Following the JD-R model (Demerouti et al., 2001) work characteristics for the development of work related burnout can be grouped into two broad categories: (1) job demands, which are psychological, physical, social, and organizational factors of work, which require a physical and mental effort that usually lasts for a longer period of time. Work-related demands include: emotional and physical stress, shift work and conflicts. They are not necessarily negative or cause negative effects. Rather, they can become stressors with negative consequences if coping with the demand requires a high level of effort over the long term and there is no adequate recovery phase, and (2) job resources, which represent the psychological, physical, social and organizational working conditions, which are relevant for the achievement of work-related goals, alleviate work demands, and promote personal growth and development (Demerouti et al., 2001). Two mental processes are assumed by the model: a health-impairment process, where stressors lead to emotional exhaustion in the long run and thus to impaired well-being; and a motivational process through which job resources, such as autonomy or social support, foster motivation and work engagement (Demerouti et al., 2001).

Meta-analytic evidence supports the validity of the model and its suitability to assess employee well-being (Lesener et al., 2019). According to the JD-R model job resources may buffer the impact of job demands on job strain (Bakker and Demerouti, 2007). An extension to the JD-R model is the inclusion of personal resources (Xanthopoulou et al., 2007). Personal resources, according to Hobfoll et al. (2003, p. 632), are “aspects of the self that are generally linked to resiliency”. They serve a similar function as job resources; when job resources are in short supply, personal resources can counteract that shortage (Bakker and Demerouti, 2017). Personal resources, such as occupational self-efficacy have been shown to alleviate stress (Grau et al., 2001).

Data analyzed in this study were collected in Germany from 2020–03–24 until 2020–05–17, shortly after the lockdown in Germany was implemented. Participants were able to register with their email address after confirming a consent form. They then received an e-mail with a link to the survey. On the survey page, demographics and job-related questions were initially collected. The next pages included items regarding working conditions, their experience during the COVID-19 situation, and well-being.

Follow-up surveys took place every week (first five surveys). At the beginning, participants were asked whether they had worked in the previous week and whether anything had changed at work. Here, the repeated survey of working conditions wasn't offered if no changes regarding their job had taken place. Participants were reminded of missed surveys at irregular intervals (up to five times). We always offered the opportunity to opt-out of participation in each invitation and reminder email.

The study received a positive assessment from the institute of psychology of the University of Duisburg-Essen's Ethics Committee.

In this study we used the weekly measurements of well-being to model our different groups of well-being trajectories. Job demands (such as telework and work intensification), as well as job and personal resources were measured at T1, the initial survey time point.

The sample consisted of 642 participants who completed the initial survey, of whom 453 (70.6%) were women. Participants had a mean age of 39.63 (SD = 12.81; Range: 18–79) and were mostly well-educated: Highschool or below: 21.3%; Apprenticeship: 12.3%; Bachelor: 17.1%; Master: 34.4%; Doctorate: 10.4%; Other: 4.4%. 392 (61.1%) of the participants had a full-time employment, while 183 (28.5%) worked part-time. The remaining 67 (10.4%) were in marginal and/or irregular employment or doing an apprenticeship. Job tenure of the sample was 8.12 years on average (SD = 9.56). One hundred sixty-five (25.7%) participants had leadership responsibilities.

While 411 (64.0%) of the participants stated, that their tasks were possible to be carried out from home, 378 (58.9%) had previous access to telework. In the initial time (T1) of the COVID-19 pandemic and its associated lockdown in Germany 403 (62.8%) of the participants were working from home. Additional descriptive statistics are available in Table 1.

We used R version 4.1.3 (R Core Team, 2021) and employed a three-step approach as described in Asparouhov and Muthén (2014). We decided to use a GBTM-approach [often referred to as latent class growth analysis (LCGA)] to assess the trajectories of our latent classes. In comparison to the often employed growth mixture modeling (GMM), which assumes the existence of distinct subpopulations, GBTM doesn't estimate within-group variability, and thus intends to approximate trajectories across population members (Nagin and Odgers, 2010). In order to deal with varying numbers of assessment time points as well as differing start times in our survey, we used the R package LCMM (version 1.9.3) to specify our models, which uses maximum likelihood estimation (Proust-Lima et al., 2017). To select the best fitting model, we first compared the BIC of several 1-class models with different link-functions (beta, linear, I-splines with varying number of knots) as well as varying specifications of time, allowing estimation of linear, quadratic and cubic trajectories. We used the gridsearch function of the package lcmm for automatic grid search, with 50 departures from initial values and 100 maximum iterations. In order to assess predictors of group assignment, measured at T1, we applied multinomial regression analysis, using multinom from the nnet (version 7.3-16) package (Venables and Ripley, 2002).

Table 2 displays the descriptive statistics and correlations among our study variables at T1. Our outcome well-being, measured with the WHO-5 well-being index, correlates negatively with job demands in terms of work intensification; and positively with personal and job resources. Job demands in terms of work intensification additionally correlate negatively with the job resource flexitime and positively with age and job tenure. As expected personal and job resources appear to be intercorrelated. Interestingly we find a negative relationship between gender and our outcome well-being and the personal resource self-efficacy as well as all job resources, except for social support, indicating slightly lower (in the case of social support higher) values for women in this study. Table A1 in the Appendix displays the correlations of well-being, as well as our focal variables at T1 with the lagged well-being measurements.

Table 3 displays the fit indices and entropy of our estimated models. Three groups should fit the data well enough, as can be seen in the elbow plot in Figure 2. Entropy was slightly lower from the two-class solution (0.76) but still acceptable with 0.75. The small decrease in BIC doesn't justify the risk of overfitting, by choosing a model comprised of more classes. Additionally we looked at the posterior probability of class membership assignment, which was above 0.8 on average for each of the classes.

The three-class model clearly differentiated three groups, which are displayed in Figure 3 using observed values with smoothed average scores. Group: low level of well-being consisted of 22% of the sample and is represented by an initially ever so slightly decreasing level of well-being as measured by the WHO-5 well-being index. Participants in this group on average had a higher level of well-being at the earlier time points, which deteriorated in the first couple of weeks, to slightly improve in the middle of April and consequently remain low throughout the the rest of the study. Group: medium level of well-being had the highest percentage of participants with 45% and is comprised of individuals with on average consistently medium levels of well-being. Similar to the low level of well-being group, we find an initial slight decline of well-being in this group, which recovers a little bit in the middle of April, to then deteriorate across the rest of the measurement time points. Group: high level of well-being consistently showed the highest level of well-being on average. This group displayed a continuous improvement of well-being up until May. Thereafter it declined slightly, but still stayed above the initial levels until the end of our study.

Table 4 displays the result of our multinomial regression analysis, predicting trajectory group membership (AIC = 1, 252.32).

Building upon the accumulation and adjustment models of stress (see Zapf et al., 1996) we had assumed to identify varying trajectory profiles of mental well-being, but had to discard all of them, except for a rather stagnant, albeit slightly improving trajectory of well-being for the high well-being group and deteriorating trajectories for the medium and low well-being groups. Using the accumulation and adjustment models to explain these findings would suggest, that during the initial weeks of the pandemic, the participants of this study did not experience much of an increase in stressors or had sufficient resources to adjust to the new situation (Zapf et al., 1996). We had expected to find more variation regarding trajectories of mental well-being, given that high day-to-day fluctuations of mental well-being have been reported (Totterdell et al., 2006). It is interesting to see that especially the persons with limited well-being suffered further losses, while the group with high well-being gained slightly in terms of well-being. This may be explained by the conservation of resources (COR) theory (see Hobfoll, 1989), which suggests, that individuals with more availability of resources are more likely to receive additional resources, while those with fewer resources are more likely to be threatened by resource-loss, similar to the Matthew effect: “For to every one who has will more be given, and he will have abundance; but from him who has not, even what he has will be taken away.” (Matthew 25:29).

Our data did not support our hypothesis that the onset of telework is a job demand, which we had assumed to due its sudden implementation and reports of increased job demands due to teleworking during the COVID-19 pandemic (Wang et al., 2021). On the contrary, it turned out that in the initial phase of the COVID-19 pandemic, telework was a resource that could buffer the consequences of job demands in terms of work intensification and, in part, compensate for a lack of personal resources. Possibly, previous literature on telework was positively biased in the sense that mainly workers were included, who desired to telework and had favorable conditions (Delanoeije and Verbruggen, 2020); and it is not entirely clear whether telework could not also serve as a demand instead of a resource. Our article contributes to this question, because due to the forced transition to telework, we were able to survey people whose job and personal conditions were not necessarily ideal for it. This perception of telework as a job resource in the early stages of the COVID-19 pandemic is also reflected by the trend of many employees calling for a continuation of telework arrangements beyond the pandemic, which holds especially true for younger workers, as well as those with higher information and communication technology (ICT) use and those with positive experiences with telework (Georgescu et al., 2021). Additionally a recent survey found that 54% of 16,000 surveyed workers, across industries and countries, would consider quitting their job if certain amenities they experienced during the pandemic, like working from home and work time flexibility, were to be retracted (Ernst & Young, 2021; Melin and Egkolfopoulou, 2021). This classification of telework as a job resource is also supported by the significant interaction of telework with the personal resource occupational self-efficacy on well-being: the compensation of no telework availability through occupational self-efficacy is in line with and supports the assumptions of the job demands-resources (JD-R) model, according to which personal resources can compensate for missing job resources (Bakker and Demerouti, 2017).

The absence of significant main effects for the remaining job resources and their interactions with telework may possibly be explained by the specific circumstances brought upon by the COVID-19 pandemic: the effects of perceived autonomy, social support, or flexitime at the first survey time point may pale behind the importance of self-regulatory strategies in terms of occupational self-efficacy, as this may have been essential for transitioning and adapting to the pandemic situation. Similar results for teleworkers were found in France, where the job resources autonomy and organizational support didn't have as much of an impact on adjustment to telework as expected (Carillo et al., 2021). This would suggest that employees' belief in their ability to perform well in their job, regardless of the work environment, is more influential on their well-being than the resources provided by their job. The relevance of personal resources for well-being during the pandemic is in accordance with the literature (e.g., Cotel et al., 2021; Joie-La Marle et al., 2021). The lack of a significant effect of job resources on mental well-being during pandemic in this study may also inform the JD-R model, by highlighting the importance of considering personal resources such as self-efficacy in addition to job resources (Xanthopoulou et al., 2007). It is important to caution against overinterpreting the findings of our study as it was conducted during a specific situation the pandemic.

Nevertheless we found a significant association of some of our control variables with well-being group membership, in the direction and context expected from the literature (i.e., children increasing the likelihood of being in the low well-being group; cohabitants decreasing it; Zamarro and Prados, 2021; Etheridge and Spantig, 2022), which speaks for the validity of our results, which held under the addition of these variables.

Although we had to discard telework as a job demand and rather found it being a job resource, we are able to add to the literature about personal resources in the JD-R model whose moderating relationship between job resources and the health improvement path had been recognized by Xanthopoulou et al. (2007), but not tested. Similar to our results, personal resources (i.e., optimism) were found to buffer the negative effect of low job resources on work engagement (Salminen et al., 2014).

What did we learn? Contrary to our assumptions, our study shows that telework is a job resource even in in its unfamiliar initial phase. Moreover we add to the JD-R literature by demonstrating positive effects of job and personal resources on trajectories of well-being over time.

Of course, our study has limitations that we would like to point out. We must caution about generalizing our results to non-pandemic periods and, although we surveyed longitudinally, we cannot specify cause-effect relationships because we examined only the effect of variables at T1 on well-being trajectory group membership. A clear strength of the study is that it captures well the onset of pandemic-induced changes, given that we started our survey immediately when lockdown mandates came to effect in Germany. It is conceivable that parts of our findings are due to influences of the particular situation rather than finding their genesis in the general onset of telework. Since the data analyzed here was limited to Germany, our research's applicability to other countries can't be assumed. This is equally true for the german work force in general. We can't rule out bias due to our retrospective approach of querying well-being (i.e., “How did you feel in the past week?”), as well as common method bias and using self-ratings instead of objective measures of well-being (Podsakoff et al., 2003; Schmier and Halpern, 2004). Additionally it is important to note that the majority of the convenience sample studied here was well educated, thus we can't assume absence of a socioeconomic bias either.

Future research should try to replicate our study design under non-pandemic conditions. Furthermore, it would be conceivable to use an experimental design in which long-term well-being is examined by means of occupational self-efficacy training, with and without the offer of telework, in order to be able to better analyze the individual effect facets.

Work intensification has been shown to be a stressor that is predictive of psychological well-being. Our results suggest that telework may contribute to improved coping with intensified work conditions. Telework can buffer the effects of work intensification and has a positive impact on well-being trajectories. In practice, it could be useful to grant stressed employees additional (or in principle) teleworking time to cope with special workloads and demands.

In addition, the study also shows that training of occupational self-efficacy can be useful to deal with special situations (e.g., when telework is not possible), since in our study the personal resource occupational self-efficacy in contrast to job resources had a sustainable effect on well-being trajectory group membership. A promising approach here could be, for example, interventions aimed at increasing psychological capital, a construct which, in addition to self-efficacy, also includes hope, optimism, and resilience (Luthans et al., 2010).

In conclusion, our study has shown that telework can have a positive impact on well-being by buffering the negative effects of work intensification. The results indicate that access to telework, where possible, can lead to improved outcomes in terms of well-being and can help employees cope with special situations such as the COVID-19 pandemic.

Furthermore, our study revealed that occupational self-efficacy is a personal resource that has a direct measurable effect beyond traditional job resources in the early period of the pandemic. The results suggest that training in occupational self-efficacy can be beneficial in helping employees deal with the unique stressors and challenges brought about by the pandemic. This highlights the importance of considering both job and personal resources in understanding employee well-being during the pandemic.

It is important to keep in mind, however, that the results of this study were obtained during a specific situation, the COVID-19 pandemic, and should not be generalized to other teleworking situations.