by following question: ‘How do you evaluate the quantity or volume of work produced today? (e.g. number of transactions completed, extent of daily task completed)’ Answer: A Likert Scale, 1 poor to 5 excellent	n.a.
WFH	Is measured by following question: ‘Do you work from home or return to workplace now?’ Answer: Dummy, 1: working from home; 0: working at workplace
Job control	Is measured by following 6 items: Con1: to what extent do you agree that your job allows you to make a lot of decisions on your own? Con2: to what extent do you agree that you have a lot of say about what happens on your job? Con3: to what extent do you agree that you can determine the order in which your work is to be done on your job? Con4: to what extent do you agree that you can determine when a task is to be done on your job? Con5: to what extent do you agree that you can determine your own work rate on your job? Con6: to what extent do you agree that you have very little freedom to decide how you do your work on the job? Answer: A Likert Scale, 1 completely disagree to 5 completely agree	.75
Job demand	Is measured by following 9 items: Dem1: to what extent do you agree that your job requires working very hard? Dem2: to what extent do you agree that your job requires working very fast? Dem3: to what extent do you agree that your job requires long periods of intense concentration? Dem4: to what extent do you agree that your job is very hectic? Dem5: to what extent do you agree that you have too much work to do everything well on your job? Dem6: to what extent do you agree that you are not asked to do an excessive amount of work at your job? (reverse scored) Dem7: to what extent do you agree that you have enough time to get the job done? (reverse scored) Dem8: to what extent do you agree that that you are free of conflicting demands that others make on your job? (reverse scored) Dem9: How frequently does your job require working under time pressure? Answer: A Likert Scale, 1 completely disagree to 5 completely agree	.77
Social support	Is measured by following question: ‘Overall, are you satisfied with your employer’s anti‐epidemic support? (e.g. financial support, emotional support from line managers, anti‐epidemic knowledge guides, clear guidelines of WFH)’ Answer: Dummy, 1: yes; 0: no	n.a.
Effective communication	Is measured by following questions: Com1: to what extent do you agree that the inter‐personal communication related to your job can accurately delivery job content? Com2: to what extent do you agree that the inter‐personal communication related to your job fully express the information? Com3: to what extent do you agree that you are well acknowledged the process of the team project? Com4: to what extent do you agree that the inter‐personal communicating message is delivered in a positive way? Com5: to what extent do you agree that the inter‐personal communicating message is delivered in a negative way? Com6: recently, communication conflicts have quite often had a negative impact on completing my daily work. Com7: I feel the relationships with my colleagues are not as close asthey used to be. Answer: A Likert Scale, 1 completely disagree to 5 completely agree	.83
Daily working hours	Is measured by following question: ‘recently, how many hours have you needed to work daily?’ Answer: Numbers	n.a.
Difference of working hours	Is calculated by: Daily working hours – Daily hours used to work Daily hours used to work is measured by following question: ‘how many hours did you need to work daily before lockdown?’ Answer: Numbers	n.a.
Working experiences	Is measured by following question: ‘How many years since you got your first job’ Answer: years	n.a.
Daily number of colleagues work with	Is measured by following question: ‘On average, how many colleagues do you need to communicate with on daily base?’ Answer: Numbers	n.a.
Daily number of leaders work with	Is measured by following question: ‘On average, how many leaders do you need to report to on a daily basis?’ Answer: Numbers	n.a.
Daily number of departments work with	Is measured by following question: ‘On average, how many departments do you need to communicate with on a daily basis?’ Answer: Numbers	n.a.
Daily commuting time	Is measured by following question: ‘On average, how many hours did you spend commuting to the workplace?’ Answer: Numbers	n.a.
Positions	Is measured by following question: ‘What is your position?’ Answer: 1: Management position, 2: R&D position, 3: Rear‐Service positions, 4: Marketing position，5:Other	n.a.
Position levels	Is measured by following question: ‘What’s the level of your position?’ Answer: 1: rank‐and‐file employee, 2: middle manager 3: top manager	n.a.
Nature of employers	Is measured by following question: ‘What’s the nature of your employer?’ Answer: 1: government 2: public institutions, 3: foreign‐funded enterprise and joint venture, 4: state‐owned enterprise; 5: private enterprise	n.a.
Age	Answer: 1: under 25, 2: 25–30, 3: 31–35, 4: 36–40, 5: 41–50, 6: over 50	n.a.
Gender	Answer: 1: male, 0:female	n.a.
Education	Answer: 1: no degree to 5: postgraduate degree and above	n.a.
Marriage & Children	Answer: 1: married, no child, 2: married, have a child or children, 3: single, no child, 4: single, have a child or children	n.a.
Days of starting work after Chinese New Year	Is measured by following question: ‘How many days since you started to work after Chinese New Year?’ Answer: Numbers	n.a.
WFH Training	Is measured by following question: ‘Do you ever have training experience working from home? (e.g., remote work apps, training on communications via online tools),’ Answer: Dummy, 1: yes; 0: no	n.a.

Analysis strategy

Our analysis consists of three steps. In Step 1, to test our hypothesis 1, we applied entropy balance and weighted mean difference Welch's t‐test (mean after entropy balance matching) methods to compare the self‐evaluated job performance between WFH and WFO employees. Following the approach of recent papers on labor economics and health (Hetschko, Schöb and Wolf 2016; Kunze and Suppa 2017; Nikolova, 2019), our strategy includes 1) data preprocessing to form comparable groups of individuals as treatment and control group (treatment group: WFH employees; control group: RTW employees) by applying entropy balance, and 2) estimating the treatment effect after matching by Welch's t‐test. We also reconfirmed the regression result (Hainmueller 2012).

In Step 2, we investigated the direct and mediating effects of job control and job demand on job performance (hypotheses 2 and 3). We applied the quasi‐Bayesian Monte Carlo method to test the mediating effect of job demand and job control, which is a technique to increase the robustness of the mediating test by employing a strategy of numerous repeated re‐sampling to build an empirical approximation of the sampling distribution and examine the indirect effects by constructing the confidence intervals (CIs; Imai, Keele and Tingley 2010). We used the package ‘Mediation’ for causal mediation analysis. In addition, to confirm the validity and reliability of mediating hypotheses results, we used structural equation modeling (SEM) as robustness check, with package ‘lavaan’ to assess the mediating effect of job control and job demand on the relationship between WFH and job performance.

In Step 3, to test the moderating effect of social support, we applied hierarchical regressions at the final step by following the classical approaches to seek the significance of interactions in a set of model tests.

All the analysis is conducted with software R.

Results

Before testing the hypotheses, a benchmark test of a binary correlation matrix is presented in Table 2. The overall coefficient is not high, and a variance inflation factor was performed at below 10, demonstrating low multicollinearity.

Table 2.

Variables correlation matrix

		1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	27	28	29	30	31	32	33
1.	Job performance – quality
2.	Job performance – productivity	.41
3.	WFH	.29	−.12
4.	Job control	.18	.24	.06
5.	Job demand	.13	.25	−.12	.18
6.	Social support	.16	.27	.00	.23	.16
7.	Effective communication	−.11	−.17	.06	−.11	−.05	−.08
8.	Daily working hours	.03	.04	−.04	−.15	−.19	−.14	−.10
9.	Difference of working hours	−.03	−.06	−.04	−.04	.04	−.05	−.02	.00
10.	Working experiences	.00	.13	−.14	.12	.03	.11	−.15	−.02	.01
11.	Daily number of colleagues work with	.01	.13	−.13	.02	.07	.03	−.11	.06	−.02	.17
12.	Daily number of leaders work with	.01	.11	.00	.04	.11	−.02	−.06	.12	−.08	.10	.51
13.	Daily number of departments work with	−.06	.06	−.06	.06	.04	.01	−.02	.11	−.07	.07	.36	.44
14.	Daily commuting time	.03	.04	.06	−.03	−.06	−.01	−.02	.08	−.04	.05	.10	.14	.11
15.	Management	.03	.06	.03	.10	.04	.00	.04	.04	.01	.04	.12	.13	.25	.03
16.	Research	.04	.05	−.03	.05	.08	.02	−.07	−.02	−.05	−.01	.02	.08	−.03	.02	−.25
17.	Service	−.08	−.01	−.12	−.01	.05	−.04	.07	−.01	.05	.01	−.04	−.06	.01	.03	−.17	−.09
18.	Marketing	−.01	−.05	.01	−.08	−.02	−.01	−.01	−.02	.02	.02	.01	−.01	−.06	−.03	−.32	−.25	−.13
19.	Other	−.03	−.05	.07	−.04	−.07	.01	.02	−.01	.01	−.05	−.08	−.09	−.05	−.02	−.26	−.15	−.07	−.17
20.	Position levels	.02	.08	−.04	.12	.10	.06	.07	−.02	−.04	.27	.21	.22	.25	.02	.31	.08	−.13	−.06	−.15
21.	Government	.02	.02	−.04	.01	.01	.01	.05	−.01	−.01	−.12	−.09	−.02	−.04	−.04	.03	.01	.05	−.09	.06	.01
22.	Public institutions	.04	.02	.08	.03	.06	.05	.00	−.01	.01	−.03	−.02	.05	.07	−.02	.01	.03	.05	−.04	.08	.01	−.06
23.	Foreign‐funded enterprise and joint venture	.05	.02	−.01	.04	.00	.03	−.01	−.01	−.06	.03	.11	.09	.09	.06	.10	.08	−.03	−.06	−.06	.12	−.07	−.15
24.	State‐owned enterprise	−.06	.01	−.03	.00	.04	.06	−.03	−.02	.01	.09	.02	.04	.05	.05	.01	.01	.03	.00	−.04	−.09	−.07	−.16	−.19
25.	Private enterprise	−.03	−.04	.00	−.07	−.08	−.10	.01	.03	.04	−.04	−.06	−.11	−.13	−.07	−.09	−.07	−.04	.08	−.02	−.03	−.16	−.37	−.43	−.45
26.	Age	.01	.11	−.06	.09	.06	.09	−.14	−.02	−.02	.72	.10	.12	.06	.04	.05	−.03	−.01	.02	−.01	.29	−.05	.10	−.02	.10	−.11
27.	Gender	.02	.03	−.04	.01	.10	.09	−.03	−.03	.03	.08	.01	.02	−.03	.00	−.04	.18	−.12	.04	.01	.14	.03	.01	−.04	.03	.01	.10
28.	Education	.06	.02	−.02	.09	−.03	−.04	−.04	−.05	−.06	−.05	.09	.13	.10	.01	.12	.24	−.17	−.14	−.10	.17	.03	.09	.09	−.02	−.13	−.09	−.02
29.	Married, no child	.05	−.03	.05	−.06	−.13	−.03	−.04	.10	.00	−.06	−.05	−.04	−.09	.07	−.04	.03	.00	−.03	.09	−.06	.12	−.03	.05	−.09	.02	−.08	.03	.08
30.	Married, have a child or children	−.05	.06	−.13	.12	.13	.12	−.07	−.04	−.03	.57	.13	.09	.13	−.01	.11	.01	.02	−.02	−.14	.31	−.10	.07	−.01	.12	−.09	.52	.03	−.03	−.40
31.	Single, no child	.02	−.04	.11	−.08	−.05	−.11	.10	−.02	.03	−.57	−.11	−.07	−.08	−.04	−.09	−.02	−.03	.03	.09	−.30	.02	−.05	−.02	−.07	.09	−.51	−.04	−.02	−.22	−.08
32.	Single, have a child or children	−.03	−.04	.00	−.05	−.01	.01	.01	.00	.00	.05	.00	.03	−.01	.01	−.02	−.03	.02	.07	−.03	.07	−.01	−.02	.02	.02	.00	.06	−.03	−.02	−.02	−.08	−.05
33.	Days of starting work after Chinese New Year	−.01	.04	−.11	.08	.06	−.01	−.01	−.02	.03	.09	.03	.05	.01	.02	.00	.07	.01	−.03	−.03	.04	.06	−.05	.07	.00	−.04	.03	.05	.09	.03	.01	−.03	−.01
34.	WFH training	.08	.07	.08	.06	.14	.13	−.02	.02	.03	−.02	.05	.11	.12	.02	.10	.08	−.02	.00	−.09	.16	.03	.03	.07	−.02	−.10	−.01	.09	.01	−.04	.09	−.08	.05	−.11

Influence of WFH on self‐reported job performance (hypotheses1a and 1b tests)

Before proceeding to test hypothesis 1 in Step 1, we first applied the entropy balance and weighted mean difference (mean after entropy balance matching) methods. The quality of entropy balance matching combined with a data description is summarized in Table 4. Before matching, WFH employees worked for <2.7 h daily on average compared with their pre‐daily working hours. Employees who had returned to work worked <0.53 h on average than their current daily work. After matching, this difference was reduced. WFH employees are used to having less colleagues to work with (mean: WFH = 2.91, RTW = 3.18), are less likely to work at back office (mean: WFH = 0.10, RTW = 0.18), are younger (mean: WFH = 2.62, RTW = 2.77), are less likely to be married and have a child or children (mean: WFH = 0.52, RTW = 0.56), and are more likely to be single and without a child or children (mean: WFH = 0.36, RTW = 0.26). In addition, WFH employees indicated that they started working after Chinese New Year a day later than WFO employees (mean: WFH = 9.36, RTW = 11.09). In particular, WFH employees experienced better interpersonal communication than RTW employees (mean: WFH = 2.74, RTW = 2.67). In entropy balance matching, we matched all conditioning variables, and the bias of each matched variables was reduced to nearly 0, supporting good quality of entropy balance matching. Moreover, differences in mean and variance between the treatment and control groups were largely reduced after weighting (see in Table A2).

Table 4.

Causal mediation analysis of job control and job demand

	via Job Control	via Con2	via Con5	via Job Demand	via Dem3	via Dem4
Regression on job performance – quality
Mediating effect	.14***	.02^	.12***	−.02**	−.03**	−.03*
Direct effect	.45***	.46***	.45***	.50***	.50***	.49***
Total effect	.59***	.48***	.57***	.48***	.48***	.48***
Prop. mediated	23.72%**	4.16%^	21.05%**	4.33%*	5.3%*	6%^
Regression on job performance – productivity
Mediating effect	.03*	. 01*	.05***	−.03***	−.03**	−.001
Direct effect	−.19***	−.17***	−.21***	−.12***	−.13**	−.16***
Total effect	−.17***	−.17**	−.17***	−.15***	−.16***	−.16***
Prop. mediated	16.4%*	5.88%^	29.41%***	21.25%	20.11%**	4.9%

^p < 0.1; *p < 0.05; **p < 0.01; ***p < 0.001.

Table A2.

Descriptive statistics before treatment, selected covariate variables, before and after matching

	Treated		Controls unmatched		Controls matched		Standardized bias %
	N = 442		N = 419		N = 419
	Mean	Variance	Mean	Variance	Mean	Variance	Unmatched	Matched
Effective communication	2.74	.37	2.67	.30	2.74	.33	.12	.00
Difference of working hours	−2.73	117.00	−.53	72.15	−2.73	97.22	.29	.00
Daily working hours	3.43	1.40	3.77	1.30	3.43	1.40	.09	.00
Working experiences	3.56	1.31	3.55	1.43	3.56	1.55	.01	.00
Daily number of colleagues work with	2.91	1.08	3.18	1.04	2.91	.85	.27	.00
Daily number of leaders work with	2.14	.64	2.15	.59	2.14	.56	.01	.00
Daily number of departments work with	2.34	.61	2.44	.63	2.34	.55	.13	.00
Daily commuting time	2.18	.67	2.09	.58	2.18	.64	.12	.00
Management	.41	.24	.38	.24	.40	.24	.12	.00
Research	.20	.16	.22	.17	.20	.16	.06	.00
Service	.10	.09	.18	.15	.10	.09	.05	.00
Marketing	.31	.22	.31	.21	.31	.22	.23	.00
Other	1.40	.35	1.44	.32	1.40	.34	.01	.00
Position levels	.15	.13	.10	.09	.15	.13	.07	.00
Government	.02	.02	.03	.03	.02	.02	.08	.00
Public institutions	.16	.15	.10	.12	.16	.15	.17	.00
Foreign‐funded enterprise and joint venture	.15	.13	.16	.13	.15	.13	.03	.00
State‐owned enterprise	.16	.13	.18	.15	.16	.13	.06	.00
Private enterprise	.50	.25	.51	.25	.50	.25	.01	.00
Age (under 25)	.44	.25	.39	.24	.43	.25	.15	.00
Age (25–30)	.25	.19	.33	.22	.27	.20	.10	.00
Age (31–35)	.10	.09	.11	.10	.10	.10	.18	.00
Age (36–40)	.07	.07	.07	.06	.07	.07	.06	.00
Age (41–45)	.02	.02	.02	.02	.02	.02	.01	.00
Age (over 45)	.02	.14	.02	.14	.02	.14	.01	.00
Gender (male)	.41	.24	.46	.25	.41	.24	.09	.00
Education (no degree)	.05	.05	.02	.02	.05	.04	.14	.00
Education (primary school)	.15	.13	.15	.13	.15	.13	.02	.00
Education (high school)	.69	.21	.73	.20	.71	.21	.08	.00
Education (undergraduate)	.11	.32	.10	.30	.11	.32	.04	.00
Education (postgraduate degree and above)	.001	.05	0	0	–	–	–	–
Married, no child	.11	.10	.09	.08	.11	.10	.10	.00
Married, have a child or children	.52	.25	.65	.23	.53	.25	.26	.00
Single, no child	.36	.23	.26	.19	.36	.23	.22	.00
Single, have a child or children	.00	.00	.00	.00	.00	.00	.00	.00
Days of starting work after Chinese New Year	9.36	69.62	11.09	59.76	9.36	51.00	.22	.00
WFH training	.71	.21	.64	.23	.71	.21	.15	.00

Then, we verified hypothesis 1 by measuring the ATT under the balanced matching conditions in Table 3. After matching, the results for hypothesis 1 are presented in Tables 5 and 6. The results show that WFH employees are more satisfied with quality (mean: WFH = 4.56, RTW = 4.11, p < 0.01). In addition, WFH employees feel less satisfied with productivity (mean: WFH = 3.86, RTW = 4.05, p < 0.01). Hypotheses 1a and 1b were supported.

Table 3.

Treatment effect of WFH before and after entropy balance matching

	Treated group	Controls unmatched	Treatment effect (unmatched)		Controls matched	Treatment effect (matched)
	Mean	Mean	Mean difference	t‐Test	Mean	Mean difference	t‐Test
Job performance – quality	4.56	4.11	.45	8.92***	4.11	.45	8.83***
Job performance – productivity	3.86	4.05	−.19	−3.41***	4.03	−.17	−3.1**
Job control	3.67	3.59	.08	1.81*	3.58	.09	2.10*
Con1	3.62	3.567	.05	.68	3.59	.03	.38
Con2	3.24	3.01	.03	2.04*	3.06	.18	2.36*
Con3	3.76	3.84	−.08	−1.12	3.83	−.07	−1.09
Con4	3.61	3.67	−.06	−.78	3.71	−.01	.21
Con5	3.69	3.17	.52	6.85***	3.16	.53	7.14***
Con6	3.60	3.60	0	−.02	3.53	.07	.028
Job demand	3.37	3.48	−.11	−3.50***	3.46	−.09	−3.07***
Dem1	2.64	2.63	.01	.16	2.64	0	−.12
Dem2	3.14	3.29	−.15	−2.27*	3.25	−.11	−1.67
Dem3	3.57	3.75	−.18	−2.77***	3.73	−.16	−2.47*
Dem4	3.25	3.50	−.25	−3.53***	3.48	−.23	−3.26**
Dem5	3.10	3.18	−.08	−1.21	3.15	−.05	−.65
Dem6	3.72	3.87	−.15	−1.84	3.87	−.15	−1.84
Dem7	3.69	3.72	−.03	−.25	3.77	−.08	−.96
Dem8	3.85	3.87	−.02	−.25	3.81	.04	.59
Dem9	4.12	4.19	−.07	.24	4.17	−.05	−.88
Social support	4.17	4.17	.00	−.08	4.17	.00	−.14

*p < 0.05; **p < 0.01; ***p < 0.001.

Table 5.

Regressions on satisfaction with job performance (quality)

	M1	M2	M3	M4	M5	M6	M7
WFH	.49 (.05)***	.48 (.05)***	.53 (.05)***	.50 (.05)***	.48 (.05)***	.53 (.05)***	.52 (.05)***
Mediators
Job control		.20 (.04)***			.23 (.21)		.27 (.22)
Job demand			.33 (.06)***			.32 (.31)	.25 (.32)
Social support				.16 (.03)***	.19 (.18)	.15 (.25)	.21 (.27)
Interactions
Job control * Social support					−.02 (.05)		.01 (.07)
Job demand * Social support						.00 (.07)	−.03 (.05)
Conditioning variables
Effective communication	−.16 (.05)***	−.14 (.05)**	−.15 (.04)***	−.14 (.04)**	−.13 (.04)**	−.14 (.04)**	−.13 (.04)**
Daily working hours	.00 (.00)	.00 (.00)*	.00 (.00)*	.00 (.00)*	.00 (.00)*	.00 (.00)*	.00 (.00)**
Difference of working hours	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)
Working experiences	.04 (.04)	.03 (.03)	.05 (.03)	.03 (.03)	.03 (.03)	.04 (.03)	.04 (.03)
Daily number of colleagues work with	.03 (.03)	.04 (.03)	.04 (.03)	.03 (.03)	.04 (.03)	.04 (.03)	.04 (.03)
Daily number of leaders work with	−.02 (.04)	−.02 (.04)	−.04 (.04)	−.01 (.04)	−.02 (.04)	−.03 (.04)	−.03 (.04)
Daily number of departments work with	−.07 (.04).	−.08 (.04)*	−.07 (.04)	−.08 (.04)*	−.08 (.04)*	−.07 (.04)	−.08 (.04)*
Daily commuting time	.00 (.03)	.01 (.03)	.01 (.03)	.00 (.03)	.00 (.03)	.01 (.03)	.01 (.03)
Management	.09 (.08)	.08 (.08)	.07 (.08)	.10 (.08)	.09 (.08)	.09 (.08)	.08 (.08)
Research	.07 (.09)	.07 (.09)	.05 (.09)	.08 (.09)	.08 (.09)	.06 (.09)	.06 (.09)
Service	−.02 (.10)	−.02 (.10)	−.04 (.10)	.01 (.10)	.00 (.10)	−.02 (.10)	−.02 (.10)
Marketing	.02 (.08)	.03 (.08)	.01 (.08)	.03 (.08)	.04 (.08)	.03 (.08)	.04 (.08)
Position levels	−.14 (.10)	−.13 (.10)	−.13 (.10)	−.14 (.10)	−.13 (.10)	−.14 (.10)	−.13 (.10)
Government	.00 (.05)	−.02 (.05)	−.02 (.05)	−.01 (.05)	−.02 (.05)	−.02 (.05)	−.03 (.05)
Public institutions	.15 (.25)	.15 (.25)	.13 (.25)	.12 (.25)	.13 (.25)	.11 (.25)	.12 (.25)
Foreign‐funded enterprise and joint venture	−.06 (.16)	−.04 (.16)	−.08 (.16)	−.09 (.16)	−.07 (.16)	−.10 (.15)	−.08 (.15)
State‐owned enterprise	.01 (.17)	.04 (.17)	.02 (.17)	.00 (.17)	.02 (.17)	.01 (.17)	.03 (.17)
Private enterprise	−.15 (.17)	−.11 (.17)	−.14 (.17)	−.17 (.17)	−.14 (.17)	−.17 (.17)	−.14 (.17)
Age	−.07 (.16)	−.03 (.16)	−.06 (.16)	−.07 (.16)	−.04 (.16)	−.06 (.16)	−.03 (.16)
Gender	.03 (.03)	.03 (.03)	.03 (.03)	.03 (.03)	.03 (.03)	.03 (.03)	.03 (.03)
Education	.01 (.05)	.01 (.05)	.00 (.05)	.00 (.05)	.01 (.05)	−.01 (.05)	.00 (.05)
Married, no child	.04 (.05)	.03 (.05)	.06 (.05)	.05 (.05)	.04 (.05)	.07 (.04)	.06 (.04)
Married, have a child or children	−.11 (.09)	−.12 (.09)	−.17 (.09)	−.11 (.09)	−.12 (.09)	−.17 (.09)	−.17 (.09)
Single, no child	−.07 (.09)	−.09 (.09)	−.11 (.09)	−.05 (.09)	−.07 (.09)	−.09 (.09)	−.10 (.09)
Single, have a child or children	−.50 (.39)	−.43 (.39)	−.54 (.38)	−.48 (.39)	−.43 (.38)	−.52 (.38)	−.48 (.38)
Days of starting work after Chinese New Year	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)
WFH training	.07 (.06)	.06 (.06)	.03 (.06)	.04 (.06)	.04 (.06)	.01 (.06)	.00 (.06)
R square	.15	.17	.17	.17	.18	.19	.20
Adjust R square	.12	.14	.14	.14	.15	.16	.17
F‐value	5.00	5.67	6.01	5.74	5.84	6.28	6.21
p‐Value	.00	.00	.00	.00	.00	.00	.00

^p < 0.1; *p < 0.05; **p < 0.01; ***p < 0.001; Standard errors in parentheses.

Table 6.

Regressions on satisfaction with job performance (productivity)

	M1	M2	M3	M4	M5	M6	M7
WFH	−.16 (.05)**	−.18 (.05)***	−.11 (.05)*	−.15 (.05)**	−.17 (.05)***	−.11 (.05)*	−.13 (.05)**
Mediators
Job control		.30 (.05)***			.22 (.21)		.32 (.22)
Job demand			.46 (.06)***			.11 (.30)	.05 (.31)
Social support				.27 (.03)***	.22 (.18)	.00 (.24)	.05 (.26)
Interactions
Job control * Social support					.00 (.05)		.03 (.05)
Job demand * Social support						−.07 (.07)*	−.08 (.07)*
Conditioning variables
Effective communication	−.21 (.05)***	−.18 (.05)***	−.21 (.05)***	−.19 (.05)***	−.17 (.04)***	−.19 (.04)***	−.17 (.04)***
Daily working hours	.00 (.00)	.00 (.00)	.00 (.00)*	.00 (.00)*	.00 (.00)*	.00 (.00)*	.00 (.00)**
Difference of working hours	.00 (.00)*	.00 (.00)*	.00 (.00)**	.00 (.00)	.00 (.00)	.00 (.00)*	.00 (.00)*
Working experiences	.06 (.04)	.05 (.04)	.07 (.04)*	.05 (.03)	.04 (.03)	.06 (.03)	.05 (.03)
Daily number of colleagues work with	.05 (.03)	.05 (.03)	.06 (.03)	.04 (.03)	.05 (.03)	.05 (.03)	.06 (.03)
Daily number of leaders work with	.02 (.04)	.02 (.04)	.00 (.04)	.04 (.04)	.03 (.04)	.01 (.04)	.01 (.04)
Daily number of departments work with	−.04 (.04)	−.05 (.04)	−.03 (.04)	−.05 (.04)	−.06 (.04)	−.05 (.04)	−.05 (.04)
Daily commuting time	.05 (.03)	.06 (.03)	.06 (.03)	.04 (.03)	.05 (.03)	.06 (.03)	.06 (.03)*
Management	.13 (.08)	.11 (.08)	.10 (.08)	.15 (.08)	.13 (.08)	.12 (.08)	.12 (.08)
Research	.07 (.09)	.07 (.09)	.03 (.09)	.09 (.09)	.09 (.09)	.05 (.09)	.05 (.09)
Service	−.04 (.10)	−.04 (.10)	−.07 (.10)	.01 (.10)	.00 (.10)	−.03 (.10)	−.04 (.09)
Marketing	−.09 (.08)	−.08 (.08)	−.10 (.08)	−.06 (.08)	−.06 (.08)	−.07 (.08)	−.06 (.08)
Position levels	−.12 (.10)	−.10 (.10)	−.12 (.10)	−.13 (.10)	−.11 (.10)	−.12 (.10)	−.11 (.10)
Government	.08 (.06)	.05 (.05)	.05 (.05)	.06 (.05)	.04 (.05)	.04 (.05)	.02 (.05)
Public institutions	.10 (.26)	.09 (.25)	.08 (.25)	.05 (.25)	.05 (.25)	.03 (.25)	.04 (.24)
Foreign‐funded enterprise and joint venture	.06 (.16)	.09 (.16)	.03 (.16)	.01 (.16)	.04 (.16)	.00 (.15)	.03 (.15)
State‐owned enterprise	−.08 (.18)	−.04 (.17)	−.06 (.17)	−.11 (.17)	−.07 (.17)	−.08 (.17)	−.05 (.17)
Private enterprise	−.02 (.18)	.03 (.17)	−.02 (.17)	−.07 (.17)	−.03 (.17)	−.06 (.17)	−.02 (.17)
Age	−.03 (.17)	.02 (.16)	−.01 (.16)	−.03 (.16)	.01 (.16)	−.01 (.16)	.02 (.16)
Gender	.06 (.04)	.06 (.04)	.06 (.03)	.07 (.03)	.07 (.03)	.07 (.03)*	.07 (.03)*
Education	−.07 (.06)	−.05 (.05)	−.08 (.05)	−.08 (.05)	−.07 (.05)	−.08 (.05)	−.07 (.05)
Married, no child	−.03 (.05)	−.04 (.05)	.00 (.05)	.00 (.05)	−.01 (.04)	.02 (.04)	.01 (.04)
Married, have a child or children	−.05 (.09)	−.08 (.09)	−.15 (.09)	−.05 (.09)	−.07 (.09)	−.14 (.09)	−.14 (.09)
Single, no child	.16 (.09)	.13 (.09)	.10 (.09)	.18 (.09)	.15 (.09)	.13 (.09)	.11 (.09)
Single, have a child or children	−.48 (.40)	−.37 (.39)	−.54 (.39)	−.44 (.39)	−.36 (.38)	−.49 (.38)	−.43 (.38)
Days of starting work after Chinese New Year	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)
WFH training	.13 (.06)*	.11 (.06)	.07 (.06)	.07 (.06)	.07 (.06)	.02 (.06)	.02 (.06)
R square	.11	.15	.16	.17	.19	.22	.23
Adjust R square	.08	.12	.13	.14	.16	.18	.2
F‐value	3.56	5.1	5.5	5.78	6.41	7.22	7.48
p‐Value	.00	.00	.00	.00	.00	.00	.00

*p < 0.05; **p < 0.01; ***p < 0.001; Standard errors in parentheses.

Mediating role of job demand and job control (hypotheses 2 and 3 tests)

Changes in job demand and job control can be observed from Step 1 in Tables 5 and 6. Under balanced matching conditions, WFH employees experience a significantly higher level of job control (ATT: WFH = 3.67, RTW = 3.58, p < 0.05). More specifically, such change is noteworthy in the job control of ‘talking right’ (con2; ATT: WFH = 3.24, RTW = 3.06, p < 0.05) and job control of ‘working rate’ (con5; ATT: WFH = 3.69, RTW = 3.16, p < 0.001). In terms of job demand, WFH employees experience a significantly lower level than RTW employees (mean: WFH = 3.37, RTW = 3.46, p < 0.001). The difference is obviously observed in terms of ‘long periods of intense concentration’ (dem3; ATT: WFH = 3.57, RTW = 3.73, p < 0.05) and ‘hecticness of the job’ (dem4; ATT: WFH = 3.25, RTW = 3.48, p < 0.01). These results imply that WFH may lead to changes in job control and job demand, which may intermediately affect job performance.

Therefore, in the second step, we tested the mediating effect by applying the quasi‐Bayesian Monte Carlo method in Table 4. The results show that in terms of quality, the mediating effect of job control and job demand is confirmed as statistically significant (job control = 0.14, p < 0.001; job demand = −0.02, p < 0.01). The proportion of mediating effect on total effect is around 23.72% and 4.33%. We also tested the mediating effect of the important items of job control and job demand. We find that the job control on ‘working rate’ (con5; 0.12, p < 0.10, prop. mediated = 21.05%), job demand on ‘long periods of intense concentration’ (dem3; −0.03, p < 0.01, prop. mediated = 5.3%), and ‘hecticness of the job’ (dem4; −0.03, p < 0.05, prop. mediated = 6%) positively mediate the relationship between WFH and satisfaction with quality.

In terms of productivity performance, the mediating effect of job control and job demand is supported (job control = 0.03, p < 0.05, prop. mediated = 16.4.5%; job demand = −0.03, p < 0.01, prop. mediated = 21.25%). However, it is noticeable, unlike in the domain of quality, that the mediating effect of job control and job demand contributes to the direct impact of WFH. Such mediating effect trades off the direct influence of WFH on satisfaction with productivity. Items such as job control on ‘working rate’ (con5; 0.01, p < 0.05, prop. mediated = 5.88%) and job demand on ‘long periods of intense concentration’ (dem3; −0.03, p < 0.01, prop. mediated = 20.11%) mediate the relationship between WFH and satisfaction with productivity. In this case, hypotheses 3 and 4 are fully supported.

In addition, the robustness check results via SEM analysis (both classical and bootstrap approach is used) is consistent with the quasi‐Bayesian Monte Carlo analysis. Accordingly, hypotheses 3 and 4 are supported as well (see details in Tables A3 and A4).

Table A3.

Robustness check of mediation effect by structure equation modelling

	Description	χ2	df	GFI	NNFI	CFI	RMSEA	SRMR
Accept values				>.90	>.90	>.95	<.05	<.08
M1	Full items model	1592.36	467	.795	.992	.994	.053	.049
M2	Dropped items model	394.16	194	.915	.998	.999	.035	.024
M3	Dropped items model (bootstrap)	394.16	194	.915	.998	.999	.035	.024
M4	Mean	15.456	1	.991	.991	1	.13	.006
M5	Mean (bootstrap)	15.456	1	.991	.991	1	.13	.006

Table A4.

Robustness check of mediation effect by structure equation modelling

	M1		M2		M3		M4		M5
	Quality	Productivity	Quality	Productivity	Quality	Productivity	Quality	Productivity	Quality	Productivity
	Path coefficient	Path coefficient	Path coefficient	Path coefficient	Path coefficient	Path coefficient	Path coefficient	Path coefficient	Path coefficient	Path coefficient
WEF	.44 (.05)***	−.19 (.05)***	.45 (.05)***	−.18 (.06)**	.45 (.05)***	−.18 (.06)**	.47 (.05)***	−.14 (.05)**	.47 (.05)***	−.14 (.06)**
Mediator
Job control	.24 (.05)***	.38 (.06)***	.20 (.05)***	.34 (.06)***	.20 (.06)***	.34 (.07)***	.17 (.04)***	.26 (.04)***	.17 (.04)***	.26 (.05)***
Job demand	.07 (.07)	.11 (.08)	.21 (.07)***	.27 (.07)***	.21 (.08)**	.27 (.08)***	.29 (.06)***	.37 (.06)***	.29 (.06)***	.37 (.07)***
Mediation effect
Via job control	.03 (.01)*	.04 (.02)	.04 (.01)*	.06 (.02)**	.04 (.02)*	.06 (.03)*	.02 (.01)*	.30 (.01)*	.02 (.01)*	.02 (.01)*
Via job demand	−.01 (.01)	−.01 (.01)	−.03 (.01)*	−.03 (.01)**	−.03 (.01)*	−.03 (.02)*	−.03 (.01)**	−.04 (.01)**	−.03 (.01)**	−.03 (.011)**
Control variables
Effective communication	−.12 (.04)**	−.15 (.05)***	−.12 (.04)***	−.16 (.05)**	−.12 (.05)**	−.16 (.05)***	−.11 (.04)**	−.15 (.05)***	−.11 (.04)*	−.15 (.04)**
Daily working hours	.00 (.00)*	.00 (.00)*	.00 (.00)*	.00 (.00)*	.00 (.00)	.00 (.00)	.00 (.00)***	.00 (.00)**	.00 (.00)	.00 (.00)
Working experiences	.03 (.03)	.06 (.04)	.05 (.03)	.08 (.04)*	.05 (.04)	.08 (.04)*	.04 (.03)	.07 (.03)	.04 (.04)	.07 (.04)
Daily number of colleagues work with	.04 (.03)	.06 (.03)	.04 (.03)	.06 (.03)	.04 (.03)	.06 (.03)	.04 (.03)	.06 (.03)	.04 (.03)	.06 (.03)
Daily number of leaders work with	−.02 (.04)	.02 (.04)	−.04 (.04)	.01 (.04)	−.04 (.04)	.01 (.05)	−.04 (.04)	.01 (.04)	−.04 (.04)	.01 (.05)
Daily number of departments work with	−.08 (.04)	−.03 (.04)	−.08 (.04)*	−.03 (.04)	−.08 (.04)	−.03 (.04)	−.07 (.04)	−.01 (.04)	−.07 (.04)	−.01 (.04)
Daily commuting time	.01 (.03)	.03 (.03)	.01 (.03)	.03 (.03)	.01 (.03)	.03 (.04)	.02 (.03)	.04 (.03)	.02 (.03)	.04 (.04)
Management	−.02 (.07)	.01 (.08)	−.02 (.07)	.01 (.08)	−.02 (.07)	.01 (.07)	−.02 (.07)	.01 (.08)	−.02 (.07)	.01 (.07)
Research	−.01 (.08)	.02 (.08)	−.02 (.08)	.00 (.08)	−.02 (.08)	.00 (.09)	−.02 (.08)	.01 (.08)	−.02 (.08)	.01 (.08)
Service	−.09 (.08)	−.04 (.09)	−.10 (.08)	−.06 (.09)	−.10 (.08)	−.06 (.09)	−.10 (.08)	−.05 (.08)	−.10 (.08)	−.05 (.09)
Marketing	−.01 (.07)	−.07 (.07)	−.01 (.07)	−.06 (.07)	−.01 (.07)	−.06 (.07)	−.02 (.07)	−.07 (.07)	−.02 (.07)	−.07 (.07)
Other	−.12 (.09)	−.09 (.10)	−.13 (.09)	−.10 (.10)	−.13 (.10)	−.10 (.10)	−.11 (.09)	−.08 (.09)	−.11 (.10)	−.08 (.10)
Position levels	.00 (.05)	.01 (.05)	−.01 (.05)	−.01 (.05)	−.01 (.05)	−.01 (.06)	.00 (.05)	.01 (.05)	.00 (.05)	.01 (.05)
Government	.15 (.21)	.18 (.22)	.13 (.21)	.15 (.22)	.13 (.21)	.15 (.20)	.14 (.20)	.16 (.21)	.14 (.19)	.16 (.20)
Public institutions	.02 (.13)	.08 (.14)	.00 (.13)	.06 (.14)	.00 (.09)	.06 (.09)	.01 (.13)	.06 (.14)	.01 (.09)	.06 (.09)
Foreign‐funded enterprise and joint venture	.07 (.15)	.01 (.15)	.05 (.15)	−.01 (.15)	.05 (.11)	−.01 (.11)	.06 (.14)	.00 (.15)	.06 (.11)	.00 (.11)
State‐owned enterprise	−.08 (.15)	.03 (.15)	−.10 (.15)	.01 (.15)	−.10 (.11)	.01 (.11)	−.10 (.14)	−.01 (.15)	−.10 (.11)	−.01 (.11)
Private enterprise	−.02 (.14)	.03 (.15)	−.03 (.14)	.02 (.14)	−.03 (.10)	.02 (.10)	−.03 (.13)	.01 (.14)	−.03 (.10)	.01 (.10)
Age	.01 (.03)	.02 (.04)	−.01 (.03)	.01 (.04)	−.01 (.03)	.01 (.04)	.01 (.03)	.02 (.03)	.01 (.03)	.02 (.04)
Gender	.02 (.05)	.00 (.05)	.01 (.05)	−.01 (.05)	.01 (.05)	−.01 (.06)	.01 (.05)	−.02 (.05)	.01 (.05)	−.02 (.06)
Education	.04 (.04)	−.04 (.05)	.05 (.04)	−.04 (.05)	.05 (.05)	−.04 (.05)	.06 (.04)	−.02 (.04)	.06 (.05)	−.02 (.05)
Married, have a child or children	−.12 (.09)	−.06 (.10)	−.14 (.09)	−.08 (.10)	−.14 (.09)	−.08 (.10)	−.15 (.09)	−.09 (.09)	−.15 (.08)	−.09 (.09)
Single, no child	−.02 (.09)	.12 (.10)	−.04 (.09)	.10 (.10)	−.04 (.09)	.10 (.10)	−.04 (.09)	.10 (.10)	−.04 (.09)	.10 (.09)
Single, have a child or children	−.36 (.37)	−.39 (.39)	−.49 (.37)	−.58 (.39)	−.49 (.23)*	−.58 (.19)**	−.38 (.36)	−.41 (.38)	−.38 (.20)	−.41 (.21)
Days of starting work after Chinese New Year	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)	.00 (.00)
WFH training	.07 (.06)	.09 (.06)	.05 (.06)	.06 (.06)	.05 (.06)	.06 (.07)	.05 (.05)	.06 (.06)	.05 (.06)	.06 (.06)

Moderating role of social support (hypotheses 4a and 4b tests)

Tables 5 and 6 present the results of the moderating analysis of social support by applying hierarchical regressions. The results from the first four regression models consider the direct impact of WFH, job control, job demand and social support on self‐reported job performance as benchmark (Models 1–4 in Tables 5 and 6). Models 5–7 test the moderating effect of employers' social support on the relationships between job control, job demand and social support with job performance. We initially find that the social support is significantly positively related to satisfaction with quality (0.16, p < 0.001) and productivity (0.27, p < 0.001). Toward the moderating effect of employers' anti‐epidemic policy, we find the interaction terms of job demand*social support to be only significant on the regressions on satisfaction of productivity (−0.07, p < 0.05). That is, hypothesis 4a is supported.

Overall, the results of testing the hypotheses are shown in Table 7 and Figure 2.

Table 7.

Results of hypotheses

Hypotheses	Findings	Accept/Reject
H1a: Employees who work from home are more satisfied with their job performance	Significance only can be seen in terms of Quality (8.83***) (Evidence from Table 3)	Partly accept
H1b: Employees who work from home are less satisfied with their job performance	Significance only can be seen in terms of Productivity (−3.1**) (Evidence from Table 3)	Partly accept
H2: Job demand, at least in part, negatively mediates the relationship between WFH and job performance	Job demand negatively mediates, in part, between the WFH and the job performance (Productivity: .02*, 12.5%; Quality: .14***, 23.72%) (Evidence from Table 4)	Accept
H3: The relationship between WFH and job performance is mediated, in part, by job control	Job control negatively mediates, in part, between the WFH and the job performance (Productivity: .03**, 15.78%; Quality: .08***, 14.28%) (Evidence from Table 4)	Accept
H4a: Social support negatively moderates the relationship between job demand and job performance	Interaction term job demand*social support is significant on the regressions on satisfaction of productivity (−.10*). (Evidence from Tables 5 and 6)	Partly accept
H4b: Social support positively moderates the relationship between job control and job performance	Non‐significance (Evidence from Tables 5 and 6)	Reject

*p < 0.05; **p < 0.01; ***p < 0.001.

Figure 2.

The hypotheses results presented in the conceptual framework

Discussion and conclusion

In responding to the inconsistent findings on the impact of WFH on job performance, the present paper found that WFH helps promote job performance in terms of quality but leads to poor job performance in terms of productivity, which indicates that WFH may not always play an ‘either‐or’ (positive or negative) role, as previous theories suggest. To explore the causal mechanism underpinning the findings, based on the JDCS model, we found that WFH affects job performance via job demand and control path, moderated by social support, which indicate that WFH leads to flexibility, and employees have more autonomy to work at any timepoint per day to finalize their job. They usually choose the timepoint to conduct work when they have a desirable working condition, consequently cultivate focus, concentration and creativity (Hunter 2019). Accordingly, job quality can be enhanced. Despite a good job quality, WFH employees devote higher job demand. Thus, it is not conducive to job productivity than WFO employees. In addition, we found the positive moderating role of social support from organizations to enhance job performance during epidemic crisis.

Theoretical implications

The present paper aims to contribute in several ways. Our study extends the JDCS model under the context of COVID‐19 by investigating whether WFH can render the change in job control and job demand and exert influence on employees' job performance with the moderating effect of employers' support. The JDCS model can also help explain why WFH plays a mixed role to affect job performance. Prior studies have mainly qualitatively discussed changes to the way that individuals work during the COVID‐19 pandemic (Wang et al. 2021), the advantages and disadvantages of enforced WFH (Hallman et al. 2021; Purwanto 2020), ICT functions that enable to offer affordance to satisfy WFH targets (Waizenegger et al. 2020), and the way to provide a resource for WFH (Hafermalz and Riemer 2021). Research that indicates why WFH can affect employees' work‐related outcomes, particularly with empirical evidence, is limited. By applying a sample collected in China, we investigated two paths (i.e. job demand and job control) and a boundary condition (support) of the relationship between WFH and job performance.

Our results show that job control and job demand positively mediate the relationship between WFH and job performance. The increased job control and decreased job demand by applying WFH can be considered one of the main reasons WFH helps enhance job quality. This finding is notable because this study tends to clarify the mixed mechanism that WFH affects work‐related outcomes from the perspective of job characteristics and provides a theoretical framework. In terms of job productivity, we find that the increased job control and decreased job demand trade off the negative effect of WFH on productivity. Therefore, when explaining why WFH compared with WFO varies in job performance, the verified mediating effect of job control and job demand underpinned by the JDCS model can only account for job quality enhancement, rather than sufficiently support why WFH lowers job productivity.

The present paper also articulates the specific job control (‘talking right’ and ‘work rate’) and job demand (‘a long time of intense concentration’ and ‘hecticness of the job’) items are vital factors in performance enhancements. On the basis of such findings, we can presume that the ‘talking right’ enhanced by WFH implies that the enforced ‘physical distance’ may shorten the ‘power distance’ inscribed in hierarchical structure, because ICT enables communication flattening information transmitting in traditional stratified management. Reciprocally, such physical distance reduces redundant commands from managers, and workplace distractions trigger WFH employees to have more autonomy on ‘working rate’. Thereafter, in the wake of alleviations on ‘a long time of intense concentration’ and ‘hecticness of the job’, performance is enhanced.

Furthermore, we applied entropy balance matching, a method that has been regarded with more advantages for controlling self‐selection bias in quasi‐experiment research. Future studies could also adopt entropy balance matching to control self‐selection from process control, especially in the crisis context.

Empirical and managerial implications

Empirically, post COVID‐19, WFH may become a vital HRM strategy. According to the Gartner CFO Survey (2020), 74% of companies plan to shift some of their employees to remote working temporarily. Our findings may imply several valuable tips for organizational employers and employees if one wants to accommodate employees to WFH for the long term. We suggest that sustained and pragmatic WFH policy in terms of ‘set working hours’ and ‘taking regular breaks’ should be designed to reduce job demands, such as ‘a long time of intense concentration’ and ‘hecticness of the job’. Furthermore, employers may leave employees more empowerment on scheduling, enhance the equality among different hierarchy people, and avoid lengthy and discursive commands while working to improve the ‘talking right’ and ‘work rate’ autonomy for employees. In addition, social support is found to be a critical boundary condition between WFH and job characteristics. Thus, it is vital that sound and feasible epidemic policies, such as providing personal protective equipment, a financial sponsored program, psychological counselling and support, are put in place and executed as crucial responsibilities (Shani and Pizam 2009). And finally, employers need to be aware that more resources should be available for increased virtual collaboration needs as WFH has now taken hold and will be around for a long time in the future.

Limitation and future research perspectives

First, even though in the present study we have controlled for a wide range of variables that may potentially relate to job performance, inevitably, it still misses some relevant variables. For example, even though we have involved communication factors under control, technology fatigue may still contribute significantly on change of job demands and subsequently affect job performance (Yang et al. 2021). Second, our dataset is a cross‐sectional one and we asked employees to rate job performance rather than multilevel respondents. The absence of lagged performance data restricts the possibility of examining the long‐term effect of WFH on job performance and relationships between the variables of interest. As already noted, the current sample was collected at the early period of ending epidemic lockdown. By applying the cross‐sectional model, identifying the potential time variance (e.g. honeymoon effect) from the targeted relationship is difficult. Thus, future studies should adopt panel data and compare the present study to test for robustness.

Funding

This work was supported by the National Natural Science Foundation of China (grant number 72102033); Shanghai 2020 Science and Technology Innovation Action Plan (grant number 21692102600); the Fundamental Research Funds for the Central Universities of China (grant number N2206012); the Humanities and Social Science Foundation of the Ministry of Education of China (grant number 21YJC630153); the Social Science Foundation of Liaoning in China (grant number L21CGL013).

Conflict of interest

We declare that we have no financial and personal relationships with other people or organizations that can inappropriately influence our work, there is no professional or other personal interest of any nature or kind in any product, service and/or company that could be construed as influencing the position presented in, or the review of, the manuscript entitled, ‘Working from home vs. working from office in terms of job performance during COVID‐19 pandemic crisis: evidence from China’.

Biographies

Jingjing Qu is an associate professor at Shanghai AI Lab, China. Her research interests include artificial intelligence governance, artificial intelligence technology innovation and well‐being of entrepreneurs.

Jiaqi Yan is a lecturer at School of Business and Administration of Northeastern University. He received his PhD degree from Tongji University and studied as a joint PhD student at the University of Sydney. His research interests include human resource management, hospitality management and entrepreneurship.