3 Connected Yet Cognitively Drained?
A Mixed-Methods Study Examining Whether Online Vigilance and Availability Pressure Promote Mental Fatigue1
3.1 Abstract
This mixed-methods study investigates whether online vigilance promotes mental fatigue, and whether this effect is greater when under pressure to be available online. Additionally, it examines whether passively sensed smartphone behavior can serve as a digital proxy for online vigilance. Data were collected from 1,315 adult participants, who received 84 experience sampling questionnaires over 14 days, providing 67,762 usable datapoints on individuals’ perceptions of momentary online vigilance, mental fatigue, and availability pressure. Additionally, the smartphone use of 834 participants was passively monitored. Findings revealed both a momentary and lagged association between self-reported online vigilance and self-reported mental fatigue. Availability pressure was not a significant moderator, but did predict mental fatigue directly and indirectly, by promoting online vigilance. We found behavioral smartphone use features were weakly associated with self-reported online vigilance and mental fatigue. Overall, this study provides initial support that online vigilance may play a role in the development of mental health conditions such as burnout via its tendency to promote one of its precursors, mental fatigue.
Keywords: online vigilance, mental fatigue, availability pressure, smartphone, log data, passive sensing, monitoring, experience sampling
3.2 Introduction
Mental fatigue is a subjective state of tiredness characterized by a reluctance for further effort (Hopstaken et al., 2015). Attention towards mental fatigue has increased in recent years as research identifies the state as a key precursor to burnout (Demerouti et al., 2002), a mental health condition that has risen sharply in workers in westernized societies (De Hert, 2020; Edú-Valsania et al., 2022; Naldi et al., 2021). Together with other characteristics, such as reduced cognitive and emotional regulation and mental distancing (Schaufeli et al., 2020), mental fatigue is believed to contribute to feelings of exhaustion and a-motivation (Demerouti et al., 2002; Edú-Valsania et al., 2022). Given the debilitating health consequences associated with it, better understanding of what factors promote mental fatigue is essential to facilitate the development of effective interventions and policies.
Vigilance is a factor known to promote mental fatigue, as protracted periods of vigilance require hard mental work (Warm et al., 2008). For instance, truck driving (Ting et al., 2008), excavating (Li et al., 2019), and monitoring various security video feeds (Hodgetts et al., 2017) are mentally fatiguing activities because they demand paying constant attention to one or more stimuli. This study questions whether in contemporary Westernized societies a new type of vigilance may have emerged alongside the culture of ubiquitous digital connectivity, namely online vigilance, that may potentially contribute to rising burnout rates by promoting mental fatigue.
Online vigilance, then, refers to a state in which one is constantly attentive to the online world (Reinecke et al., 2018). Characterized by cognitive pre-occupation and a constant monitoring of and reactivity to the online world, we posit that online vigilance is a mentally taxing state; after all, performing day-to-day activities while simultaneously keeping constant tabs on what happens online, repeatedly checking various digital streams for incoming messages, responding immediately to incoming notifications, and making oneself available online at all hours of the day, may come with cognitive costs, including the costs of increased cognitive load (e.g., Gilbert et al., 2023) and of multitasking (e.g., switch costs, attention residue) (Freytag et al., 2021). Furthermore, because social pressures to be available online threaten autonomy (Halfmann & Rieger, 2019), and autonomy threats are fatiguing (Ryan & Deci, 2008), the impact of online vigilance on mental fatigue may be especially pronounced when perceiving greater pressure to be available online.
Drawing from a large-scale mixed-methods study (N = 1,315) that combines passive smartphone monitoring and experience sampling (67,762 data points), the aim of the present study is to test the idea that online vigilance promotes mental fatigue, especially when perceiving greater availability pressure. We test this hypothesis using both self-report and passively sensed behavioral measures of online vigilance, the latter helping to overcome validity and reliability concerns with self-report measures (e.g. Ellis et al., 2019; Parry et al., 2021), and potentially opening the door to the development of just-in-time-adaptive-interventions for early-onset detection of burnout (e.g., Wang & Miller, 2020).
3.3 Theoretical Framework
3.3.1 Online Vigilance as the Subjective Experience of Being ‘On’
Ubiquitous connectivity serves obvious purpose and utility, making communication possible at great distances, enabling immediate satisfaction of various needs and desires, and offering various novel functionalities. But with these opportunities an ‘always on’ culture has emerged (McDowall & Kinman, 2017; Vanden Abeele, 2021), in which people are ‘permanently online, permanently connected’ (Vorderer et al., 2018), and find it increasingly difficult to switch off (Nguyen, 2021, 2023). Recent research shows that a repercussion of this ‘always on’ lifestyle is the tendency for individuals to be highly attuned to their online world, i.e., to be online vigilant (Johannes et al., 2021; Reinecke et al., 2018).
Online vigilance is argued to comprise three distinct components (Reinecke et al., 2018): salience, which reflects a cognitive preoccupation with the online world (e.g., thinking about incoming emails, Twitter feed, WhatsApp messages), reactibility, which reflects how quickly one responds to and prioritizes events and cues from the online sphere relative to offline demands, and monitoring, which reflects how often one actively enters their online sphere to check for new updates and information. Although online vigilance can be understood as a trait characterizing between-person differences in people’s mindsets towards the online world (Reinecke et al., 2018), ample research also supports its state-like nature: Studies demonstrate that people vary substantially moment-to-moment in online vigilance, and this variability correlates with various well-being outcomes (e.g., Johannes et al., 2019; Freytag et al., 2021; Gilbert et al., 2023; Johannes et al., 2021; Reinecke et al., 2018). The central hypothesis guiding this study is that one such well-being outcome affected by online vigilance is mental fatigue.
3.3.2 Online Vigilance as a Precursor to Mental Fatigue
The claim that online vigilance elevates individuals’ experiences of mental fatigue is grounded in a strong theory and evidence base in the field of psychology on the cognitive cost of being in a state of vigilance. Vigilance, a term often used interchangeably with sustained attention, can be understood as “the ability of observers to maintain their focus of attention and remain alert to stimuli over prolonged periods of time” (Warm et al., 2008, p. 115). Although originally believed to require minimal mental effort (Grier et al., 2003; Warm et al., 2008), recent empirical literature concludes that vigilance promotes fatigue — and does so frequently and sometimes to a non-minor extent (Finomore et al., 2006; Matthews et al., 2010; Smith et al., 2019; Warm et al., 2008), by depleting cognitive resources (Eisert et al., 2016). In line with resource expenditure theory (i.e. overload theory) prolonged vigilance is found to increase cognitive load, thereby leading to a faster and greater depletion of cognitive resources (Head & Helton, 2014). This depletion becomes even more pronounced when multitasking is involved, as dual-task performance requires allocating additional cognitive resources to the control of attention over two (or more) stimuli (Courage et al., 2015).
Applying this evidence and reasoning to the online sphere, it seems plausible that constantly attuning oneself to the online sphere, i.e., being online vigilant, is fatiguing. A key ingredient here is the ‘always on’ yet fragmented nature of online communication (Oulasvirta et al., 2012; Reinecke et al., 2018). To remain fully abreast of content and notifications — which often update moment-to-moment and become outdated just as quickly — one needs to keep the online sphere top of mind, as well as monitor and potentially immediately react to incoming information or communication. The latter activities require alternating attention from the ongoing task/activity to the digital sphere and back (Peng & Tullis, 2021). As such, individuals may face a depletion of cognitive resources not just because of being attentive to what is happening online increases the overall cognitive load, but also from the capacity costs associated with switching attention to and from communication channels, a cognitively depleting experience that has been well-documented in the literature on media multitasking (e.g., Baumgartner & Sumter, 2017; Rioja et al., 2023).
Circumstantial support for the fatiguing effect of online vigilance is visible, among others, in research on the work of journalists, who identify the constant monitoring of social media spaces as mentally taxing (Bossio & Holton, 2021). There is also a growing body of evidence on online vigilance that indirectly supports its fatiguing nature: Freytag et al. (2021) found that cognitive preoccupation with online communication (i.e., the salience component of online vigilance), together with media multitasking, depletes working memory and situational coping capacities. Similarly, Gilbert et al. (2023) found that online vigilance increases people’s communication load, and Johannes et al. (2021) found that online vigilance predicted a measure of affective wellbeing that included feeling tired (vs. awake) as one of its indicators. To date, however, research linking online vigilance directly to the experience of mental fatigue is still lacking. Moreover, to our knowledge, studies have not yet explored whether the effect of online vigilance lingers, which would strengthen causal claims over online vigilance and its potential to accumulate over time, thus perhaps contributing to the development of feelings of severe exhaustion, as witnessed in the case of burnout. Extant findings on the effect of online vigilance also result from studies involving student-heavy samples (e.g., Freytag et al., 2021; Gilbert et al., 2023). Accordingly, the first objective of this study is to bridge these knowledge gaps by formally testing the contemporaneous and lagged associations between online vigilance and mental fatigue in an adult sample:
H1: Online vigilance positively predicts mental fatigue at the same (H1a) and the next time-point (H1b).
3.3.3 Availability Pressure as Moderator
Digital communication comes with normative expectations attached (Bayer et al., 2016), among others expectations to be ‘always on’ (Nguyen, 2021). Individuals can perceive these expectations in the form of social pressure to be available to others online (Halfmann & Rieger, 2019). Prior research suggests that such perceived pressure exacerbates the negative effect of online vigilance on well-being outcomes. Gilbert et al. (2023), for instance, found that the negative impact of online vigilance on stress is greater when availability pressure is high. They explain this effect through the lens of self-determination theory, a theory that gives a comprehensive account of how internal and external pressures can rid individuals of their sense of autonomy (Ryan & Deci, 2008). Halfmann et al. (2021; see also Halfmann & Rieger, 2019) already observed that availability pressure can frustrate individuals in their autonomy needs. Building further on this observation, Gilbert et al. (2023) argued that under conditions of heightened availability pressure, people likely appraise the stress response resulting from online vigilance differently as they experience reduced autonomy to cope with it; Therefore, the debilitating effect on stress may be greater.
A similar pattern may exist for mental fatigue. First of all, autonomy-frustrating experiences are themselves often judged as fatiguing (e.g., Sheldon et al., 1996). Hence, if we consider availability pressure an autonomy-frustrating boundary condition, availability pressure itself may elicit fatigue. Organizational literature provides evidence for this, showing perceived expectations to be available after-hours to predict burnout symptoms including exhaustion (Hendrikx et al., 2023) and a more negative work–life balance (Belkin et al., 2020).
Because availability pressure in itself leads to a depletion of resources, however, situations of heightened pressure may also lead to an overall diminished capacity to regulate behavior so that energy is maximally preserved; circumstantial evidence to support this reasoning was found by Hall (2017). He observed that ‘mobile entrapment’, a construct similar to availability pressure, negatively impacts a measure of subjective well-being comprising items reflecting how energetic and fatigued individuals feel. Moreover, mobile entrapment also may act as a boundary condition that weakens the positive effect of texting on well-being, leading Hall to conclude that availability pressure shapes individual experiences. In the context of online vigilance and mental fatigue, the resource depletion resulting from heightened availability pressure may mean that individuals have less resources left to cope with and protect themselves against the fatiguing effect of online vigilance. The second objective of this study is to test this idea, by examining whether the association between perceived online vigilance and subjective fatigue is moderated by perceived availability pressure (H2a), and whether this relation holds true when looking at lagged effects (H2b).
H2: Availability pressure exacerbates both the momentary (H2a) and lagged (H2b) positive association between online vigilance and mental fatigue.
3.3.4 Behavioral Measures and Objective Assessments of Online Vigilance
To date, studies examining online vigilance have predominantly relied on self-report data. However, online vigilance can arguably be measured behaviorally, by examining how quickly people respond to notifications, or how often they check or pick up their smartphone, as these behavioral tendencies seem likely to capture this sense of vigilance to one’s online world (Johannes et al., 2021; Reinecke et al., 2018).
There is clear value in the development of behavioral online vigilance indicators, as they open up the possibility of implementing just-in-time interventions to regulate excess vigilance through passive sensing. Also, they overcome limitations of self-report measures of online vigilance, as we may assume it is difficult for people to accurately recall or capture how vigilant they were over the past few hours.
To our knowledge, only Johannes et al. (2021) has explored whether objective assessments of smartphone use behavior match self-perceptions of the behavioral components of online vigilance. Unfortunately, due to technical issues the authors used secondary operationalizations for which a theoretical link was less intuitive, and their study therefore did not show convincing evidence. This leaves room for new analyses that explore the potential behavioral correlates of online vigilance. Hence, a final objective of this study is to explore whether two of the behavioral components of perceived online vigilance, reactibility and monitoring, have behavioral correlates in smartphone use (RQ1), and whether these correlates perform similarly in predicting mental fatigue.
RQ1: Can passively sensed behavioral smartphone use patterns serve as behavioral proxies for online vigilance, more specifically for reactibility and monitoring?
RQ2: How do these behavioral patterns perform in predicting momentary and temporally lagged fatigue?
In line with these objectives, we will extend our investigation by assessing our four hypotheses H1a, H1b, H2a, and H2b, using behavioral measures of online vigilance in addition to self-report measures. These hypotheses represent H3a, H3b, H4a, and H4b, respectively.
3.4 Method
3.4.1 Participants
This preregistered study uses data gathered from 1,315 adult participants (mean age = 38.83 years; SD = 11.72), who participated voluntarily in a two-week mixed-methods study in the Fall of 2022. The study took place in the context of a citizen-science project, and was advertised with the aid of a large newspaper (with a daily reach of around 450,000 readers) who invited people to participate in our study through both their print and online news channels (see @#sec-appendix-a for further details). The project received approval from the institutional review board of Ghent University.
The 1,315 participants included in formal analyses provided 67,762 ESM datapoints, which equates to an average of 51.53 questionnaires completed per participant. Most participants (62%) identified as female (37% male; 1% non-binary/other), were highly educated (93% had at least a bachelor’s degree), and were actively working (77%; 8% were students; 15% were not in paid employment, e.g., retired, sick leave, etc.). Most participants used an Android device (70%), a percentage higher than the general adult population in this region (62% in 2021 (Sevenhant et al., 2022)). The remaining 30% used an iOS device.
Although 916 participants were Android users and installed the tracking app so their smartphone could be logged, due to various technical difficulties (e.g., smartphone operating system blocking tracking access, see @#sec-appendix-a for more information), log data could only be collected from 834 participants (91%). More than 7 million unique phone-events were collected from these participants, comprising their app, screen, notification and keyboard activity.
3.4.2 Procedure
All participants were asked to install m-Path, an app to receive experience sampling questionnaires. Android users were additionally invited to install a dedicated app built for the project, that passively monitored participant smartphone behavior2 (see below for further details). See the section ‘Behavioral measures’ for details.
In the experience sampling app, over a period of 14 days participants received 6 questionnaires per day between 7.30 am and 10.45 pm. Each notification was randomly scheduled within a 90 minute timeslot, and remained available for 45 minutes (with a reminder after 30 minutes). Time intervals varied from minimally 1 hour and 15 minutes to maximally 4 hours and 15 minutes between two subsequent questionnaires; on average responses were separated by 3 hours and 4 minutes (SD = 16.66 minutes). Additional details about the onboarding procedure and the ESM questionnaire can be found in @#sec-appendix-a.
3.4.3 Measures
3.4.3.1 Self-report Measures
The ESM items relevant to the present study were measured on a 7-point Likert scale ranging from 1 (‘Not at all’) to 7 (‘All the time’). The first questionnaire of the day had items that referred to the time period since waking up (e.g., “Since getting up this morning, I felt […]”), while all other questionnaires had items that referred to the time period since the previous questionnaire they received (e.g., “Since the last questionnaire, I felt […]”).
Online Vigilance. The online vigilance measure included three items that were adapted from Reinecke et al. (2018), reflecting monitoring (‘[…] I continuously monitored my emails and messages.’), reactibility (‘[…] I gave incoming messages and emails my immediate attention.’), and salience (‘[…] my thoughts were continuously with my emails, messages and social media.’).
Availability Pressure. Availability pressure was measured with one item, stating ‘[…] I felt pressure to be digitally available to others.’
Mental fatigue. Mental fatigue was measured with one item, stating ‘[…] I felt mentally drained.’
Wake-up time. To mark the onset of the first time window of the day, wake-up time was also measured in the first questionnaire of each day with the following item (time-entry field): ‘At what time did you get up today (hh:mm, e.g., 07:30)?’
3.4.3.2 Behavioral Measures
We passively monitored Android users’ smartphone use, gathering information on start and stop times of smartphone sessions and app activities, as well as timestamped notification data. Three digital proxies were calculated from these data, with one of these features representing the monitoring dimension of online vigilance, and two of them representing the reactibility dimension of online vigilance. These features were computed for each ESM time interval. For instance, if the fourth ESM questionnaire of the day asked participants how mentally fatigued they felt since the previous questionnaire they received, each digital proxy of online vigilance covered that exact same period to facilitate formal analyses. The time window for each survey corresponded to the period from the sending of the previous survey, as clarified in both the participation instructions and the baseline survey.
Monitoring Behavior. To capture the monitoring dimension, we computed a feature reflecting the number of times a participant unlocked and relocked their phone again within 5 seconds, without directly reacting to/accessing the phone via a notification pop-up. This action can be seen as a “self-initiated” or voluntary phone unlock that merely serves the purpose of ‘checking’.
Responsiveness. To capture the reactibility dimension, we computed a first feature representing the percentage of notifications reacted upon, calculated as the proportion of (a) number of notifications clicked (= reacted upon) divided by (b) number of notifications received.
Reaction Time. Also, to capture the reactibility dimension, we computed a second feature reflecting the time it took to react to notifications, calculated by computing the average amount of time between receiving an app notification and clicking it.
These last two features were computed based on the notification data from communication and social media apps (e.g., e-mail, (work) communication, social media apps). Further information on data cleaning and processing can be found in @#sec-appendix-a.
3.4.4 Data Analysis
ESM data is nested (observations within participants). Thus, a multilevel model was required to prevent standard error underestimation (Hox, 1998). Also, given that our models include latent factors, SEM was required to model associated errors. To meet both requirements, a multilevel SEM was conducted. To enable these analyses using the R package lavaan, level 2 of our model was saturated (see https://lavaan.ugent.be/tutorial/multilevel.html for details). This step also ensured model fit indices concerned level 1 of our model, our primary focus. As part of these analyses, random intercepts were used with fixed slopes. Prior to testing our structural model, however, we conducted confirmatory factor analysis (CFA) to test our measurement model for online vigilance, both as indicated by self-report and by behavioral variables. Model fit indices were used to assess our structural models given present degrees of freedom, with adequate fit determined based on the following three criteria: (a) CFI: above .90: moderate but acceptable model fit, above .95: very good model fit. (b) RMSEA: lower than .06 good model fit, between .06 and .08 moderate but acceptable model fit, and above 0.08 poor model fit. (c) SRMR: .05 or below demonstrates good model fit. These criteria align with current standards (Hu & Bentler, 1999; MacCallum et al., 1996; Marsh et al., 2004). If model fit criteria could not be met, our models were adapted in line with theory and output modification indices. Lastly, we employed two-tailed testing, and considered effects significant if p-values were .05 or less.
3.4.5 Transparency and Openness
Information about our preregistration, code and analysis, as well as the larger project in which this study is situated, can be found on osf (https://osf.io/4hb2c/). Our preregistration stated that when structural models would not reach adequate fit, theoretically and methodologically relevant adaptations would be implemented. Unfortunately, this was the case for all but one of our models. We specify at the start of the confirmatory results section which adaptations were made, and for which reasons.
3.5 Results
3.5.1 Descriptives
Descriptives for all measures per ESM time window and for daily smartphone use are included in Table 3.1. On average participants were very rarely to rarely vigilant to their online world (self-report variables), under pressure to be available online, or experiencing mental fatigue. Participants spent on average around 180 minutes on their smartphone per day, opened approximately 107 apps, unlocked their devices 85 times and received about 41 notifications each day.
| n | Mean | SD | Min | Max | |
| Person-level means descriptivesa | |||||
| saliences | 1315 | 2.45 | 0.94 | 1.00 | 6.24 |
| monitorings | 1315 | 2.75 | 0.98 | 1.00 | 6.16 |
| reactibilitys | 1315 | 3.08 | 1.01 | 1.00 | 6.51 |
| mental fatigues | 1315 | 2.83 | 1.20 | 1.00 | 6.86 |
| availability pressures | 1315 | 2.45 | 1.04 | 1.00 | 6.62 |
| monitoring sessionsl | 767 | 9.61 | 6.36 | 1.00 | 59.00 |
| responsivenessl (%) | 663 | 0.30 | 0.18 | 0.01 | 1.00b |
| reaction timel (sec.) | 655 | 641.20 | 455.58 | 0.71 | 3565.61 |
| Daily smartphone use descriptivesc | |||||
| screentime (min.) | 702 | 179.80 | 80.98 | 11.59 | 502.18 |
| app frequency | 702 | 106.90 | 54.52 | 8.60 | 337.08 |
| screen unlocks | 761 | 84.77 | 42.00 | 3.00 | 262.14 |
| notifications | 736 | 33.62 | 29.32 | 1.57 | 261.33 |
Note. s denotes self-report variables; l denotes behavioral variables. a Person-level means per ESM time window. b One participant’s average was excluded due to a technical issue with their data. c Daily smartphone descriptives were measured for participants with at least seven days of data.
3.5.2 Self-report Models
We first report the results of the self-report models. The analyses for these models were based on data from the total sample of 1,315 adults. As mentioned above, we deviated from our pre-registered analysis plans when model fit was unsatisfactory. We did so in three ways:
First, we included covariances between variables measured at the same time-point as model fits improved greatly when doing so, and this also makes theoretical sense. Second, for the temporally lagged models, we added a cross-lagged path from mental fatigue (T0) to online vigilance (T1). We did so because model fit of the original model was poor, whereas the cross-lagged model fitted well and, in our view, only provides greater support for the hypotheses tested. Third, theory indicated a direct impact of availability pressure on mental fatigue, which we had not preregistered. Again, adding this effect greatly improved model fit. Hence, we adapted our models accordingly. To keep the results section concise, only the results of the finalized models are discussed below. The original models (with poor model fit) are added in @#sec-appendix-a, and the step-by-step procedure is detailed on osf.
3.5.2.1 Confirmatory Factor Analysis of Self-Report Online Vigilance Items
We first performed a confirmatory factor analysis to test whether our three online vigilance items loaded onto one factor3. The (just-identified) model showed that the factor loadings of the salience, monitoring and reactibility self-report items were significant with β’s ranging from .68 to .81 (see Table 3.2). Additionally, these items demonstrated high within-person (r = .50 to .59) and between-person (r = .82 to .89) correlations. These findings (also detailed in Table 3.3) broadly support using a one-factor structure for the self-report measure of online vigilance.
| Latent Factor | Indicator | B | SE | Z | p | β |
| Self-reported Online Vigilance | ||||||
| salience | 1.00 | .00 | .68 | |||
| monitoring | 1.36 | .01 | 148.65 | <.001 | .81 | |
| reactibility | 1.26 | .01 | 149.75 | <.001 | .73 | |
| Behavioral Online Vigilance | ||||||
| responsiveness | 1.00 | .00 | .23 | |||
| monitoring sessions | −4.02 | 2.95 | −1.36 | .173 | −.83 | |
| reaction time | .35 | .08 | 4.67 | <.001 | .08 | |
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | |
| 1. saliences | --- | .89*** | .82*** | −.10*** | −.16*** | .14*** | .43*** | .84*** |
| 2. monitorings | .55*** | --- | .90*** | −.12*** | −.10*** | .09*** | .34*** | .79*** |
| 3. reactibilitys | .50*** | .59*** | --- | −.11*** | −.14*** | .12*** | .30*** | .74*** |
| 4. reaction timel | −.03*** | −.05*** | −.06*** | --- | .02*** | −.05*** | −.08*** | −.07*** |
| 5. responsivenessl | −.07*** | −.09** | −.09*** | .03* | --- | −.15*** | −.09*** | −.16*** |
| 6. monitoring sessionsl | .09*** | .10*** | .10*** | −.07*** | −.19*** | --- | .05*** | .12*** |
| 7. mental fatigues | .17*** | .17*** | .14*** | −.01 | −.03* | .02*** | --- | .46*** |
| 8. availability pressures | .45*** | .47*** | .44*** | −.03*** | −.11*** | .09*** | .18*** | --- |
Note. * p < .05, ** p < .01, *** p < .001. Within-person correlations are shown below the diagonal; between-person correlations above the diagonal. These correlations do not take the multilevel structure into account, and hence will be slightly inflated. s denotes self-report variables; l denotes behavioral variables.
3.5.2.2 Hypotheses Tests with Self-reported Online Vigilance
Hypothesis 1 posited that self-reported online vigilance would predict mental fatigue, both at the same timepoint (H1a) and lagged (H1b). Hypothesis 1a was supported by the data (see Table 3.4), showing a clear effect of online vigilance on mental fatigue (β = 0.17, p < .001), controlling for mental fatigue at the previous time-point. The model fit the data well. Hypothesis 1b was also supported (see Table 3.4), with online vigilance predicting greater mental fatigue at the next time-point (β = .07, p < .001), controlling for mental fatigue at the present time-point (β = .39, p < .001). It should be noted that we also found a cross-lagged effect of mental fatigue on online vigilance (β = .02, p < .001).
We performed an additional (non-preregistered) exploratory analysis to examine whether there was the effect of online vigilance on mental fatigue lagged even further (lag-2 effect). Controlling for lag-1 online vigilance, we found that lag-2 online vigilance also predicted mental fatigue (β = .01, p = .049), thus showing that the fatiguing effect of online vigilance lingers on, leaving a residue at least 4 to 6 hours later.
Hypothesis 2 stated that availability pressure would moderate the aforementioned contemporaneous (H2a) and temporally lagged (H2b) associations, so that in instances where availability pressure was higher, the association between online vigilance on mental fatigue would be stronger. The hypothesis was not supported (see Table 3.4): availability pressure did not significantly moderate the contemporaneous (β = .01, p = .631), nor the lagged (β = .01, p = .666) association between online vigilance and mental fatigue.
When fitting the moderation models, we observed that availability pressure had a direct effect on mental fatigue in both the contemporaneous and lagged models, and that it also covaried strongly with online vigilance. A similar observation was made in recent work from Gilbert et al. (2023), in which the authors suggest to explore mediation models. Based on our observations and following this suggestion, we performed an additional exploratory analysis, modelling a lagged effect of availability pressure on online vigilance, calculating both its direct lagged and its indirect effect (via online vigilance) on mental fatigue. Findings support a direct lagged effect of availability pressure on mental fatigue (β = 0.06, p < .001), and an indirect effect via online vigilance (β = 0.05, p < .001; see Table 3.5).
| Hypothesis | Regressions | Indicator | B | SE | 95% CI | p | β | CFI | RMSEA | SRMR |
| H1a — Outcome: mental fatigue | ||||||||||
| Predictor | online vigilance | .25 | .01 | [.24, .27] | <.001 | .17 | .99 | .05 | .05 | |
| lagged mental fatigue | .38 | .00 | [.38, .39] | <.001 | .39 | |||||
| H1b — Outcome: mental fatigue | ||||||||||
| Predictor | lagged online vigilance | .10 | .01 | [.08, .11] | <.001 | .07 | .96 | .09 | .08 | |
| lagged mental fatigue | .39 | .00 | [.38, .40] | <.001 | .39 | |||||
| Predictor (online vigilance) | lagged online vigilance | .48 | .01 | [.47, .50] | <.001 | .49 | ||||
| lagged mental fatigue | .02 | .00 | [.01, .02] | <.001 | .02 | |||||
| Covariance | mental fatigue — online vigilance | .12 | .00 | [.12, .13] | <.001 | .17 | ||||
| lagged mental fatigue — lagged online vigilance | .20 | .01 | [.19, .21] | <.001 | .21 | |||||
| H2a — Outcome: mental fatigue | ||||||||||
| Predictor | online vigilance (ov) | .17 | .01 | [.15, .19] | <.001 | .13 | .99 | .04 | .05 | |
| lagged mental fatigue | .38 | .00 | [.37, .39] | <.001 | .39 | |||||
| availability pressure (ap) | .05 | .01 | [.03, .08] | <.001 | .05 | |||||
| ov × ap | .00 | .00 | [−.01, .01] | .631 | .01 | |||||
| Covariance | online vigilance — availability pressure | .58 | .01 | [.56, .59] | <.001 | .60 | ||||
| H2b — Outcome: mental fatigue | ||||||||||
| Predictor | lagged online vigilance (ov) | .06 | .01 | [.04, .09] | <.001 | .04 | .99 | .04 | .06 | |
| lagged mental fatigue | .39 | .00 | [.38, .39] | <.001 | .39 | |||||
| lagged availability pressure (ap) | .03 | .01 | [.01, .05] | .008 | .03 | |||||
| ov_lagged × ap_lagged | .00 | .00 | [−.01, .01] | .666 | .01 | |||||
| Predictor (online vigilance) | lagged online vigilance | .49 | .01 | [.47, .50] | <.001 | .49 | ||||
| lagged mental fatigue | .02 | .00 | [.01, .03] | <.001 | .03 | |||||
| Covariance | online vigilance — mental fatigue | .12 | .00 | [.11, .13] | <.001 | .17 | ||||
| lagged online vigilance — lagged mental fatigue | .05 | .00 | [.05, .06] | <.001 | .06 | |||||
| Regressions | Indicator | B | SE | 95% CI | p | β | CFI | RMSEA | SRMR |
| Outcome: mental fatigue | |||||||||
| availability pressure | .06 | .01 | [.05, .08] | <.001 | .06 | .99 | .03 | .01 | |
| lagged availability pressure (direct) | .06 | .01 | [.05, .07] | <.001 | .06 | ||||
| lagged availability pressure (indirect) | .06 | .00 | [.05, .06] | <.001 | .05 | ||||
| online vigilance | .24 | .01 | [.22, .26] | <.001 | .16 | ||||
| Outcome: online vigilance | |||||||||
| lagged availability pressure | .23 | .00 | [.23, .24] | <.001 | .33 | ||||
| Covariance | |||||||||
| availability pressure — online vigilance | .45 | .01 | [.44, .47] | <.001 | .54 | ||||
| availability pressure — lagged availability pressure | .40 | .01 | [.39, .42] | <.001 | .32 | ||||
3.5.2.3 Exploratory Day-Level Analysis
Finally, we added an exploratory analysis to examine whether the associations between availability pressure and online vigilance uphold at the day-level. This was the case (see Table 3.6): Average daily availability pressure (β = 0.15, p < .001) and online vigilance (β = 0.17, p < .001) predicted average daily mentally fatigue; there was an indirect effect of availability pressure via online vigilance (β = 0.12, p < .001).
| Regressions | Indicator | B | SE | 95% CI | p | β | CFI | RMSEA | SRMR |
| Outcome: daily mental fatigue | |||||||||
| daily availability pressure | .17 | .01 | [.14, .19] | <.001 | .15 | .99 | .05 | .01 | |
| daily availability pressure (indirect) | .14 | .01 | [.12, .15] | <.001 | .12 | ||||
| daily availability pressure (total) | .30 | .01 | [.29, .32] | <.001 | .27 | ||||
| daily online vigilance | .25 | .02 | [.21, .28] | <.001 | .17 | ||||
| Outcome: daily online vigilance | |||||||||
| daily availability pressure | .55 | .01 | [.54, .57] | <.001 | .70 | ||||
3.5.3 Behavioral Models
3.5.3.1 Confirmatory Factor Analysis of Behavioral Online Vigilance Items
The next step in our analysis was to explore whether we could infer behavioral proxies for online vigilance from the smartphone log data. We first performed a CFA with the three behavioral indicators that we calculated, reflecting monitoring (operationalized as pro-active checking of the phone) and reactibility (operationalized as relative responsiveness to notifications and response speed). Interestingly, factor loadings did not support the features loading onto one factor, with β’s ranging from .08 to −.83 (see Table 3.2), and unexpectedly low within- (r = .03 to −.19) and between-person (r = .02 to −.15) correlations (see Table 3.3). We further explored associations between the self-report online vigilance items and the behavioral features, and here also, correlations were surprisingly weak at both the within (r = .04 to .07) and between-person level (r = .09 to −.16; see also Table 3.3).
Although, based on the evidence, we cannot claim that the selected features serve as proxies for the experience of online vigilance, we nonetheless examined each feature’s capacity to predict mental fatigue, both contemporaneously and lagged, and with and without availability pressure as a moderator.
3.5.3.2 Hypotheses Tests with Behavioral Online Vigilance Measures
Parallel to our first hypothesis, our third hypothesis posited that, controlling for previous fatigue levels, behavioral online vigilance would predict mental fatigue, both contemporaneously (H3a) and lagged (H3b). We tested models for each separate feature. These base models were just-identified, hence their fit could not be examined. After modelling covariances as free parameters, however, both models showed good fit (see Table 3.7). No support for an effect was found.
Adding availability pressure as a moderator in the contemporaneous (H4a) and temporally lagged (H4b) models yielded similar results, with one exception for reaction time, which showed a small but significant, and counter-intuitive contemporaneous effect (β = .07, p = .048) on mental fatigue after controlling for availability pressure and the interaction effects. This finding suggests that when people respond more slowly to notifications, they are more fatigued. While theoretically not implausible, overall, these results lead us to reject all four behavioral hypotheses (H3a to H4b).
| Hypothesis | Regressions | Indicator | B | SE | 95% CI | p | β | CFI | RMSEA | SRMR |
| H3a — Outcome: mental fatigue | ||||||||||
| Predictor | reaction time | .00 | .02 | [−.03, .04] | .835 | .00 | 1.0 | .00 | .01 | |
| responsiveness | −.03 | .02 | [−.07, .01] | .086 | −.02 | |||||
| monitoring sessions | .01 | .02 | [−.03, .04] | .630 | .01 | |||||
| lagged mental fatigue | .39 | .01 | [.36, .41] | .000 | .39 | |||||
| H3b — Outcome: mental fatigue | ||||||||||
| Predictor | lagged reaction time | −.01 | .02 | [−.05, .03] | .613 | −.01 | .99 | .02 | .02 | |
| lagged responsiveness | −.01 | .02 | [−.05, .03] | .659 | −.01 | |||||
| lagged monitoring sessions | .01 | .02 | [−.03, .05] | .541 | .01 | |||||
| lagged mental fatigue | .38 | .02 | [.35, .41] | .000 | .38 | |||||
| H4a — Outcome: mental fatigue | ||||||||||
| Predictor | reaction time | .10 | .05 | [0, .19] | .048 | .07 | .99 | .05 | .05 | |
| responsiveness | −.01 | .04 | [−.1, .07] | .738 | −.01 | |||||
| monitoring sessions | .05 | .04 | [−.03, .12] | .215 | .04 | |||||
| lagged mental fatigue | .37 | .01 | [.35, .4] | .000 | .38 | |||||
| availability pressure (ap) | .19 | .03 | [.13, .25] | .000 | .18 | |||||
| reaction time × ap | −.09 | .05 | [−.19, .01] | .067 | −.07 | |||||
| responsiveness × ap | .00 | .05 | [−.09, .09] | .961 | .00 | |||||
| monitoring sessions × ap | −.06 | .04 | [−.14, .03] | .206 | −.05 | |||||
| H4b — Outcome: mental fatigue | ||||||||||
| Predictor | lagged reaction time | .05 | .05 | [−.04, .13] | .316 | .04 | .98 | .07 | .06 | |
| lagged responsiveness | −.02 | .04 | [−.1, .06] | .654 | −.01 | |||||
| lagged monitoring sessions | .01 | .04 | [−.06, .09] | .715 | .01 | |||||
| lagged mental fatigue | .37 | .02 | [.34, .4] | .000 | .37 | |||||
| lagged availability pressure (ap) | .10 | .03 | [.04, .16] | .000 | .09 | |||||
| lagged reaction time × ap | −.06 | .05 | [−.15, .03] | .199 | −.05 | |||||
| lagged responsiveness × ap | .02 | .04 | [−.06, .11] | .587 | .02 | |||||
| lagged monitoring sessions × ap | −.01 | .05 | [−.1, .08] | .831 | −.01 | |||||
3.6 Discussion
Psychological research on offline contexts provides ample evidence that state vigilance promotes mental fatigue, yet to our knowledge, this study is among the first to investigate whether online vigilance affects individuals similarly. This is an important question to investigate. After all, online vigilance has already been found to predict stress (Freytag et al., 2021) and reduce affective well-being (Johannes et al., 2021). If it also mentally fatigues individuals, then this further justifies societal concerns over the harms associated with being ‘always on’. Most notably, given the strong link between mental fatigue and burnout (Edú-Valsania et al., 2022), knowing that online vigilance promotes mental fatigue brings immediate implications for how we understand and tackle this mental health condition.
Drawing from the data of 1,315 adult individuals, we investigated this question by exploring the momentary and lagged impact of online vigilance on mental fatigue, and additionally examined whether the pressure to be constantly available might exacerbate this link. Finally, we explored the potential of behavioral smartphone features to serve as ‘passively sensed’ proxies for online vigilance, which would open the door to detect the early onset of mental fatigue.
Overall, our self-report findings lend strong support for the hypothesis that online vigilance promotes mental fatigue, both momentarily and lagged: constantly attending to, monitoring and reacting to the online world comes with an energy-sapping cost, not only ‘in the moment’ but also further in time. In line with psychological theories on vigilance (e.g., Warm et al., 2008), we may thus conclude that online vigilance represents a form of ‘mental work’ with fatiguing qualities. This fatiguing effect may linger in time because fatigued individuals may lack the resources necessary to actively regulate themselves (see also Inzlicht et al., 2014). With mental fatigue also predicting online vigilance, there may thus be a risk for mutual reinforcement that could ultimately lead to a negative spiral of fatigue. Together with the extant evidence on other maladaptive outcomes of online vigilance such as stress (e.g., Freytag et al., 2021; Gilbert et al., 2023; Reinecke et al., 2018), our findings support broader concerns over being ‘always on’ and its repercussions for individual health and well-being.
Important to note here, is that the magnitude of the identified effects here is small; an individual would have to become much more vigilant of their online world (e.g., if an individual moves from stating they ‘very rarely’ were vigilant of their online world in the past hours, to stating they were ‘very often’ vigilant of it) to move even 1-point upwards on the mental fatigue scale (e.g., from ‘rather true’ to ‘true’). Yet, although the effect is very small in magnitude, and intriguingly may be so small that individuals may not easily perceive it, the lagged associations do indicate that mental fatigue effects may not get fully regulated, and could instead accumulate over-time, similar to how small calorie increases can snowball into more consequential downstream effects (e.g.: Most et al., 2017). Hence, small effects per se should not be discounted out of hand, particularly when effects pertain to very narrow timescales (e.g., a few hours). That stated, further research is needed before concluding a pronounced impact on longer-term outcomes such as burnout.
A second aim of our study was to examine whether availability pressure exacerbates the impact of online vigilance on mental fatigue. After all, self-determination theory suggests internal or external pressures have a depleting quality (see also Ryan & Deci, 2008), hence it would be plausible that under greater availability pressure individuals lack the necessary resources to cope with and regulate the fatiguing effect of online vigilance effectively. No moderating effect was found, however, although availability pressure did predict mental fatigue directly (thus supporting its depleting effect), both acutely and downstream (i.e., next time-point) and also indirectly, through online vigilance. This result shows a remarkable parallel with the findings from Gilbert et al. (2023) recent study exploring the link between online vigilance, and we agree that “studies with an even more granular temporal level, or experimental studies are needed that are able to clearly delineate the various roles these concepts play” (p. 452). A tentative explanation for the nonsignificant moderating effect that could be explored in such future research, is the proposition that availability pressure itself might not always be negatively valenced; for instance, one might perceive that others expect them to be available in anticipation of good news (e.g., a birth announcement). Such availability expectations might not necessarily reduce or frustrate autonomy needs, but may on the contrary reflect belongingness, which may in turn drive individuals to remain vigilant over their communication channels.
A third and final aim of this study was to examine whether three behavioral smartphone use features which we believed would give expression to the behavioral dimensions of online vigilance, namely the speed at which an individual responds to notifications (reactibility), the percentage of notifications they responded to (reactibility), and the amount of self-initiated smartphone unlocks (monitoring), would predict mental fatigue. The value of these features lies in their possibility to be passively monitored, which overcomes issues associated with self-report data such as recall bias and common method variance, but also opens the door to digitally phenotyping online vigilance, and building automatized interventions on those digital phenotypes (e.g., sending the user a prompt, informing them they might benefit from a pause). Unexpectedly, however, our behavioral features appeared a poor measure of self-reported online vigilance: They showed a very weak association between themselves, as well as with the corresponding self-reported online vigilance items. Likely as a result of the latter, they did not perform well in predicting mental fatigue.
Various explanations can be sought for this unexpected finding. A first obvious possibility is that the chosen features have low validity: Perhaps they do not discriminate well between state online vigilance and, alternatively, the normal waxing-and-waning of task engagement. After all, human beings have an innate motivation to balance ‘exploitation and exploration’ (Inzlicht et al., 2014) — i.e., to alternate task engagement with the pursuing of alternative, potentially online activities (Wiradhany et al., 2021). For instance, short social media breaks may help individuals recover from a taxing work task. It may be that, at a behavioral level, the features that we identified as representations of online vigilance may share overlap with what smartphone behavior looks like when relaxing from work, creating a major confound in the association between these behaviors and the subjective experience of online vigilance. Future lab-based and observational research is needed to explore this tentative explanation.
Second, the lack of behavioral findings raises the question of how online vigilance manifests behaviorally if not via the theorized means. Initially, it may be that the smartphone features that we developed are not good representations of online vigilant behavior, and that we ought to have operationalized this construct differently. An exploration of alternative operationalizations, however, did not immediately lead to markedly different results. Additionally, perhaps smartphone behavior alone is not indicative enough, and we should also glean other behavioral data, such as people’s computer use. After all, (work) e-mails may be something adults are highly online vigilant over, but these e-mails might be dealt with on the computer more so than the smartphone. Also, many popular communication channels are simultaneously available on smartphones and other devices, and there may be a ‘handover’ from one device to the other depending on the context one is in. Hence, to develop valid behavioral features of online vigilance, we might have to look at all digital media behaviors.
Finally, one crucial conclusion that we might draw from our collective evidence is that subjective experience trumps behavior in explaining the downstream implications of digital media use. This perspective aligns with recent findings by media scholars [such as Ernala et al. (2022); Lee et al. (2021); Lee & Hancock, 2023; Vanden Abeele (2021)]. Their work underlines the significance of subjective beliefs, experiences and mindsets in relation to digital connectivity in explaining the impact of media use on different psychological outcomes. Online vigilance might, similarly, represent such a mindset or media experience which is better captured subjectively, rather than through the use of different behavioral features.
This study is not without limitations, of which the first is undoubtedly the use of a self-selected sample. By cooperating with a local newspaper, advertisements for the study were solely published on their website and in print, to an audience that is, on average, older and more highly educated. Individuals also participated voluntarily, which likely indicates their interest in the topic. This approach has likely narrowed down our potential research population and may have potentially led to certain biases. Alternatively, intensive longitudinal studies on online vigilance in adult (rather than in student) populations are generally less prevalent, and the fact that participants in our study were not financially incentivized for their participation has likely improved data quality overall.
We wish to remark here that extending this line of research to a clinical population, in particular, could provide insights into potential implications of online vigilance on mental health and wellbeing in vulnerable groups. Such research could also consider the role of accumulated fatigue in the relationship between online vigilance and well-being, allowing a more nuanced understanding of its long-term impacts, tipping points, and potential resilience factors.
A second major limitation of this study is that participation in this study required smartphone ownership and use, for the completion of daily ESM questionnaires. This has two major ramifications. First, during the recruitment process, the research team received a number of requests for participation from adults who did not own or use a smartphone. Because managing the data collection was already extremely challenging, we did not act upon these requests, but it is important to remark that this subpopulation who deliberately chooses not to use a (smart) mobile phone may provide crucial insight into digital well-being phenomena. Second, in the end-of-study questionnaire, a number of individuals pointed towards the irony of using a smartphone for a research project on the smartphone and indicated that their ‘missing values’ were often due to deliberate decisions to go offline. Some participants also noted being vigilant for the study notifications, causing an obvious confound. Here we lay bare an intrinsic methodological problem of ESM research for the study of digital well-being and digital disconnection, an observation also made by Klingelhoefer et al. (2023), namely that you intervene on the very reality that you aim to capture, thus refuting the claim of ecological validity of ESM research.
Third, in our study we observed a high correlation between availability pressure and online vigilance, suggesting a notable conceptual overlap that warrants further consideration. This overlap may imply that in certain contexts, the perceived pressure to be available online and the resultant vigilance in monitoring digital communications are so closely intertwined that they might influence each other more profoundly than initially anticipated. This finding raises important questions about the independence of these constructs and the potential for a shared underlying factor. It also suggests that the measures used may capture elements of both constructs, thus potentially affecting the clarity of the distinctions drawn between them. Future research should explore this further, potentially by exploring through lab experimentation if a meaningful distinction can be drawn.
Finally, while our study explored lagged relationships to address causality concerns, reverse causality as well as the presence of unmeasured third variables influencing the observed outcomes need further exploration. It is plausible, for instance, that participants were more vigilant to their online world during periods of elevated stress, and that this stress contributed to increased mental fatigue.
Notwithstanding these limitations, our study also has several strengths: Our findings draw from a very large sample (1,315 participants, utilizing around 42,000 datapoints to test contemporaneous and lagged models that control for mental fatigue’s autoregressive effect), that included a large number of workers, with ages spanning from 18 to 82 years old, and showed effects that were directionally consistent across models. Thus, we have some confidence that the effects in our study are unlikely to reflect type 1 errors, and that the identified effects can tell us something about what excessive online vigilance may lead to in populations particularly prone to experiencing burnout, namely, those who are actively employed. Moreover, a positive link between online vigilance and mental fatigue was established both after controlling for likely confounds (e.g., autoregression effects) and after temporally separating online vigilance from mental fatigue, thus offering some suggestion of potential causality. Our use of experience sampling also reduced concerns about recall bias, a critical issue in some previous research that examined the link between online vigilance and outcomes conceptually related to mental fatigue (Freytag et al., 2021; Reinecke et al., 2018), whereby participants had to recall their experiences from many more hours prior. Lastly, our work adds to extant scholarship work on online vigilance and adjacent outcomes such as stress (Freytag et al., 2021; Gilbert et al., 2023; Reinecke et al., 2018) or mind-wandering and mindfulness (Johannes et al., 2018). As such, it further showcases the importance of the subjective experience of being ‘always on’ impacting well-being. To summarize, then, putting aside the necessary caution one must have for a first test of effects, the strength of the present findings edges forward the idea that increased levels of online vigilance are likely to elevate mental fatigue, and further down the line, if unchecked, the broader symptoms of burnout.
Published as: Van Gaeveren, K., Murphy, S. L., de Segovia Vicente, D., & Vanden Abeele, M. M. P. (2024). Connected Yet Cognitively Drained? A Mixed-Methods Study Examining Whether Online Vigilance and Availability Pressure Promote Mental Fatigue. Communication Research, 0(0). https://doi.org/10.1177/00936502241248494↩︎
This app functions solely on Android devices and, to our knowledge, no iOS alternatives exist which can capture detailed trace data.↩︎
Because this model was just-identified, model fit could not be examined using standard model fit indices (e.g., CFI, RMSEA, etc.; Hu & Bentler (1999); MacCallum et al. (1996); Marsh et al. (2004)).↩︎