Out of Sight, Out of Mind? A Longitudinal Investigation of Smart Working and Burnout in the Context of the Job Demands–Resources Model during the COVID-19 Pandemic

The academic interest in smart working, a form of flexible work characterized by the use of technology to conduct one’s work, has dramatically increased over recent years, especially during the COVID-19 pandemic. Building on the job demands–resources (JD-R) model, in this study we investigate whether smart working affects the longitudinal association between perceived work characteristics, such as workload and social support (SS), and workers’ health and well-being, in terms of exhaustion. Overall, 185 workers completed a self-report questionnaire at two time points (fourmonth time-lag) during the COVID-19 outbreak. The results from moderated multiple regression analysis partially support our predictions. The longitudinal association between workload and exhaustion was positive—although marginally significant—for smart workers, but nonsignificant for in-person workers. Contrarily, the longitudinal association between SS and exhaustion was negative for in-person workers, but nonsignificant for smart workers. Overall, this study suggests that, to support employees’ health and productivity, work characteristics—both physical and psychosocial— should fit the new way of working as well as remote workers’ specific needs and expectations. Hence, to promote sustainable work, interventions should be aimed at helping smart workers to manage their workload effectively, as well as reducing professional and social isolation.


Introduction
Technological advancements that have occurred over recent years have facilitated the redefinition of work processes, including the diffusion of alternative work arrangements [1] that provide employees with more flexibility in scheduling their job tasks, with respect to time, place and method of execution [2]. In this scenario, telecommuting, which has been referred to as remote work, telework or distance work, among other labels [3], represents the most common flexible work arrangement; by using information and communication technologies (ICTs), telecommuting allows employees to work away from the traditional workplace (e.g., the office) [2]. However, although the various terms adopted in the literature overlap to some extent [3,4], some differences across the distinct conceptualizations of telecommuting need to be acknowledged [3]. In their review of existing research on telecommuting, Allen et al. [3] proposed the following definitions: telecommuting usually "involves working some portion of time away from the conventional workplace, often from home, and communicating by way of computer-based technology" (Allen et al., 2105, p. 43; see also Golden, 2006) [3,5]. Furthermore, as noted by Allen et al. (2015, p. 43) [3], the term telework, most frequently used in European or Australian literature, usually identifies a consequence, managers exercise more control on smart workers' activities to sustain their productivity [16].
The second research area focuses on intensification of work, which refers to the increased "effort employees put into their jobs during the time that they are working" [17]. Two types of effort can be distinguished: "extensive" and "intensive" efforts [18]. Whereas extensive effort refers to the time workers spend on work tasks, intensive effort relates to physical and mental effort, and smart working may implicate an intensification of both [19,20]. In turn, work intensification may give rise to employees' stress, psychological impairment and work-family conflict [19]. Furthermore, the use of ICTs to remain connected to one's work, also in terms of the tendency to continuously monitor notifications or to check and react immediately to emails, can lead employees to work beyond normal working hours (e.g., during evenings, the night and weekend) [3], with the risk of having difficulties in detaching from work [21] and negative consequences on the work-family balance [3].
The aforementioned issue is closely related to the third research area, which is about work-life interference, in terms of both work-family conflict and family-work conflict. On the one hand, the opportunity to work anywhere and anytime generates the expectation, in colleagues at headquarters and in management, of being constantly connected and reachable, with negative consequences in terms of work-family conflict [7]. The expectation of immediate availability related to the use of ICTs undermines one of the most emphasized advantages of smart working-that is, the possibility to conciliate work and private life. On the other hand, given their temporal flexibility, smart workers receive more requests from family members that would not be made in the presence of more defined working hours [7], thus resulting in greater family-work conflict. Again, the findings concerning the relationship between smart working and the work-family interference are mixed [3]. On the one hand, the physical and temporal flexibility due to technology enables workers to switch between work and family domains, which can result in reduced work-family conflict [22]. On the other hand, the blurred boundaries between work and family domains may also increase inter-role distractions and interruptions, thus fostering role conflict [23]. In this respect, the meta-analysis conducted by Gajendran and Harrison [4] highlights that time is necessary to adapt to the changes associated with remote working: work-family conflict appeared to be lower in those who have been working remotely for more than one year than in those who have been adopting this work arrangement for less time. Additionally, past research has shown differences in work-life interference based on gender and parental responsibilities [24][25][26].
Finally, the last line of research addresses, the risk of isolation for smart workers. Previous studies showed that remote workers may experience feelings of social and professional isolation [27,28]. In fact, some authors proposed that virtual connections would not be able to compensate for the reduction in social contacts, in terms of face-to-face interactions and social proximity with colleagues and supervisors, which may lead to social isolation [7]. The frequency of remote working seems to affect the association between remote working and working relationships [4]. Specifically, spending more than 2.5 days per week working away from the office resulted in lower relationship quality. Moreover, physical distance from colleagues and supervisors may undermine opportunities for personal and professional development (e.g., career opportunities), leading employees to feel professionally isolated [27][28][29].
In line with this extensive body of literature, and also based on recent research [8], in this study, we conceptualize smart working as a "context" that shapes remote employees' work experience, rather than simply as a work arrangement related to specific work characteristics or individual outcomes (i.e., a predictor), such as autonomy or job satisfaction, respectively (please see the Discussion Section for more details). Specifically, we believe that smart working, by being associated, for example, with work intensification, work-family conflict and isolation, may interfere with the achievement of one's work goals and with employees' personal and professional development. Hence, the aim of this longitudinal study was to investigate whether the work arrangement (i.e., smart working vs. in-person working) affects the association between perceived work characteristics and employees' health and well-being over time. Specifically, building on the JD-R model [30,31], we posited that workload and social support (SS)-as relevant job demands and resources, respectively-would be longitudinally associated with exhaustion, a central aspect of job burnout [32]. Furthermore, we expected the positive association between workload and exhaustion to be stronger, while expecting the negative association between SS and exhaustion to be weaker, for smart workers. We describe in detail the theoretical rationale underlying our hypotheses in the following sections of the article.

Smart Working, Job Demands/Resources and Exhaustion: The Current Study
Although originally proposed to explain job burnout [31], the JD-R is a flexible theoretical model that specifies the relationship between classes of constructs-including job and personal characteristics, motivation, health and performance-and that can be applied in several organizational contexts (for a critical and updated review of the JD-R model, see Bakker and Demerouti [30], and Schaufeli and Taris [33]). According to the JD-R model, job characteristics can be classified either as job demands or job resources [30,33,34]. On the one hand, job demands are "those physical, psychological, social, or organizational aspects of the job that require sustained physical and/or psychological effort and are therefore associated with certain physiological and/or psychological costs" (p. 274) [30]. On the other hand, the term job resources refers to "those physical, psychological, social, or organizational aspects of the job that are functional in achieving work goals, reduce job demands and the associated physiological and psychological costs, or stimulate personal growth, learning, and development" (p. 274) [30]. According to the JD-R model, job demands and job resources trigger a health impairment process and a motivational process, respectively. In the health impairment process, as investigated in this study, chronic job demands (i.e., risk factors) require effort from workers and deplete their resources, both psychological and physical, possibly leading over time to psychophysical symptoms and job burnout [30,34], a "syndrome conceptualized as resulting from chronic workplace stress that has not been successfully managed" [35]. Furthermore, a lack of job resources (i.e., protective factors) precludes the achievement of one's work goals, which may give rise to job burnout and health complaints over time [33,36]. Hence, according to the JD-R model, burnout may particularly arise when workers are encumbered by elevated job demands and experience a dearth of job resources [33]. Specifically, in this study we concentrated on the exhaustion dimension of burnout, which refers to "feelings of being overextended and depleted of one's emotional and physical resources" (p. 399) [32]. Exhaustion is the central feature of burnout [32] and represents its energetic component [33]. Notably, exhaustion has a key role in the development of burnout [37], as well as in the health impairment (or energetic) [33] process of the JD-R model [30,38], as examined in this study. Finally, although burnout syndrome is frequently reported among healthcare professionals providing care to COVID-19 patients [39,40], recent research has shown that burnout is also a relevant phenomenon in the general working population during the COVID-19 pandemic [41][42][43].
When investigating antecedents of exhaustion, workload and SS are often identified in the literature as relevant job demands and resources in the general working population [44][45][46][47][48]. Not surprisingly, both workload and SS play a key role in several theoretical models of work-related stress and well-being, including the Demand-Control-Support model [49,50], the Effort-Reward Imbalance model [51], the Conservation of Resources theory [52], the Health and Safety Executive's Management Standards [53], as well as the JD-R model [34]. Although different conceptualizations exist [54], workload generally refers to the amount of work to be conducted in a given time [44]. In line with the health impairment process of the JD-R model, workload requires effort (e.g., to fulfil job demands) and drains workers' mental and physical resources, such as attention, energy or time [46]. Employees who are constantly exposed to high workload (and/or have insufficient opportunities for recovery) may develop exhaustion over time [55,56]. Empirical studies, both cross-sectional and longitudinal, supported the idea of an association between workload and negative outcomes for the individual [44], including exhaustion [46,57]. Hence, based on the health impairment process of the JD-R, according to which job demands may lead to stress outcomes over time, and in line with previous empirical results, we hypothesized that workload at T1 will positively predict exhaustion at T2. In particular, higher levels of workload will be associated with higher levels of exhaustion four months later, controlling for initial levels of exhaustion.
Social support refers to interpersonal support from other individuals at work, including supervisors and colleagues [58]. As a job resource, SS is functional in accomplishing tasks effectively and achieving work goals: for example, a worker may be helped by their supervisor or a colleague to overcome temporary challenges or difficulties in order to complete the work as required [59,60]. Hence, SS can contribute to preventing exhaustion in the long run [36]. Furthermore, SS can help employees to maintain or improve other valuable resources at work [45,52], including personal resources, such as self-efficacy and optimism [61,62], which can protect workers from negative outcomes of job stress [63]. Accordingly, employees who work in a supportive environment may feel more capable to control their work environment and to handle unforeseen events at work [30,62], which may result in lower levels of exhaustion over time [63,64]. In line with this reasoning, previous cross-sectional and longitudinal studies have shown a negative association between SS and exhaustion [59,65,66]. Overall, based on the idea that job resources help to protect workers against stress outcomes, and in line with previous empirical evidence, we hypothesized that SS at T1 will negatively predict exhaustion at T2. Specifically, higher levels of SS will be associated with lower levels of exhaustion four months later, controlling for initial levels of exhaustion.
Finally, in this study, we also hypothesized that the work arrangement might affect the longitudinal relationship between work characteristics, in terms of workload and SS, and exhaustion. On the one hand, smart workers may have difficulty in managing their workload due to technology-related problems (e.g., network connections, securing devices) [67], frequent interruptions [68], high email quantity and poor email quality [69], and an overwhelming amount of virtual meetings [70], which may ultimately result in information overload [71] and loss of control over their work flow (see also [72] for a review). Additionally, smart workers may find it difficult to focus on their work due to family demands [7], and they usually have to perform multiple tasks at a time (both workand family-related), or switch from one task to another (i.e., multitasking) [73]. In this perspective, smart workers may be less able to effectively manage their current workload. In line with the health impairment process of the JD-R model, this implies a greater effort to complete their tasks, higher psycho-physiological costs for the individual and, eventually, higher levels of exhaustion over time. Hence, we hypothesized that the positive association between workload at T1 and exhaustion at T2 will be stronger for smart workers.

Hypothesis 3 (H3)
. Work arrangement will moderate the association between workload at T1 and exhaustion at T2, which is expected to be stronger for smart workers.
On the other hand, with respect to SS, it should be considered that communications, tasks and interpersonal collaborations are mostly mediated by technology when working from home [8]. Hence, smart working may give rise to social and professional isolation [27,28], thus weakening employees' ability to establish direct and enduring relationships with colleagues and supervisor [74,75]. Similarly, compared to in-person workers, smart workers have fewer opportunities for informal, face-to-face interactions with col-leagues and supervisors, which are essential for the development of emotions, knowledge, shared values and mutual trust [7,16]. Hence, by hindering social and professional exchanges at work, smart working may result in poor and less effective social support, which does not fully meet remote workers' specific needs and expectations [16]. For example, smart workers may be less able to obtain from colleagues or supervisors specific resources or information, tailored to their unique needs, which would be useful to overcome workrelated problems and successfully complete their tasks [28,76]. Hence, by receiving less effective support from colleagues or supervisors, smart workers may benefit less from SS. This implies that smart workers may be less able to achieve their work goals and less protected against exhaustion. Therefore, we hypothesized that the negative association between SS at T1 and exhaustion at T2 will be weaker for smart workers.

Hypothesis 4 (H4)
. Work arrangement will moderate the association between SS at T1 and exhaustion at T2, which is expected to be weaker for smart workers.
Finally, past research has shown that demographic variables (i.e., gender and age) and work-related factors (i.e., type of contract) may relate to burnout and exhaustion. Concerning gender, past studies suggested that women tend to experience higher levels of burnout-and, specifically, exhaustion-than men [77,78]. A possible explanation may involve gender roles, with women experiencing higher levels of work-family conflict, a wellknown risk factor for exhaustion [79], due to conflicting demands from work and home [78]. Similarly, age has been shown to be associated with job burnout, although results are mixed. For example, a positive association between age and burnout may exist: the components of burnout require time to emerge, since they arise from chronic workplace stress that have not been successfully managed [35]. Accordingly, burnout and its components may manifest in the later stages of one's career, as a result of cumulated effects of prolonged stress [80]. It is also possible that age may be negatively associated with burnout and its components, given that workers tend to develop effective coping skills over time or because individuals experiencing higher burnout tend to leave their job earlier [81,82]. Interestingly, the association between age and burnout may be affected by gender [80,83]. Finally, regarding the type of contract, previous studies suggest a relationship between temporary employment and exhaustion, albeit with some inconsistent results [84][85][86]. For example, it is possible that temporary employees experience unfavorable employment conditions, including less job security, low wages and involuntary part-time, which may have negative consequences on workers' health [87,88]. In the light of a possible association between gender, age, type of contract and exhaustion, the hypothesized associations were examined both including and omitting these demographic and work-related characteristics.

Participants and Procedures
This study was conducted during the COVID-19 pandemic in Italy. Workers from several organizational context were asked to fill out an online questionnaire aimed at determining the constructs under investigation. They were also informed that a second questionnaire would be administered four months later. The first wave (i.e., Time 1, T1) occurred between the end of October 2020 and the first half of November 2020, while the second wave (i.e., Time 2, T2) occurred between the end of February 2021 and the first half of March 2021. Before taking part in the study, all participants provided written informed consent. The project was approved by the Ethical Committee for the Psychological Research of the University of Padua, Italy (protocol n. 3842). Overall, 295 participants took part in the first wave of the study at T1, and 185 (62.7%) took part in both waves (i.e., T1 and T2). There were no differences in main demographics or study variables between those who dropped out of the study and those who completed the study.

Self-Report Measures
The questionnaire included the following measures: Exhaustion was determined at both T1 and T2 by a scale taken from the Italian adaptation of the Maslach Burnout Inventory [89]. The scale included nine items aimed at detecting the feelings of being emotionally exhausted by one's work. A sample item is: "I feel emotionally drained from my work". The items were rated on a response scale from 1 (never) to 6 (always), and higher scores reflected higher levels of exhaustion. Cronbach's alpha was 0.93 at T1 and 0.92 at T2.
Workload was assessed at T1 using four items, reflecting mostly quantitative workload [54], taken from the Q u -Bo test, a self-report, standardized questionnaire developed for the Italian context [90]. An example of an item is: "Your job requires you to work very fast". The 6-point response scale ranged from 1 (strongly disagree) to 6 (strongly agree), with higher scores reflecting higher levels of workload. Cronbach's alpha was 0.84 at T1.
Social support was determined at T1 using a scale taken from the Safety at Work (SAPH@W) Questionnaire [91], an instrument aimed at assessing perceived safety at work during the COVID-19 pandemic. The items were rated on a response scale from 1 (not at all) to 10 (completely), with higher scores referring to high levels of SS. Cronbach's alpha was 0.77 at T1.
Work arrangement was detected by asking each participant to indicate their work arrangement at T1. The item discriminates those who work in-person from those who work-in whole or in part-remotely (i.e., smart workers).

Data Analysis
First, to evaluate the psychometric properties of the self-report instruments adopted in the study, a confirmatory factor analysis (CFA) was performed using the maximum likelihood estimation with robust standard errors and a scaled test statistic [92]. In a fourfactor model, exhaustion (at both T1 and T2), workload and SS were measured by the respective scale items. Residuals for indicators that were repeated over time (i.e., items of the exhaustion scale) were freely estimated [93]. Since χ 2 is affected by the sample size, additional fit indices were considered: the root mean square error of approximation (RMSEA), the comparative fit index (CFI) and the standardized root mean square residual (SRMR). A model shows a good fit to data if χ 2 is nonsignificant. Additionally, values close to or less than 0.08 for RMSEA and SRMR, as well as values close to or greater than 0.90 for CFI, indicated an acceptable fit [94]. Construct validity (i.e., convergent and discriminant validity) and reliability were assessed using the average variance extracted (AVE) [95] and coefficient ω [96], respectively. AVE was used to assess convergent as well as discriminant validity. An AVE greater than 0.50 indicated adequate convergent validity. Additionally, discriminant validity was established if the AVE of any two constructs was greater than their squared correlation, which reflected their shared variance [95]. For ω, values greater than 0.70 suggested satisfactory reliability [97].
Then, the relationships hypothesized in the study were tested using moderated multiple regression analysis [98]. In Model 1 (M1), exhaustion at T2 was regressed on exhaustion, workload, SS and work arrangement at T1. In Model 2 (M2), the two interaction terms between workload and work arrangement, as well as between SS and work arrangement, were also included. Models 3 (M3) and 4 (M4) were similar to M1 and M2, respectively, with the exception that the focal relationships were estimated controlling for the effect of gender, age and type of contract. The variables included in the models (except dichoto-mous variables) were mean-centered. If significant interactions were found, then a simple slope analysis was conducted to determine whether the associations between predictors at T1 (i.e., workload or SS) and exhaustion at T2 were significant across different work arrangements (i.e., in-person vs. smart working). Finally, to enable easier interpretations, significant interactions were plotted following Aiken and West [98]. Data analysis was conducted using R software (version 4.1.2, The R Foundation for Statistical Computing, Vienna, Austria) [99]. In particular, the CFA was performed through the lavaan package version 0.6-10 [92], whereas the AVE and the coefficient ω were computed using the semTools package version 0.5-5 [100] for R software.

Confirmatory Factor Analysis
First, a CFA was performed to evaluate the psychometric properties of the self-report questionnaires administered in the study. The model showed a less than adequate fit to data: χ 2 (237) = 542.30, p < 0.001; RMSEA = 0.083, CFI = 0.876, SRMR = 0.067. Modification indices indicated that an error covariance-between items 1 and 2 of exhaustion, which shared a similar wording-should be freely estimated at both time points. A new CFA was conducted, and fit indices showed an acceptable fit to data: χ 2 (235) = 480.15, p < 0.001; RMSEA = 0.075, CFI = 0.901, SRMR = 0.068. Furthermore, the revised model showed a better fit to data compared to the original model, ∆χ 2 (2) = 31.65, p < 0.001. The AVE ranged from 0.58 (exhaustion at T2) to 0.67 (SS), whereas coefficient ω ranged from 0.80 (SS) to 0.92 (exhaustion at T1). The correlation between the latent factors ranged from −0.48 (between SS and exhaustion at T1) to 0.60 (between exhaustion at T1 and exhaustion at T2), and AVE for each pair of latent factors was greater than their squared correlation. Overall, the self-report questionnaires administered in this study showed good psychometric properties in terms of factor structure, construct validity (i.e., convergent and discriminant validity) and reliability.

Hypothesis Testing
Correlations and descriptive statistics are presented in Table 1. All variables had univariate skewness and kurtosis that fell within the acceptable range of ±2.0 and ±7.0, respectively [101]. As expected, there was a positive, large-sized correlation [102] between exhaustion at T1 and exhaustion at T2 (r 183 = 0.61, p < 0.001). Interestingly, a positive, small-to-medium-sized correlation between workload at T1 and exhaustion at T2 emerged (r 183 = 0.23, p < 0.01), as well as a negative, medium-to-large-sized correlations between SS at T1 and exhaustion at T2 (r 183 = −0.39, p < 0.001). With respect to the control variables, no differences emerged in the variables under investigation by gender, age and type of contract. The results of the moderated regression analyses (i.e., M1 to M4) are shown in Table 2. In M3 and M4, there was no association between control variables at T1 and exhaustion at T2, with the exception of age, which was positively associated with exhaustion over time. However, the results did not substantially differ across models that included or omitted control variables. Hence, the results of the more parsimonious M1 and M2 are discussed hereafter [103], although the results of all the tested models are shown in Table 2.  In M1, the predictors at T1 accounted for 40% of the variance in exhaustion at T2 (R 2 = 0.40, F(4, 180) = 29.48, p < 0.001). In this model, exhaustion at T1 positively predicted exhaustion at T2 (b = 0.52, SE = 0.07, p < 0.001, β = 0.53), suggesting that exhaustion is relatively stable across waves. Workload at T1 did not predict exhaustion at T2 (b = 0.02, SE = 0.06, ns, β = 0.02), whereas SS at T1 negatively predicted exhaustion at T2 (b = −0.09, SE = 0.04, p < 0.05, β = −0.17). Hence, Hypothesis 2 was supported, but Hypothesis 1 was not. In M2, the interaction terms accounted for an additional 2.5% of the variance in exhaustion at T2, F change (2, 178) = 3.81, p < 0.05. The interaction between workload and work arrangement was marginally significant (b = 0.21, SE = 0.11, p < 0.10, β = 0.22), whereas the interaction between SS and work arrangement was significant (b = 0.15, SE = 0.07, p < 0.05, β = 0.27). Simple slope analysis revealed that the association between the workload at T1 and exhaustion at T2 was positive and marginally significant for smart workers (b = 0.15, SE = 0.09, p < 0.10, β = 0.16), but not significant for in-person workers. Furthermore, the association between SS at T1 and exhaustion at T2 was negative and significant for in-person workers (b = −0.15, SE = 0.04, p < 0.001, β = −0.28), but not significant for smart workers. The interaction between workload and work arrangement is shown in Figure 1, whereas the interaction between SS and work arrangement is represented in Figure 2. Smart working strengthened the positive-albeit nonsignificant-association between workload and exhaustion but weakened the negative association between SS and exhaustion. Overall, Hypothesis 4 was supported, and Hypothesis 3 was partially supported.

Discussion
In this longitudinal study, we investigated whether smart working affects the association between perceived work characteristics and workers' health and well-being over time. More specifically, building on the JD-R model, we hypothesized that workload, as a job demand, would positively predict exhaustion, whereas SS, as a job resource, would negatively predict exhaustion. We also hypothesized that work arrangement would moderate these longitudinal associations. In particular, we expected that the positive association between workload and exhaustion would be stronger, while the negative association between SS and exhaustion would be weaker, for smart workers.
The results partially support our hypotheses. Contrary to our expectations, workload at T1 did not predict exhaustion at T2 (i.e., four months later), controlling for exhaustion at T1. Conversely, SS negatively predicted exhaustion over time, which implies that higher levels of SS at T1 were associated with lower levels of exhaustion at T2, controlling for initial levels of exhaustion. With respect to the moderating role of smart working, the

Discussion
In this longitudinal study, we investigated whether smart working affects the association between perceived work characteristics and workers' health and well-being over time. More specifically, building on the JD-R model, we hypothesized that workload, as a job demand, would positively predict exhaustion, whereas SS, as a job resource, would negatively predict exhaustion. We also hypothesized that work arrangement would moderate these longitudinal associations. In particular, we expected that the positive association between workload and exhaustion would be stronger, while the negative association between SS and exhaustion would be weaker, for smart workers.
The results partially support our hypotheses. Contrary to our expectations, workload at T1 did not predict exhaustion at T2 (i.e., four months later), controlling for exhaustion at T1. Conversely, SS negatively predicted exhaustion over time, which implies that higher levels of SS at T1 were associated with lower levels of exhaustion at T2, controlling for initial levels of exhaustion. With respect to the moderating role of smart working, the results show that the interaction between workload and work arrangement is marginally

Discussion
In this longitudinal study, we investigated whether smart working affects the association between perceived work characteristics and workers' health and well-being over time. More specifically, building on the JD-R model, we hypothesized that workload, as a job demand, would positively predict exhaustion, whereas SS, as a job resource, would negatively predict exhaustion. We also hypothesized that work arrangement would moderate these longitudinal associations. In particular, we expected that the positive association between workload and exhaustion would be stronger, while the negative association between SS and exhaustion would be weaker, for smart workers.
The results partially support our hypotheses. Contrary to our expectations, workload at T1 did not predict exhaustion at T2 (i.e., four months later), controlling for exhaustion at T1. Conversely, SS negatively predicted exhaustion over time, which implies that higher levels of SS at T1 were associated with lower levels of exhaustion at T2, controlling for initial levels of exhaustion. With respect to the moderating role of smart working, the results show that the interaction between workload and work arrangement is marginally significant. Specifically, smart working strengthened the positive-albeit nonsignificant-association between workload at T1 and exhaustion at T2, which was positive and marginally significant for smart workers, but not significant for in-person workers. On the contrary, the results show that the interaction between SS and work arrangement is significant: smart working weakens the negative association between SS at T1 and exhaustion at T2, which is negative and significant for in-person workers, but non-significant for smart workers. Notably, these results did not change after controlling for age, gender and type of contract.

Theoretical Implications
Overall, we believe our study offers a novel insight into the complex phenomenon of smart working, with both relevant theoretical and practical implications. With respect to the former, as pointed out by Wang et al. (2020) [8], previous research in the field predominantly followed two approaches. The first focuses on identifying the types of work that are most suited to smart working (e.g., jobs involving complex tasks or low task interdependence) [104], while the second concentrates on how engaging in smart working influences employees' outcomes through perceived work characteristics (e.g., smart working is associated with greater perceived job autonomy, which positively influences job satisfaction and performance) [4]. Although these two approaches provide extremely useful information about smart working, in this study, we went a step further. Specifically, in line with the third approach described by Wang et al. (2020) [8], we considered smart working as a "context" that influences the meaning of work characteristics and the ability of workers to cope effectively with demands. From this perspective, smart working does not necessarily affect employees' well-being and performance through perceived work characteristics (e.g., increased autonomy or work-family balance). Instead, to foster performance and well-being, what matters is that work characteristics fit the new way of working as well as smart workers' specific needs and expectations. On the contrary, "unintended outcomes might arise when virtual work characteristics fail to meet individual and/or task requirements" (p. 22) [8]. Notably, as noted by the authors, this approach is particularly useful in the context of the current COVID-19 pandemic, in which smart working is no longer an option, but a necessity, and the meaning of some job demands/resources may be considerably influenced by the extraordinary pandemic context [8].
Interestingly, in our study, smart workers did not report higher levels of workload (M = 4.05, SD = 1.10) than in-person workers (M = 3.85, SD = 1.23), t(182.59) = −1.16, p = 0.25. They also did not report lower levels of SS (M = 7.15, SD = 1.91) than in-person workers (M = 7.11, SD = 2.16), t(182.72) = −0.12, p = 0.90. Contrarily, and in line with the above reasoning, our research has shown that smart working, as a "context" that shapes remote employees' work experience, affected the association between job demands/resources and exhaustion. On the one hand, the association between workload and exhaustion over time was nonsignificant for in-person workers, but a positive (albeit marginally significant) longitudinal association between workload and exhaustion emerged for smart workers. A possible explanation is that smart working is associated with both work intensification and extensification [19,20]; due to difficulties in managing their workload [67][68][69][70], smart workers may end up investing more effort-physical and mental-and spending more time in their work to perform their tasks effectively and achieve their goals. Additionally, given the blurred boundaries between work and family domains, they may also experience higher levels of work-family/family-work conflict [3,105]. This may result in greater inability to switch-off from work and impaired recovery experiences [106], possibly leading to exhaustion over time [56].
On the other hand, our study showed that, as expected, there was a negative association between SS and exhaustion over time for in-person workers, but the longitudinal association between SS and exhaustion was nonsignificant for smart workers. A possible explanation is that in-person workers have many opportunities for social interactions with colleagues and supervisors at work, making it easier for them to ask for and ob-tain effective support to complete their tasks successfully, which leads to a lower risk of burnout. Conversely, since remote working hinders social and professional exchanges at work [16,27,28], traditional strategies of SS are less effective for smart workers, who have specific needs and expectations. These include, for example, the need for instrumental support from supervisors/colleagues to accomplish tasks when working from home or to handle problems related with ICTs [8]. Smart workers might also need emotional support from supervisors/colleagues to overcome social isolation and to maintain bonds with other work team members [107]. Of course, it is also possible that traditional forms of SS simply cannot be applied to smart workers, and new strategies need to be implemented (we briefly describe this aspect in the practical implications section of this article).
Overall, our results are in line with the JD-R model, with some unexpected and intriguing exceptions. In fact, contrary to the model assumptions, the association between workload-a job demand-and exhaustion over time was not significant for in-person workers. Similarly, SS-a job resource-was not associated with exhaustion over time in smart workers. However, it should be noted that the JD-R model is a descriptive model that outlines the relationships between classes of constructs, including job demands/resources and health/motivation, without describing the underlying theoretical mechanisms [33]. Hence, following Bakker and Demerouti [30], we draw on the conservation of resources (COR) theory [52,108] to outline a more comprehensive discussion of our findings.
According to the COR theory, individuals seek to acquire, retain and protect things they value, namely, resources. These include, for example, conditions (e.g., health, employment), objects (e.g., technological equipment), personal characteristics (e.g., self-efficacy, key skills) or energy resources (e.g., energy, time). In this context, stress occurs when resources are actually lost or threatened with loss, or when people do not acquire adequate resources in front of significant resources investment. A principle of the COR theory is that individuals must invest resources to protect/recover from resource loss, as well as to gain resources. Hence, individuals with more resources are more capable of resource gain and less vulnerable to resource loss, whereas those with limited resources are more vulnerable to resource loss and less capable of resource gain. Another principle of COR is that resource gains become more salient in the context of resource loss, meaning that resource gains are especially valued in conditions of heavy losses (i.e., the gain paradox principle) [108]. Finally, it should be considered that resources do not exist individually but rather aggregated in packs (i.e., resource caravans).
In line with the COR theory, workload may directly deplete employees' resources, such as time or psycho-physical energy, or it may interfere with the acquisition of new resources, such as the case when a worker does not have adequate time to cultivate social relationships (e.g., with family members or friends), thus resulting in exhaustion [46]. When confronted with a high workload, workers may invest additional resources to protect against resource loss: for instance, they may use resources to cope effectively with an elevated job demand. At this stage, workers may also seek SS, which, in a COR perspective, is desired on its own (as a resource) and because it contributes to the maintenance of large resource reservoirs (e.g., by being functional in obtaining other useful resources) [52], thus helping individuals to cope with demands as well as prevent exhaustion. This point was clarified by Halbesleben et al. [109], who argued that resources hold value for an individual to the extent that they are perceived to be helpful in achieving one's goals, meaning that the value of a specific resource can vary significantly across different contexts. From this perspective, the employees who thrive are not necessarily the ones with more resources, but the ones that are best able to allocate their resources to maximize their fit with the work environment [109]. As mentioned before, smart workers may actually be supported by colleagues and supervisors, but they may benefit less from the support they receive, which is not tailored to their specific needs and expectations. From this perspective, by being scarcely useful both in acquiring other resources and achieving one's work objectives, SS may be a less valuable resource for smart workers. Hence, in-person workers, who benefit most from effective support from colleagues/supervisors, are less vulnerable to resource loss and, consequently, exhaustion. Contrarily, smart workers, who derive little benefit from less effective SS at work, are more vulnerable to resource loss and less protected against exhaustion. This explains why, according to the COR theory, SS is not associated with exhaustion over time in smart workers.
However, to better understand the lack of a longitudinal association between workload and exhaustion for in-person workers, it is necessary to consider the context in which the study occurred, characterized by the COVID-19 pandemic. Specifically, from a COR perspective, the COVID-19 outbreak resulted in considerable resource loss for individuals, in terms of economic (e.g., temporary loss of income), social (e.g., loss in relationships with family and friends), leisure (e.g., loss of opportunity to travel), psychological (e.g., loss on control over one's future) and health-related (e.g., loss in physical and mental health) resources [110,111]. Compared to remote working, office working is associated with a pool of relevant resources (i.e., resource caravan) [109], such as social interactions and rewards, opportunities for professional development and clear boundaries between work and home, which can facilitate physical and psychological detachment from work [109]. Furthermore, in line with the gain paradox principle, the resource gains associated with in-person work are especially valuable during the COVID-19 pandemic, to compensate for actual resource loss or prevent further loss. For example, as noted by Wang et al. [8], being socially connected with others at work may be especially relevant during the COVID-19 outbreak, in which social gatherings are made difficult by restrictions and social distancing. In line with this reasoning, since office working is associated with a pool of relevant resources, and given that individuals with greater resources are less vulnerable to resource loss, inperson workers may be more resilient and better able to cope with job demands (e.g., high workload) compared to smart workers, which explains the lack of a longitudinal association between workload and exhaustion for the former during the COVID-19 pandemic.
All in all, we believe that the approach proposed in this study, as well as its findings, will contribute to providing a starting point for future research in the field of work and organizational psychology. Although the COVID-19 pandemic has produced a sharp acceleration in the adoption of smart working, hybrid work arrangements-that combine office work and smart working [112]-will be increasingly prevalent in the future [8]. Hence, forthcoming research needs to consider both situational (e.g., social, organizational and policy-related) as well as individual (e.g., personal characteristics) factors that may shape hybrid workers' experiences at home versus the office.

Limitations and Future Research Directions
This study has some limitations. First, it is possible that the four-month time-lag between the two measurement occasions represents a rather short time frame to investigate the longitudinal association between workload/SS and exhaustion, which stems from prolonged workplace stress that has not been adequately managed. This also offers a potential alternative explanation for the lack of association between workload and exhaustion. However, the choice of this time-lag is consistent with previous empirical research in the field [65,113] and is based on the assumption that the impact of job demands/resources on health outcomes (e.g., well-being, psychophysical strain) may be weakened if the time interval between waves is too long [114,115]. Second, the longitudinal reversed effect of exhaustion on workload/SS was not investigated in this study. It can be argued that workers with higher levels of exhaustion may also experience higher levels of workload and lower levels of SS over time, because they have no or limited resources (e.g., psychological or physical energies) to handle their job demands or seek support from others, respectively. Hence, although our hypotheses are in line with relevant models of work-related stress-including the JD-R model-and previous empirical research [116][117][118], future studies could examine the reciprocal association over time between workload/SS and exhaustion (e.g., using a cross-lagged panel model) across different work arrangements. Third, our sample levels of exhaustion were moderate-below the scale midpoint-at both time points, which is perhaps due to the relatively low mean age of the study participants [80]. Hence, further research is recommended to replicate and extend our findings among aging workers and in occupational contexts characterized by chronic interpersonal stressors and elevated levels of work-related stress, such as, for example, teachers. Fourth, the two measurement occasions substantially overlapped with two different stages of the COVID-19 pandemic in Italy, namely, the second and third waves [119]. Hence, it is possible that external contingencies may have affected our findings. However, the first measurement occasion was conducted approximately eight months after the emergence of the new coronavirus, and exhaustion was relatively stable across waves, suggesting that the pandemic should not be responsible for major changes in exhaustion over time. All in all, we believe our study offers an interesting insight into smart workers' work experiences and well-being during the COVID-19 outbreak, in which millions of people around the world were suddenly forced to work from home, giving rise to a "global experiment" of smart working [8]. Fifth, in this study, we mostly focused on quantitative workload-that is, the amount of work to be conducted in a given time, the most common aspect of workload considered in the literature [44]. However, future research could also address qualitative workload (e.g., the necessity to constantly solve new problems at work) [120], which reflects the "difficulty or complexity of the job, for which the worker is not trained or does not have enough resources to deal with" (p. 2) [120], as well as its consequences over time across different work arrangements. Similarly, we also adopted an overall measure of SS in this study. In line with recent research [58], we encourage researchers to distinguish between different types of social support in future research, including, for example, instrumental, emotional, and informational support. Sixth, in this study, we exclusively relied on self-report measures. Although a temporal separation between the measurement of the independent and dependent variables should alleviate concerns about common method bias in the current study [121], future research may benefit from the integration of different measurement methods, including observer rating of demands/resources or objective measures of ill-being (e.g., biomarkers of stress, such as pro-inflammatory cytokines or hair cortisol) [122][123][124]. Finally, although not a limitation per se, we acknowledged that the COVID-19 pandemic may have influenced our results. By conceptualizing smart working as a "context" that shapes employees' work experience, it should be considered that this experience depends on whether remote working is created by the organization (e.g., to reduce costs), sought by workers (e.g., to increase efficiency and work performance, to manage the work and family domains and to help the environment by saving energy and reducing traffic as well as air and noise pollution) or both [2,3]. However, during the COVID-19 pandemic, smart working was not considered an option, but rather a necessity to contain the spread of SARS-CoV-2 (e.g., by reducing physical proximity and social interactions) [8]. Hence, additional research is warranted to replicate and extend our results.

Practical Implications
Despite the aforementioned limitations, we believe our study has several practical implications. Conceptualizing smart working as a context that shapes remote employees' work experience, our study suggests that organizations should primarily adopt good management practices aimed at targeting those working conditions-both physical and psychosocial-associated with stress among smart workers (i.e., primary prevention). On the one hand, organizations should encourage supervisors to develop new skills that effectively support smart workers and help them to manage their workload. These include, for example, e-leadership [125], which encompasses communication skills, in terms of communication clarity and management of communication flow (e.g., email, virtual meetings); technological skills, such as technological knowledge, which can help workers to cope with technology-related problems (e.g., network connections); and trustworthiness. Similarly, supervisors should be encouraged not to shift their focus from supporting smart workers' performance to excessive monitoring and control of their work activities [16]. Moreover, with the aim of containing the tendency to exceed normal working hours, organizations should define specific policies that limit the use of work-related technologies during leisure time [126]. On the other hand, smart workers might need emotional support from supervisors/colleagues to reduce feelings of social isolation and maintain social bonds with other team members. Hence, attention should be devoted to creating opportunities for non-task interactions to allow social connections and continuity among team members [107]. Finally, supervisors should provide clear information about work objectives as well as personal and career development, to reduce feelings of professional isolation resulting from the central workplace's physical distance. Additionally, our study suggests that interventions should also be aimed at the identification and training of those workers who are particularly at risk of burnout, including smart workers (i.e., secondary prevention). From this perspective, interventions could be directed towards the replenishment of resources that are depleted at work (e.g., by fostering recovery), the proactive modification of job characteristics (e.g., through job crafting) and the effective management of the boundary between work and non-work domains [127]. First, based on the idea that resources built during leisure time may spill over into the working life, individuals should be encouraged to proactively create changes in their non-working lives (i.e., off-job crafting), whose favorable effects may in turn manifest at work [127]. For example, employees should learn the benefits of mentally "switching off" from one's work-related thoughts during off-job time (i.e., detachment), as well as the importance of being closely related and emotionally connected to relevant others-such as family members or friends-when not working (i.e., affiliation) [128]. A recent study suggested that detachment and affiliation may be useful in replenishing psychophysical resources usually depleted at work, thus contributing to preventing burnout [127]. Second, interventions should be aimed at promoting job crafting, meaning that workers are encouraged to proactively optimize their own job design [129], for example, by increasing structural job resources (e.g., seeking greater clarity of tasks, in which roles and tasks are well-defined). Recent research during the COVID-19 outbreak has shown that job crafting can be useful in dealing with the negative effects of work overload, especially when working from home [130]. Finally, since the worktime boundary between work and private life could gradually disappear when working from home [131], interventions should be aimed at helping workers to preserve a clear physical or psychological distinction (i.e., segmentation) between work and non-work domains (e.g., walking around the block as an alternative commuting strategy when working from home) [132].

Conclusions
All in all, our study found that the longitudinal association between workload and exhaustion was positive-although marginally significant-for smart workers, but nonsignificant for in person workers. On the contrary, the longitudinal association between SS and exhaustion was negative for in-person workers, but nonsignificant for smart workers. Stated differently, workload appeared to be a more critical job demand, whereas SS was a less valuable resource, for smart workers during the COVID-19 outbreak. Overall, these findings are consistent with the JD-R model [30,31], albeit with aforementioned exceptions, as well as the conceptualization of smart working as a "context" that shapes remote employees' work experience [8]. This study suggests that, to support smart workers' well-being and performance, work characteristics-both physical and psychosocial-should fit the new way of working as well as remote workers' specific needs and expectations. Hence, possible negative aspects of smart working-as implemented in the pandemic-related emergency-need to be thoroughly considered and addressed by organizations. However, smart working will probably continue to play a central role in the "new normal" after the COVID-19 pandemic. Hence, organizations should encourage supervisors to develop new skills and leadership styles (e.g., e-leadership) [125] to support smart workers and help them to manage their workload effectively. Similarly, organizations should encourage supervisors to foster trust and reduce isolation, both social and professional, among team members.