‘You’re still on mute’. A study of video conferencing fatigue during the COVID-19 pandemic from a technostress perspective

ABSTRACT

The global social restrictions necessitated by the COVID-19 pandemic has resulted in a dramatic increase in the use of video conferencing for activities such as work, study, and personal relationships. Alongside its many benefits, video conferencing can also have adverse effects on users. Video conferencing fatigue is a commonly cited problem, especially for those individuals forced by COVID-19 to adopt the technology. Drawing from the technostress perspective, this paper examines the causes and consequences of VCF during a pandemic situation. A research model is developed and tested quantitively with data collected from 429 users of common video conferencing tools such as Zoom, Teams, and WebEx. The results suggest the relationship between video conferencing stressors and the outcomes of user satisfaction and continuance intentions, are mediated by video conferencing fatigue. In addition, the strengths of these relationships vary depending on whether video conferencing is mainly used for work, study, or social purposes.

KEYWORDS:

• Video conferencing fatigue
• Zoom
• technostress
• communication efficiency
• COVID-19

1. Introduction

The COVID-19 pandemic has led to significant changes globally in the way people conduct their lives. Different measures have been deployed by the governments to curb the spread of the virus including the closing of borders, bans on travel, social distancing requirements, and periods of strict lockdown. Such public health measures force organisations and individuals to radically amend their activities to adapt to the new situation. Organisations have adopted new modes of operation, the most salient being remote working from home for their employees. Individuals too have changed how they socialise, gain education, and engage with traditional leisure activities.

Adhering to the pandemic health guidelines requires the use of tools enabling communication from a distance. More than ever, video conferencing (VC) applications are being adopted to facilitate traditional human activities. Just before the pandemic, 10 million people attended meetings on Zoom. By the end of April 2020, the number of users rose to 300 million (Morris 2020). Other video conferencing apps have experienced similar growth patterns. Around the world, video meetings are happening daily in different collaborative settings such as business teams, classrooms, and family meet-ups. Even as social distancing restrictions ease as the majority of people become vaccinated, it is likely VC will become the ‘new normal’ for many traditional social activities. However, this sudden and dramatic rise in the usage of a technology unfamiliar to many, requires more research to gain insights about its characteristics, the associated benefits and downsides, so that society can continue to function in a healthy manner. During a short period of time, people have had to change their daily habits, learn new tools, and get used to meetings through videos. People who previously did not use VC now must learn to use them. This change is not necessarily voluntary, but mandated in order to stay safe in the pandemic situation.

While platforms such as Zoom, WebEx, and Teams have allowed many businesses to continue to operate and people to socialise during lockdown, it has also been widely reported that the extensive use of VC has taken a toll on users (Carillo et al. 2020; Hacker et al. 2020; Waizenegger et al. 2020). One of the most commonly cited problems is videoconferencing fatigue (VCF), which is defined as the degree to which people feel exhausted or tired attributed to engaging in VC (Bennett et al. 2021). In the mainstream media, VCF is more commonly referred to as ‘Zoom fatigue’. While the problem of VCF is well documented in mainstream media outlets, scientific research is only beginning to reveal its causes and consequences. This lack of knowledge is largely due to the VCF problem only gaining prominence in 2020 with the forced work-from-home policies mandated by COVID-19.

Motivated by the COVID-19 mandated adoption of technology, the purpose of this current study is to better understand the causes and consequences of VCF. Specifically, we draw from theories of technostress (Ayyagari, Grover, and Purvis 2011; Califf and Sarker 2020; Tarafdar et al. 2020) to provide one perspective as to how VC engenders fatigue, and its implications for VC user satisfaction, communication efficiency, and intentions to continue using VC applications. Extrapolating from our findings, we also offer recommendations to VC users and application developers on how to reduce or prevent VCF.

2. Theory development

2.1. Video conferencing fatigue

One pertinent question asked is whether VCF is any different to the general fatigue which people would experience during their normal working day. Adopting a multilevel modelling investigation, Bennett et al. (2021) conclude that VCF is indeed different to general work fatigue, and thus represents a unique construct worthy of study. Specifically, VCF tends to be experienced closer to the VC meeting, whereas work fatigue typically emerges at the end of the day. When compared to meetings held through other media, such as phone calls and written communication, VC meetings were again found to be more fatiguing (Shoshan and Wehrt 2021).

Different scientific disciplines have begun to investigate VCF. In terms of the causes, one stream of enquiry has focused on how users engage with the salient features afforded by VC platforms. Having the camera switched on creates pressure on the individual to maintain a professional looking background in their home, which manifests as greater fatigue (Shockley et al. 2021). Relatedly, the self-view feature, a default setting in most VC platforms, contributes to fatigue especially in females who are generally more concerned by their appearance (Fauville et al. 2021). Having the microphone constantly on is associated with higher levels of post meeting VCF (Bennett et al. 2021), possibly because of the millisecond delay in virtual verbal responses which necessitate our brains to work harder (Johnson 2020). Qualitative research also reveals how the social affordances of VC platforms enhances perceptions of fatigue (Waizenegger et al. 2020). The timing of VC meetings also plays a role in VCF. Neuroscientific research has found that back-to-back VC meetings with no breaks, which has become normal for many remote workers, escalates stress and mental exhaustion (Microsoft 2021). When VC meetings are held around midday, workers reported feeling less VCF, but report greater VCF when they participate in meetings in the late afternoon (Bennett et al. 2021).

In contrast to the causes, the consequences of VCF have received less attention in the emerging scientific literature, possibly because VCF is often considered to be a dependent variable in on itself. Perhaps unsurprisingly, studies that have considered VCF consequences find it is associated with poorer meeting performance (Shockley et al. 2021). People also end up paying less attention when VCF emerges, and can engage in multitasking, but work performance may not be affected (Singh Chawla 2021).

We conclude from our review of the existing VCF literature that very few empirical studies have considered the follow-on consequences of VCF. Studies on the effects of ICT use during COVID lockdown do suggest negative consequences for the user's wellbeing as well as other problems (Kayis et al. 2021; Laato et al. 2020; Panisoara et al. 2020; Waizenegger et al. 2020), but VCF has not been a central construct in these studies. We address this knowledge gap by investigating how VCF relates to user satisfaction, communication efficiency, and intentions to continue using VC applications. Additionally, the users’ psychological perceptions of VC platforms have yet to be unpacked to reveal when and how engagement with VC platforms generates fatigue and the follow-on consequences. To provide this insight, we draw from the stress and technostress literatures.

2.2. Technostress

Stress was originally conceptualised as a biological reaction to challenging stimuli (Selye 1956). Contemporary theoretical paradigms now consider stress to be a process whereby a person constantly appraises their environment and whether it exceeds their resources and jeopardises wellbeing (Lazarus and Folkman 1984). Appraisal is a key element for understanding stress-relevant transactions. Emotional stress is dependent on the contrast between actual expectancies and the significance and outcome of a specific encounter (Lazarus and Folkman 1984). Differences in appraisal mechanisms explain why people experience differences in the valence, intensity, and duration of stress in environments that are objectively equal. For example, a complicated set of features on a VC platform could be appraised as stressful for one individuals, as the variance between expectation and outcomes is large, but not as stressful for another individual with different expectations. In contrast to the initial conceptualisations which did not consider perceptions of stress to be important, perception or situational awareness are central in this process-based view of stress (Fischer, Reuter, and Riedl 2021). Aligned to the process-based view is the concept of ‘stressors’, the sources of stress formed through the ongoing relationship between an individual and their environment, and ‘strain,’ which refer to the adverse outcomes related to stressors (Cooper, Dewe, and O’Driscoll 2001; Lazarus and Folkman 1984).

Technostress is defined as a phenomenon of stress experienced when people engage with ICT (Ayyagari, Grover, and Purvis 2011; Pirkkalainen et al. 2019; Tarafdar et al. 2020). For example, a malfunctioning ICT system has been shown to generate significant stress in the user (Riedl et al. 2012). Extending initial perspectives of technostress, Riedl et al. (2012, 18) suggests the indirect interactions, the ‘ … perceptions, emotions, and thoughts regarding the implementation of ICT in organisations and its pervasiveness in society in general’ should also be captured in a technostress definition. The perception addition to technostress is important for the current study, as the mandated use of VC platforms as a result of the pandemic may be associated with enhanced perceptions of stress and strain.

The transactional theory of stress (Lazarus and Folkman 1984) is the central theoretical foundation underpinning technostress research (Grummeck-Braamt et al. 2021). This view considers stress as a phenomenological process reflected in the relationship between stressors and strain (Fischer and Riedl 2022) where the strain experienced can be psychological (e.g. Maier et al. 2015; Pflügner, Maier, and Weitzel 2021; Whelan, Golden, and Tarafdar 2022) or physiological (Riedl et al. 2012; Tams et al. 2014). While there are other stress theories relevant to studies of technostress, such as coping theory (Carver, Scheier, and Weintraub 1989) and cybernetics (Edwards 1992), these approaches focus on explaining how people cope and regulate stressful situations, which is not directly relevant to the current study. Our purpose here is to understand the process by which the stressors associated with VC platforms are related to the strain of fatigue and follow-on outcomes. Below, we describe the stressor-strain-outcome (SSO) model (Koeske and Koeske 2010), a particular instantiation of the transactional theory of stress, as an appropriate theoretical lens to study technostress and VCF.

Transactional theory views two core components of the technostress process – technostressors and technostrains. Technostressors are defined as the events, properties of events or stimuli encountered by individuals that create stress (Tarafdar, Tu, and Ragu-Nathan 2010). Early technostress literature highlight complexity, pattern, overload, invasion, uncertainty, and insecurity as common stressors (Ragu-Nathan et al. 2008; Tarafdar et al. 2007). More recent works also validate contemporary digital stressors, such as conflicts (i.e. how ICT blurs the boundaries between work and life domains) and social environment (i.e. how ICT creates unwanted social norms and expectations), as having stress potential for ICT users (Fischer, Reuter, and Riedl 2021). Technostrains are the adverse responses to these technostressors (Laumer and Maier 2021; Maier et al. 2012), of which fatigue has been identified as a pertinent strain in a number of technostress studies (Bright and Logan 2018; Dhir et al. 2018; Ravindran, Yeow Kuan, and Hoe Lian 2014; Whelan, Najmul Islam, and Brooks 2020). For this study, VCF is conceptualised as the strain associated with using VC. The SSO model posits a relationship between stressors and outcomes, where strain acts as a mediating factor explaining negative outcomes (Cao et al. 2018; Tandon et al. 2021). We have grounded our work in the SSO model for a number of reasons. Firstly, the model has helped advance our understanding of technostress in multidisciplinary contexts including the workplace, education, and social media research (Cao et al. 2018; Tandon et al. 2021; Whelan, Najmul Islam, and Brooks 2020). Secondly, originating from the field of occupational health psychology, SSO provides a simple yet effective framework to explain the dynamics of technostress and its impact on individuals outcomes (Tandon et al. 2021; Q. Wang and Yao 2021). Thirdly, the SSO model fits the objective of this study, which is to examine the relationships between VC technostressors, fatigue, and adverse outcomes.

In this study, we focus on the technostressors of complexity (i.e. VC platforms perceived as difficult to use) and pattern adaptation (i.e. VC platforms forcing users to modify their desired habits and routines) as emerging research on VCF suggests users experience such issues when forced to adopt VC (Hacker et al. 2020; Waizenegger et al. 2020). While we agree with prior research which highlights that many technostressors are relevant and can potentially be part of future investigations (Fischer, Reuter, and Riedl 2021), we initially focus on these two antecedents as they are most relevant to the characteristics of VCF. Due to the COVID-19 pandemic, people have to quickly shift from direct communication to VC. During this process, they must learn to use new tools and must adapt with the new modes of operating. Thus, the two stressors of complexity and pattern adaptation are most relevant for our study.

2.3. Technostress and video conferencing fatigue

The connection between stress and fatigue is well established in the general stress literature (Klusmann et al. 2020; Lepine, Lepine, and Jackson 2004), a relationship that also carries through to technostress studies. Fatigue has been defined in these technostress studies as feeling one's emotions being overextended and feelings of being exhausted due to stressful situations (Maier et al. 2012), or feeling low on energy and unable to concentrate on important tasks after a period of heavy ICT use (Whelan, Islam, and Brooks 2020). Fatigue can also be expressed through feeling annoyed, overwhelmed, disappointed, and frustrated (Ravindran, Yeow Kuan, and Hoe Lian 2014). There are a number of reasons why technostress engenders feelings of exhaustion. Recent studies of SNS stress suggest sleep quality is adversely affected which results in sleepiness carrying through to the next day (Salo, Pirkkalainen, and Koskelainen 2019; van der Schuur, Baumgartner, and Sumter 2018). For employees, the relationship between the characteristics of workplace technology and fatigue is explained by concerns such as work-family conflict, work overload, and role ambiguity (Ayyagari, Grover, and Purvis 2011).

The consequences of technostress are many. In the organisational context where the use of technology is largely compulsory, technostress leads to role conflict, role stress (Tarafdar et al. 2007, 2011), reduced productivity (Tarafdar et al. 2007, 2010, 2011), reduced job satisfaction (Ragu-Nathan et al. 2008; Suh and Lee 2017; Tarafdar et al. 2011), decreased job commitment (Ragu-Nathan et al. 2008; Tarafdar et al. 2011), work-family conflict (Benlian 2020), and lower wellbeing (Califf and Sarker 2020). In a non-working context where technology usage is voluntary and more for leisure purposes, technostress also arises and leads to consequences such as reduced user satisfaction (Maier et al. 2012, 2014; Suh and Lee 2017), lower intentions to continuing using the technology (Maier et al. 2012, 2014; Masood et al. 2020), exhaustion (Maier et al. 2014, 2015), and burnout (Panisoara et al. 2020).

As VC is used commonly for both working and non-working purposes, in the present study we examine the consequences related to the user experience. Specifically, we consider users’ intention to continue using VC and their satisfaction with the VC experience. In addition, we consider if VCF has an impact on users’ communication efficiency when they engage in meetings through VC. It is well documented that stress has a negative impact on communication (Ledermann et al. 2010). When stressed, people become angry or frustrated, withdraw from other people, hold their ideas rather than express it, all of which lead to misunderstandings and communication problems. Stress also reduces the information processing capability of the brain (Stress Management Institute 2017) which can further reduce communication efficiency.

3. Research model and hypotheses

In theorising the links between technostress, VCF, and outcomes, we ground our research model in the SSO framework (Koeske and Koeske 2010). Previous IS studies have applied the SSO framework to explain how the stresses associated with ICT use translate into life outcomes (Cao et al. 2018; Reinecke 2009; Whelan, Islam, and Brooks 2020). As depicted in Figure 1, we apply the SSO framework to explain how VC technostressors lead to VCF and ultimately lower usage intentions, satisfaction, and communication efficiency.

Figure 1. The research model.

The dramatic rise of VC usage has been driven by the sudden onset of the COVID-19 pandemic. With mandated remote working, people who had limited engagement with VC had to quickly place the technology at the heart of their work, study, and social communications. This dramatic change in communication comes with two main issues. First, people had limited time to learn how to effectively use this new technology. VC platforms come with many features (e.g. share screen, scheduling, polls, breakout rooms) which can be difficult to learn. An analysis of 3.2 million tweets related to VC use during the pandemic reveals that VC platforms lacked the features users needed and people frequently struggled to set up and configure VC platforms (Hacker et al. 2020). This difficulty is captured by the techno complexity stressor (Fischer, Reuter, and Riedl 2021; Tarafdar et al. 2007, 2015) which has been found to be correlated with fatigue (Maier et al. 2012, 2015). Additionally, the dramatic increase in VC platforms has forced VC providers to frequently update and modify features. While such changes are often designed to enhance the user experience, frustration often results when users have to relearn how to execute certain actions in a system that were once familiar to them (Lazar, Jones, and Shneiderman 2007). Therefore, we hypothesise;

H1: VC techno complexity is positively associated with VCF.

The adoption of VC has altered peoples’ daily patterns of work, collaboration, and communication. To compensate for the lack of face-to-face interactions, some organisations have held more ‘Zoom calls’, which has had the unintended consequence of creating fatigue and annoyance as people perceive them as being unnecessary and intrusive (Hacker et al. 2020; Waizenegger et al. 2020). As a result, some employees have adapted their daily routine to include frequent mental breaks in order to cope with the onset of VCF (Hacker et al. 2020). Remote workers also have to balance a dramatically altered home environment (e.g. home schooling, competing for internet access and office space with partners) with VC meetings, which enhances stress and frustration (Carillo et al. 2020). In studies of social networking sites, the change in pattern brought about by platform use was strongly linked to feeling tired (Maier et al. 2015). Building on these insights, the use of VC to comply with social distancing requirements forces users to adapt their habits and behaviours which can trigger VCF.

H2: VC pattern adaptation is positively associated with VCF.

When people become fatigued, they develop the intention to change their behaviour and the current situation (Chen et al. 2019; Laumer and Maier 2021). This often results in the discontinuance of the stressful behaviour (Ajzen 1991). VCF is a strain which frustrates users and reduces their overall experience of VC. As a result, users may try to avoid engagement with VC by using alternative communication tools, reducing usage frequency, or stop using VC altogether. Indeed, ICT induced fatigue plays a role in explaining why people intend to switch to an alternative system when they become overloaded (Chen et al. 2019). Discontinuous usage intention has been confirmed as a consequence of technostress in other communication tools such as social networking sites (Maier et al. 2012, 2014, 2015). People may have no choice but to engage with VC when remote working, but due to their experience of VC fatigue, may intend to reduce their VC usage once the pandemic passes and VC becomes more of a choice. Consequently, we hypothesise the following:

H3: VCF negatively impacts VC continuous usage intentions.

VCF is a strain that leads to users reporting a negative experience with the system (Carillo et al. 2020; Hacker et al. 2020; Waizenegger et al. 2020). In consumer research, it is well documented that a negative experience results in lower customer satisfaction (Bui, Krishen, and Bates 2011; Tsiros and Mittal 2000). This relationship transcends to the use of ICT. A poor experience leads to users reporting low satisfaction with tools such as social networking sites (Nawaz et al. 2018), websites (Sørum, Andersen, and Vatrapu 2011), and workplace ICT systems (D’Arcy et al. 2014). In terms of fatigue in particular, a number of studies report how digital fatigue reduces the user’s satisfaction with a particular digital technology (Lee, Lee, and Suh 2016; Maier et al. 2012, 2014). While most IS fatigue studies focus on social networking sites and mobile instant messaging applications, other studies show that the onset of screen fatigue in a proofreading experiment also results in lower user satisfaction (Park et al. 2019). If users experience fatigue from using VC, it is likely they will attribute the source of this discomfort to the VC platform, which would result in a lower opinion of VC. Likewise, when fatigued from VC, users will make more errors when using the VC features. Thus, they may use the tools less efficiently which consequently lower their satisfaction levels with the tools. Consequently, we hypothesise the following:

H4: VCF negatively impacts user’s satisfaction of the VC tool

As VC tools are most often used for the collaborative and conversational purposes, the efficiency of communication is important to users. In the experiment of Alexander et al. (2012), the modality of communication had an impact on communication. Compared to direct communication, video conversation was less efficient as people must talk more to each other to achieve the same result. However, this experiment did not point out the causes of reduced communication efficiency. VCF could have an impact on communication, the reason comes from the study of the impacts of strain and fatigue on communication ability. According to Stress Management Institute (2017), strain has a negative impact on communication in different ways. First, when strained, people could easily get angry or frustrated which consequently makes it difficult for them to choose words carefully or express themselves in an appropriate manner. Also, strain makes a person withdraw from other people; thus, they may hold their ideas rather than express it. The particular strain at the centre of this study is fatigue. In the airline industry, where the efficient communication of safety information is critical, it has been found that communication suffers drastically when maintenance crews are fatigued (Wang and Chuang 2014). Message fatigue has been a prominent research theme in health behaviour research with relevant studies reporting on its associations with message avoidance (So, Kim, and Cohen 2016) and disengagement (Kim and So 2017). Neuroscience scholars have also found that activity in the brain’s executive control network is low when a person is fatigued, which ultimately hampers information processing capabilities (Togo et al. 2015). Drawing from these studies, we suggest that VCF is associated with less efficient communication in VC meetings.

H5: VCF negatively impacts VC communication efficiency

4. Research methods

The research model was tested using survey data gathered through the Qualtrics survey platform. The initial survey was pilot tested with five VC users and two academics experienced in survey research. Some minor changes to the question phrasing were made as a result of this process. In total, 429 completed responses were received for the final survey. Of these, 285 participants were recruited through Facebook and Zalo (a prominent Vietnamese social network). The original version of the survey was in English and was translated into French and Vietnamese versions. Respondents had the option to complete the survey in the language they are most comfortable with. To ensure our research model was sufficiently powered, a further 144 participants were recruited through the MTurk platform. A comparison of responses between the Facebook/Zalo and MTurk recruited samples revealed no significant differences were present. To be included in the sample, participants had to confirm they were at least 18 years old and currently use VC platforms at least once a week. The demographic of this sample is provided in Table 1.

Table 1. Respondent demographics.

Demographics	Item	Percent
Gender	Male	41
	Female	59
Age	Less than 25	15.6
	25 – under 30	31.7
	30 – under 35	24.4
	More than 35	28.3
VC usage purpose	Job & business	58.2
	Study & training	22.8
	Personal (family, friend, etc.)	19.0
VC usage frequency	Every day	17.5
	One to several times per week	42.1
	One to several times per month	30.2
	Less frequent than every month	10.2
Estimated length of a typical VC session	Less than 30 min	13.7
	30 min–1 h	49.1
	1–2 h	27.7
	2–3 h	7.7
	More than 3 h	1.8

To determine if the sample size of 429 was adequate, we used the G-power sample size test (Faul et al. 2009). Since our model has four predictors, using G-power with an effect size of 0.05, alpha of 0.05, and a power of 0.95, the minimum sample size needed was 191. Thus, we can conclude that the sample size of 429 provides sufficient power to support the statistical findings.

All measurement items were taken from prior validated studies. A five-point Likert scale used to measure the key constructs of the research model. All survey items are provided in Appendix 1. The items for techno complexity (e.g. I need a long time to understand and use VC tools) were adapted from Maier et al. (2015) with the items for pattern adaptation (e.g. Due to VC, I am forced to adapt my daily schedules) adapted from Maier et al. (2012). A slight rephrasing of the survey question was required as these previous studies focused on social networking sites. VCF items (e.g. I feel drained from activities that require me to use VC) were adapted from the strain scale used by Moore (2000). The measures for continuance usage intentions (e.g. I intend to continue using VC rather than discontinue their use), user satisfaction (e.g. Overall, I am satisfied with using VC) and communication efficiency (e.g. My time in VC meetings is used efficiently) were adapted from Venkatesh, Thong, and Xu (2012), Maier et al. (2012), and Davison (1999) respectively.

Following previous studies of digital fatigue (Bright and Logan 2018; Dhir et al. 2018; Ravindran, Yeow Kuan, and Hoe Lian 2014), age, gender, purpose of VC use, the intensity of VC use, and perceptions of media richness were included in the model as control variables.

5. Analysis

The analysis was conducted using the PLS-SEM approach with SmartPLS software version 3.3.9 (Ringle, Wende, and Will 2015). The PLS approach to structural equation modelling (SEM) enables researchers to estimate complex models with many constructs, indicator variables and structural paths without imposing distributional assumptions on the data (Hair, Ringle, and Sarstedt 2011). PLS is an appropriate methodology when the goal of the study is both to evaluate the validity of a research model, and to test new theoretical relationships within that model (Hair et al. 2017).

We evaluated convergent validity by examining item loadings, composite reliabilities (CR), and average variance extracted (AVE) values. Regarding item loadings, it is recommended that values be at least 0.7 to be acceptable (Fornell and Larcker 1981). Based on this criterion, one item from the continuance intention construct was removed. Keeping this item did not change the results. However, we preferred to remove it following the 0.7 threshold guideline.

The CRs being above 0.8 and AVE values exceeding 0.5 further support satisfactory convergent validity. The loadings, CRs and AVEs are shown in Appendix 1. We evaluated the discriminant validity by comparing the square roots of AVE values to the inter-construct correlations. Table 2 shows the correlation matrix with the square root of AVE values presented diagonally. As can be seen from the table, the square roots of the AVE values for the variables are consistently greater than the off-diagonal correlation values, suggesting satisfactory discriminant validity between the variables. Appendices 2 and 3 shows that all items have cross loading coefficients lower than the factor loading on their respective assigned latent variable, suggesting that discriminant validity on the item level is met for all the constructs.

Table 2. Correlations between latent variables (square root of AVEs in the main diagonal).

	1	2	3	4	5	6	7	8	9	10	11
1. Age	1.00
2. Communication	0.09	0.90
3. Continuance	0.11	0.45	0.88
4. Freq	−0.02	−0.09	−0.22	1.00
5. Gender	0.02	0.08	0.18	−0.07	1.00
6. Media richness	0.08	0.59	0.38	−0.01	0.07	0.80
7. Pattern	−0.20	−0.29	−0.25	−0.06	0.01	−0.39	0.79
8. Satisfaction	0.09	0.47	0.65	−0.11	0.15	0.52	−0.24	0.89
9. Tech complex	0.02	−0.23	−0.25	0.15	0.01	−0.02	0.24	−0.17	0.83
10. VCF	−0.18	−0.37	−0.54	0.01	−0.13	−0.39	0.47	−0.57	0.27	0.85
11. Time	−0.11	−0.14	−0.14	−0.07	0.082	−0.14	0.25	−0.09	0.04	0.20	1.00

Henseler, Ringle, and Sarstedt (2015) propose the heterotrait-monotrait (HTMT) ratio of correlations criterion as a more accurate indicator discriminant validity. HTMT values below 0.90, or the more conservative 0.85 threshold, indicate discriminant validity is present (Henseler, Ringle, and Sarstedt 2015). The highest HTMT value in our data was 0.58, thus discriminant validity is established.

We also investigated possible gender differences in the composite scores of the constructs using t-tests. There were no overall gender differences in the composite score for VCF, media richness, continuance use intentions, satisfaction, and communication efficiency.

To detect possible model misspecification, we examined how well our model fitted the data (Henseler and Sarstedt 2013). To this end, we followed Henseler and Sarstedt (2013) and used standardised root mean square residual (SRMR) statistics. For SRMR, we obtained a value of 0.06. According to Hu and Bentler (1998) SRMR below 0.10, or the more conservative 0.08 level, indicates good model fit. As a result, we conclude that our model exhibited good fit to the data.

To evaluate the risk of common method bias (CMB) (Podsakoff et al. 2003) in our data, we conducted several tests. First, we conducted Harman's (1976) single factor test. We conducted a principal component analysis and found no single construct accounted for most of the total variance. Second, Pavlou et al. (2007) suggest a correlation above 0.9 is an indication of possible CMB. We examined the correlation matrix shown in Table 2 and observed that the correlations ranged from 0.01–0.65. These tests ensure that CMB is not a major concern in our study.

The significance of path coefficients was determined via a bootstrapping procedure by setting the number of bootstrap samples to 5,000 (Hair et al. 2017). Figure 2 shows the results of the structural model test. The model explained 25% of variance in VCF, 38% variance in user satisfaction, 23% in continuance usage intention, and 49% in communication efficiency. Techno complexity had a significant effect on VCF (H1; β = 0.18, p < 0.001), thus H1 is supported. Similarly, pattern adaptation was significantly and positively related to VCF (H2; β = 0.38, p < 0.001), supporting H2. As hypothesised in H3 and H4, VCF was negatively correlated with intentions to continue using VCF (H3; β =  −0.31, p < 0.001) and user satisfaction (H4; β =  −0.35, p < 0.001). However, the hypothesised relationship between VCF and communication efficiency was not supported (H5; β =  −0.07, p > 0.05).

Figure 2. Model results.

Of the control variables, VC use intensity was negatively related to continuance intention (β =  −0.22, p < 0.001) and use satisfaction (β =  −0.10, p < 0.05). Perceived media richness was positively related to continuance intentions (β =  0.29, p < 0.001), use satisfaction (β =  0.55, p < 0.001), and communication efficiency (β =  0.64, p < 0.001). No other control variables were significant related to dependent variables.

5.1. Post hoc analysis

During the data collection process, we received some feedback from participants that their usage behaviour and experience with VC vary significantly depending on the purposes of the meetings. In our survey we requested data on the main purpose of participants’ VC use. The three main reasons people used VC were for work purposes (58% of sample), education (23% of sample), and social purposes (19% of sample). Using the multigroup analysis feature in SmartPLS, we compared the results of the research model for the three groups independently. The multigroup analysis allows to test if pre-defined data groups have significant differences in their group-specific parameter estimates (e.g. outer weights, outer loadings and path coefficients) (Hair et al. 2019). A power analysis reveals that the minimum group size of 54 is needed to detect R² values of around 0.25 at a significance level of 5% and a power level of 80% for multigroup analysis (Hair et al. 2017). As the social purposes group is the smallest with 82 respondents, the minimum sample size requirements for PLS multigroup analysis is met.

The strength of the path coefficient in the relationship between techno-complexity and VCF was significantly stronger in social purposes group when compared to the work purposes group (β =  0.49 vs 12, p < 0.05). In the relationship between pattern adaptation and VCF, the path coefficient strength was significantly weaker for the education purpose group when compared to work purposes group (β =  0.14 vs 0.42, p < 0.05). The negative relationship between VCF and user satisfaction was significantly stronger for education users when compared to work users work purposes group (β =  −0.57 vs −0.16, p < 0.05). All other hypotheses comparisons between the three groups were not significant. The implications of these findings are discussed further below.

6. Discussion

The sudden adoption of VC technology during the COVID-19 pandemic has left many users reporting feelings of exhaustion. Indeed, the phrase ‘Zoom fatigue’ has now entered popular parlance. Drawing from theories of stress, this study examined the antecedents and outcomes associated with VCF.

6.1. Contributions to the literature

We apply the lens of technostress to investigate the processes associated with VCF. Although the phenomenon of technostress has shed light on the strain inherent in using different technologies (Ayyagari, Grover, and Purvis 2011; Ragu-Nathan et al. 2008; Tarafdar, Tu, and Ragu-Nathan 2010, 2015, 2020), to the best of our knowledge, only one other study (i.e. Panisoara et al. 2020) has applied the technostress lens to VC platforms. As VC platforms are now amongst the most widely used technologies across the globe, it is vital for user wellbeing and performance outcomes that the causes and consequences of VCF are better understood. Such insight is needed in order to develop effective interventions combating VCF.

Our research model aligned with the established SSO framework (Koeske and Koeske 2010) which underpins existing technostress studies (Cao et al. 2018; Reinecke 2009; Whelan, Islam, and Brooks 2020). When building the model, we applied the constructs most appropriate to the technology under investigation and the current pandemic situation. The constructs included are techno complexity, pattern adaptation, VCF, user satisfaction, and continuous usage intention. As a consequence of technostress, our model also includes the new construct of communication efficiency. People are now substituting face-to-face communications with VC, thus it is important to understand how the psychological impacts of using VC influence communication efficiency. However, our results suggest communication efficiency is not inhibited by VCF, which opens up further questions as to the specific boundary conditions where the relationship may be evident. Acoustic scholars suggest the millisecond delay in virtual verbal responses necessitates our brains to work harder (Johnson 2020). The lack of perceived reward relative to cost during videoconferencing has also been offered as a primary psychological mechanism of VCF (Lee 2020). However, those arguments are not based on empirical investigations into VCF, but on theoretical arguments. While our empirical data indicates that VCF does not necessarily hinder communication efficiency, future studies need to investigate this link further and substantiate the underlying mechanisms linking VC use to communication efficiency.

As hypothesised, VCF mediates the relationship between the VC stressors of complexity and pattern adaptation and the outcomes of continuance use intentions, and user satisfaction. As such, our study contributes to opening up the ‘black box’ of technostress (Ayyagari, Grover, and Purvis 2011) by revealing how it is that the effects of VC stressors translate to hinder important user outcomes. Our findings also highlight the broad and complicated nature of technostress and research of this phenomenon should separate stressors into those emanating from direct use of the technology itself, such as complexity, and those which indirectly lead to strain, such as having to adapt daily routines to align with the requirements of the technology.

Our study also makes contributions to the emerging literature on VCF. While both VC stressors were significantly related to VCF, pattern adaptation was a stronger predictor (β =  0.38 vs 0.18). This suggests that it is the effects mediated by using VC, rather than the technology itself, that generates strain. People struggle more when they have to adapt with the new situation related to the usage of the technology rather than the technical aspects of the technology itself. A similar finding was reported in relation to social networking sites (Maier et al. 2015).

In addition, our post hoc analysis revealed that the effect of techno complexity on VCF was much stronger for people who mainly used VC platforms for social purposes. This is possibly because people who use VC for work or educational purposes had previous experiences of using VC platforms (online learning in university for example). As explained by the transactional theory of stress (Lazarus and Folkman 1984), such users would have set their expectations of how complex VC platforms are to use at a level somewhat similar, but still lower, than their actual experiences. Social users in contrast may have been forced into using VC platforms for the first time in order to interact with friends and family because of the pandemic mandated social lockdowns. Thus, they may be less experienced in using VC and the appraisal of their expectations are in variance to actual outcomes, hence the stronger effect of the techno complexity stressor on VCF for these users. In contrast, having to adapt one’s pattern because of VC was not such a stressor when the main purpose of VC use was for educational reasons. The effect of the pattern adaptation stressor on VCF was significantly weaker for educational users. This is logical as live classes in university were still scheduled at the same times, even though they moved online. Indeed, many universities recorded their live classes which afforded students flexibility and the ability to adapt VC to their daily schedule, rather than the other way round. For work related users, the effect of pattern adaption on VCF was quite strong (β =  0.42). This finding is consistent with recent qualitative research which suggests that VC is viewed by workers as intrusive as it clashes with their professional and private schedules (Waizenegger et al. 2020). Meetings were typically scheduled when workers were in the physical office but ‘virtual taps on the shoulder’ seem to be more common when working through VC (Waizenegger et al. 2020). Thus, our findings show that the effect of technostress on VCF is not equal and depends largely on the users motivations and needs for using VC platforms.

While there are between group differences in the antecedents of VCF, the deleterious effect of VCF on all our outcomes were similar for the three subgroups. Regardless of the user’s purpose for using VC, negative outcomes emerge once fatigue is experienced. This finding is consistent with previous understandings of technostress in communication technology (Lee, Lee, and Suh 2016; Maier et al. 2012, 2015).

6.2. Implications for practice

Our findings have implications for developers, employers, and users of VC platforms. If VC developers want customers to keep using their tools, they need to identify and modify the stress inducing features. When the VC platform is complex to use, this leads to fatigue and ultimately the user not wanting to engage with the technology further. As the numbers of VC users increased dramatically as a result of COVID-19 lockdowns, developers of VC tools such as Zoom, Teams, and Webex were constantly changing features and the user interface. If these changes are dramatic, for example the share screen feature is moved to a different location, this generates stress, fatigue, and ultimately a poorer user experience. While some amount of stress is inevitable, it is particularly important when people are forced into using digital technology for the first time that system modifications are made in a piecemeal process. Similarly, negative outcomes emerge when VC forces users to change their behavioural pattern. VC tools should allow for greater flexibility to help users cope with the changes in their schedule, habits, and conditions.

Users of VC want to have the best experience with the technology and low levels of VCF. Pattern adaptation was the strongest predictor of VCF in our study. Thus, users could significantly reduce VCF by tackling this stressor. VC can only disrupt a person’s behavioural pattern if they allow it to do so. Workers, as well as their managers, should try to schedule any VC meetings well in advance and ensure the timings take into account work from home routines. In traditional office settings, workers would spend some time walking between different meetings, thus avoiding back-to-back meetings with no breaks. A similar routine can be enabled on some VC platforms where users are prevented from joining a meeting for a period of time after an earlier meeting ends. Our findings suggest this would help reduce VCF and the associated negative outcomes. MS Teams has recently embedded a ‘virtual commute’ feature which ensures the user is not contactable before or after work hours (Microsoft 2021). This is designed to reflect pre-Covid times when workers used their physical commute to transition between the work and home domains. Engaging with such features could assist users in maintaining their pattern and prevent excessive strain emerging.

6.3. Limitations and future research directions

As with all research, our findings are limited in several ways. Firstly, we did not differentiate between different VC platforms. It has been reported in the mainstream media that Zoom provides a higher quality experience over MS Teams (Forbes 2021). Thus, future research could empirically compare the three main VC platforms – Zoom, Teams, and WebEx – to determine if perceptions of media richness vary across these systems and how this influences VCF. Second, our study was correlational in nature and we can only infer causation through theoretical arguments. To validate the causal stress–strain-outcome pathways associated with VC use, future research should adopt experimental approaches. Building on our insights, researchers could manipulate the complexity of the VC platform (e.g. making features like share screen difficult to find) to determine if these interventions affect VCF. Additionally, future studies will find value in employing neuroscientific approaches such as electroencephalogram, electrodermal activity, and eye tracking, to objectively measure stress and fatigue. Thirdly, our study only applied two constructs from technostress theory as these were the most relevant to VC and the pandemic situation. It is likely that recently validated digital stressors, such as conflicts and unreliability (Fischer, Reuter, and Riedl 2021), will also be relevant to VCF, particularly so as VC developers continue to evolve the capabilities of their platforms. Future research should expand upon our model and include these alternative predictors of VCF.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Appendices

Appendix 1

Table A1. Descriptive statistics for variables.

Construct	Item	Mean	SD	Loading
Techno complexity (CR 0.76; AVE 0.69; adapted from Maier et al. 2015)	COMPLEX_1: I need a long time to understand and use VC tools	2.13	0.95	0.79
	COMPLEX_2: I do not find enough time to upgrade my technology skills to use VC tools	2.09	1.00	0.79
	COMPLEX_3: Younger people are better at using VC tools than I am	2.76	1.29	0.74
	COMPLEX_4: I often find VC tools too complex to use	1.99	0.90	0.83
Pattern adaptation (CR 0.70; AVE 0.63; adapted from Maier et al. 2012)	ADAPT_ 1: Due to VC, I am forced to adapt my daily schedules	2.44	1.12	0.83
	ADAPT_2: During VC, I am forced to change my habits	2.77	1.13	0.81
	ADAPT_3: I am forced to adapt my communication patterns to VC	3.19	1.16	0.71
Perceived media richness (CR 0.80; AVE 0.64; adapted from Sheer 2011)	RICH_1: VC allows me to express my feelings fully	2.88	1.01	0.74
	RICH_2: VC allows me to freely say what I intend to say	3.15	0.99	0.75
	RICH_3: VC allows me to explain complex/difficult ideas well	3.09	0.97	0.79
	RICH_4: VC allows me to use multiple cues (e.g. text, audio, visual)	3.58	0.99	0.70
	RICH_5: VC gives me a feel of face-to-face communication	2.91	1.11	0.71
VCF (CR 0.88; AVE 0.73; adapted from Moore 2000)	VCF_1: I feel drained from activities that require me to use VC	2.48	1.00	0.87
	VCF_2: I feel tired from my VC activities	2.56	1.07	0.85
	VCF_3: Using VC is a strain for me	2.19	0.95	0.75
	VCF_4: VC activities stress me out	2.17	0.97	0.77
Communication efficiency (CR 0.88; AVE 0.81; adapted from Davison 1999)	COMMU_1: My VC meetings are typically result orientated	3.55	0.89	0.85
	COMMU_2: My time in VC meetings is used efficiently	3.49	0.82	0.87
	COMMU_3: Issues raised in VC meetings are discussed thoroughly	3.40	0.88	0.87
Continuous usage intention (CR: 0.649; AVE 0.77; adapted from Maier et al. 2012)	CONTI_1: I intend to continue using VC rather than discontinue their use	3.99	0.83	0.72
	CONTI_2: My intentions are to continue using VC than use any alternative communication tool	3.20	1.02	0.72
	CONTI_3: If I could, I would like to discontinue my use of VC (reverse coded)	3.79	1.00	0.41^a
Satisfaction (CR: 0.917; AVE 0.80; adapted from Maier et al. 2012)	SATIS_1: Using VC is a delight	3.46	0.79	0.79
	SATIS_2: Using VC is very pleasing	3.54	0.78	0.80
	SATIS_3: Using VC is very contenting	3.48	0.80	0.81
	SATIS_4: Overall, I am satisfied with using VC	3.69	0.77	0.74

^a Item removed.

Appendix 2

Table A2. Rotated component matrix for independent variables.

Item	Richness	COMPLEX	ADAPT
COMPLEX_4	−0.031	0.834	0.079
COMPLEX_1	0.017	0.799	0.088
COMPLEX_2	−0.043	0.795	0.068
COMPLEX_3	−0.015	0.624	0.137
ADAPT_2	−0.099	0.057	0.832
ADAPT_3	−0.257	0.118	0.744
ADAPT_1	−0.137	0.206	0.679

Appendix 3

Table A3. Rotated component matrix for dependent variables.

	SATISFACTION	VCF	COMMUNICATION
SATISFACTION_3	0.813	−0.250	0.081
SATISFACTION_2	0.801	−0.330	0.176
SATISFACTION_1	0.791	−0.291	0.217
SATISFACTION_4	0.737	−0.333	0.299
CONTINUE_1	0.624	−0.301	0.315
CONTINUE_2	0.623	0.112	0.092
CONTINUE_3	0.502	−0.368	0.190
VCF_1	−0.126	0.865	−0.057
VCF_2	−0.181	0.849	−0.077
VCF_4	−0.223	0.774	−0.254
VCF_3	−0.329	0.749	−0.188
COMMUNICATION_2	0.236	−0.140	0.869
COMMUNICATION_3	0.179	−0.177	0.866
COMMUNICATION_1	0.207	−0.110	0.853