Homebound happiness? Teleworkability of jobs and emotional well-being during labor and non-labor activities

Over the past decades, telework has seen substantial growth, evolving from a niche practice to a mainstream employment model. Driven by technological advances as well as changing societal expectations and catalyzed by the COVID-19 pandemic, remote work has become an integral element of modern employment relationships (Adams-Prassl et al., 2022).

Several studies have explored the association between telework and the subjective well-being of workers. These studies typically focus on cognitive aspects of well-being, such as life satisfaction or job satisfaction. The findings are mixed, ranging from positive pre-COVID findings on job satisfaction (Vega et al., 2015; Bloom et al., 2015; Smirnykh, 2023) to negative effects on life satisfaction and mental health during COVID (Gueguen and Senik, 2023; Senik et al., 2024; Cheng et al., 2021). Using experimental variation in return-to-office policies during the COVID pandemic, Choudhury et al. (2026) find that the association between telework and job satisfaction is nonlinear, with workers in hybrid conditions reporting the highest levels of satisfaction.

Telework has been linked not only to cognitive well-being but also to workers’ emotional experiences, that is their affective well-being (AWB). Using the American Time-Use Survey (ATUS), Giménez-Nadal et al. (2020) find that teleworkers spend less time on work-related activities (e.g. commuting) and work more at irregular times of the day (evenings, weekends). They also find that teleworking men on average feel better than commuters, while they find no evidence for a clear relationship between telework and emotional well-being for women. Song and Gao (2020) point out that one should also distinguish between two different kinds of telework. When workers spend full days working from home, they free up the time that would otherwise be devoted to commuting. However, they might also take office work home and thus devote additional time to working at home on evenings or weekends. Telework during evenings is found to be harmful to emotional well-being, while regular telework is not. This negative effect seems to be even stronger for parents. Using the UK Time-Use Study, Lu and Zhuang (2023) find that men who often work from home enjoy their working time more and are more satisfied with their jobs than men who work primarily outside their homes. They do not detect such differences for women. Using the same data, Giménez-Nadal and Velilla (2024) find that teleworking is associated with less working time overall. However, they find that remote work is associated with less enjoyment of working time for men, while they observe no difference among women. Both men and women enjoy their leisure time less when working from home, which the authors attribute to the blurring of boundaries between work and non-work obligations.

Many employees value flexible work options highly, often prioritizing them over salary increases (Mas and Pallais, 2017; Vij et al., 2023). When asked directly about their preferences for working from home, people generally seem to have a positive view of teleworking, in particular because they expect it to reduce work–family conflicts (Moens et al., 2022). Surveys indicate that, when opting for remote work, employees are even willing to trade off job rewards for greater family satisfaction (Moller et al., 2024). This suggests that the full impact of telework on employee well-being cannot be identified by simply comparing episodes of remote work with days spent at the regular workplace for the same individual. Rather, the relevant margin is whether employees have the option to adjust their place of work in response to competing demands across life domains. Teleworkability can be understood as an important component of job quality, reflecting the degree of autonomy and control that employees have over when and where they perform their tasks. By expanding the feasible set of work–leisure arrangements, it enables workers to reallocate time across work, family, leisure and other activities in a utility-enhancing way. In this sense, what matters for well-being is not the realized work location at a particular point in time, but the structural flexibility – and the autonomy it represents – embedded in the job itself. Regardless of whether this flexibility is exercised through working from home on a given day, shifting hours or reorganizing tasks, the availability of a broader choice set can be expected to have a non-negative effect on overall well-being.

While the option to work from home allows workers to reschedule their work and non-work activities so as to reduce conflicts between them, this does not necessarily imply that they enjoy every activity more than they would without teleworkability. An optimal use of this flexibility may involve shifting a particular activity to a time slot in which it is less pleasant if doing so frees up time that can be spent more enjoyably on another activity. For example, parents might move some work hours from the late afternoon to the late evening, even if working at that time is less enjoyable, in order to spend more time with their children earlier in the day. Their overall well-being increases (otherwise they would not choose to reschedule), even though the enjoyment of their working time falls.

Public debate often overlooks that telework increases flexibility for both employees and employers, potentially fostering on-call demands and permanent availability. This increase in telepressure has been found to negatively associate with employee well-being by Tatar and Erdil (2024). Even if unused, this latent flexibility can affect well-being by influencing autonomy, employer expectations and workers’ ability to adapt to unexpected personal circumstances – including during on-site work or leisure time. Lastly, this expansion of options illustrates the paradox of choice (Schwartz, 2004): while greater autonomy and flexibility are generally positive aspects of job quality, an abundance of options can create decision overload and uncertainty, particularly for workers balancing multiple roles such as paid work and family responsibilities.

Taken together, these considerations imply that focusing on observed telework alone would miss important channels through which flexibility affects well-being. What matters is the structural potential to work remotely – teleworkability – which constitutes an important dimension of job quality through its provision of autonomy and control. By shaping substitution patterns, spillovers into non-telework activities and the well-being effects of latent flexibility, teleworkability affects employees’ overall work–life balance, regardless of whether remote work is actually realized at a given point in time.

We examine the relationship between the possibility to work remotely and workers’ AWB. Using data from the well-being (WB) module of the ATUS, we analyze the emotional experiences of employees during paid work as well as outside of work, for example when doing chores, meeting friends or pursuing their hobbies. We focus on the teleworkability of an occupation, capturing how feasible it is to perform job tasks remotely such that workers potentially have the freedom to choose the location of their work. Teleworkability is analytically relevant in its own right, as it reflects a structural property of a job, independent of employees’ actual decisions to work remotely. While actual remote-work behavior is shaped by short-term organizational policies and individual circumstances, teleworkability captures the inherent potential for locational flexibility in the job’s task profile. This potential may shape worker well-being not only when remote work is actively used but also through the option value it creates – for example by influencing how work schedules are negotiated and how employers organize supervision.

While the ATUS provides detailed diaries of people’s activities and emotional experiences, it does not contain information on the general availability of remote work. Hence, we match the individual ATUS diaries with data from the Occupational Information Network (O*NET), which provides a large set of occupational characteristics on nearly the entire universe of occupations in the US economy. This match allows us to identify to what extent the AWB of workers differs between occupations in which employees have more or less flexibility regarding their work location. We conduct gender-specific analyses because the effects of telework and structural flexibility may interact with differences in unpaid work responsibilities and prevailing social norms. For example, women often bear a disproportionate share of household and caregiving duties, which may amplify the benefits or the stresses associated with flexible work arrangements. Understanding these interactions is crucial for interpreting gendered patterns in well-being across occupations (e.g. Arntz et al., 2022; Senik et al., 2024; Giménez-Nadal et al., 2020; Giménez-Nadal and Velilla, 2024).

Our results indicate that there is a significant negative association between teleworkability and AWB during labor activities for women. No significant associations are observed for men during labor activities, and there is no evidence of systematic extension of these associations into other activities – such as unpaid work or leisure – for either gender. Furthermore, there is no significant relationship between a job’s teleworkability and the day-average affective or cognitive well-being of men or women, providing evidence that teleworkable occupations may not be as uniformly advantageous for workers’ AWB as theoretically predicted and publicly discussed.

Our analysis extends existing research on the relationship between job characteristics, in particular flexible work arrangements, and subjective well-being. We make two specific contributions to this literature. First, instead of examining only globally evaluative well-being measures (e.g. life or job satisfaction), we analyze workers’ AWB at specific points in time during a day (cf. Kahneman and Krueger, 2006). This allows us to distinguish between the direct association of teleworkability with the emotional experience of work itself and potential extensions into domestic work and leisure activities. Second, and contrary to other studies that also use the ATUS to analyze telework (Giménez-Nadal et al., 2020; Song and Gao, 2020), we use data on the general availability of telework instead of its actual use on a specific day. Studying teleworkability rather than actual remote work distinguishes structural job characteristics from employees’ behavioral choices. This extends our theoretical understanding of the broader welfare implications of occupational flexibility: two jobs with identical current remote work usage can differ starkly in teleworkability (or vice versa), implying different capacities to adapt to worker needs. Additionally, teleworkability allows employees to change the way they structure their workdays and workweeks. The consequences might not only be felt on days when workers actually work remotely but could also affect how they experience their work and leisure on other days. Hence, examining the general availability of remote work extends previous studies and, together with their findings, allows for a more comprehensive assessment of its importance for workers’ well-being.

We use data from the ATUS, which has been conducted by the US Bureau of Labor Statistics annually since 2003. The ATUS is the first federally administered, continuous survey on time use in the United States. ATUS participants are randomly selected from households participating in the Current Population Survey. Respondents participate once and report how much time they spend on various everyday activities. They are asked to structure the day preceding the interview into separate episodes. For each episode, they report the start and end time, what they did (i.e. the type of activity), where they were and who they were with. In 2010, 2012, 2013 and 2021 ^[1], the ATUS additionally included a Well-Being (WB) module based on the “Day Reconstruction Method” (Kahneman et al., 2004). In this module, participants are asked to report on the emotions they felt during three randomly chosen episodes to measure their emotional/affective well-being during these activities (see below for more details). The activities had to be at least 5 min long to be included, and activities in the categories sleeping, grooming and personal activities have been excluded.

We restrict the sample to all employed respondents aged 18–64 who participated in the ATUS WB module. We additionally restrict the sample to respondents with valid information on all relevant variables. The sample consists of 45,981 observed episodes (with well-being information) for a total of 15,698 individuals. Summary statistics for the estimation sample on the individual level can be found in column (1) of Table A.1 in the Supplementary Material. Forty-six percent of the individuals in our sample are women. The average age is 40.39 years, 84% are born in the United States, 59% have a partner in the household and 15% live with a child under the age of 6. Columns (2) and (3) contain summary statistics separately for men and women.

Figure 1 and Table A.2 in the Supplementary Material give an overview of the frequencies and average cumulative duration of all observed episodes of the individuals in our sample, grouped by activities, as well as the frequencies of only those episodes observed in the WB module ^[2]. We group the activity codes provided in the ATUS into four categories: labor/paid work, unpaid work, leisure and other activities ^[3]. We define unpaid work as any care or household-related activity, including child as well as eldercare, housework (such as cleaning and cooking), as well as all sorts of household-related errands, such as grocery shopping and using household or professional services (e.g. banking).

The figure shows that, due to the exclusion of sleeping and grooming activities from the WB module, most other activities are slightly over-represented in the estimation sample, with the exception of care activities. While, for example, 8.94% of episodes in the full diaries are work or work-related activities, this share would be 12.86% in the WB module, and given that work-related episodes tend to be long on average, these episodes make up for 21.95% of the total day. The most commonly observed activity among all episodes is traveling (work-related and not work-related) (21.91%), followed by sleeping and personal care (19.58%). Nevertheless, traveling episodes seem to be relatively short on average while sleep episodes are very long which is why the cumulative time of all sleep periods makes up for 39.22% of the total day, while traveling only accounts for 1.44% (work-related) and 4.01% (other) of the total time. Table A.3 in the Supplementary Material contains more detailed summary statistics on the activity level for the estimation sample.

The outcome variable in our empirical analyses is AWB. ATUS respondents are asked to rate how much they experienced specific emotions on a scale from 0 to 6. The questionnaire captures how happy, meaningful, sad, stressed, in pain and tired they felt during each queried episode. Studies using the ATUS to investigate well-being differ widely in the emotions they analyze and in whether or how they combine them into composite measures of AWB (Hoang and Knabe, 2021; Giménez-Nadal et al., 2020). We decide to include only the most commonly used emotions and, thus, to exclude meaning and pain in our main specification ^[4]. We follow Kahneman et al. (2004) and Kahneman and Krueger (2006) and construct three aggregate measures of AWB. Positive affect captures how happy respondents rated an episode. Analogously, negative affect is the average rating of the emotions sad, stressed and tired. The net affect is then calculated by subtracting the negative from the positive affect score. In addition to the net affect experienced during single episodes, we also determine respondents’ overall level of AWB. This is computed as the duration-weighted mean of the net affects of all the episodes that are observed in an individual diary (day-average AWB).

Lastly, we also analyze the cognitive well-being of ATUS respondents. Psychologists have long argued that cognitive and AWB constitute theoretically distinct dimensions of subjective well-being (Diener, 1984; Diener et al., 1999). There is also empirical evidence that cognitive and AWB are distinct phenomena and have different determinants (Kahneman and Krueger, 2006; Kahneman and Deaton, 2010). While AWB exhibits more short-run variability than cognitive well-being, long-run averages of AWB and cognitive well-being are strongly correlated (Berlin and Fors Connolly, 2019). Cognitive well-being is captured in the ATUS WB module in the years 2012, 2013 and 2021 with Cantril’s Self-Anchoring Scale (scale 0–10). Distributions of the episode-level and day-average net affect, cognitive well-being as well as episode-level positive and negative affect scores (again differentiated by the three underlying emotions) are shown in Figure A.1 in the Supplementary Material.

Figure 2 shows the weighted mean net, positive and negative affect for different types of activities ^[5]. In line with the earlier literature, we can see that labor activities (both at the workplace as well as remote) and work-related traveling are among the least enjoyable activities (Kahneman et al., 2004; Knabe et al., 2010). Contrary to Kahneman et al. (2004), respondents in our sample seem to experience relatively high levels of emotional well-being during care activities. This is largely driven by a high level of positive affect.

The focus of this study is on the relationship between well-being and the teleworkability of jobs, that is the theoretical possibility of locational flexibility. Teleworkability should not be interpreted as a general measure of job autonomy or overall flexibility but more narrowly as the extent to which the core tasks of an occupation can be performed remotely and therefore without continuous physical presence at the employer’s premises. In principle, ATUS provides information on realized remote work. Since we are able to observe the location at which an activity has been performed, we can distinguish between work episodes that took place at the workplace or at home, which we also do to look into work episodes at the workplace and remote work separately. Nevertheless, we do not use this information to construct the explanatory variable for our research question for various reasons. First and foremost, our theoretical reasoning implies that we are not interested in actually realized remote work activities but in the potential benefits and costs of theoretically being able to choose the location of one’s work freely. As we discussed above, locational flexibility does not necessarily have to affect well-being only while actually working from home, but allows more flexible scheduling that could be beneficial or costly for well-being also while working at other locations or during non-work activities.

Secondly, using teleworkability instead of remote work also has methodological advantages. When comparing work activities at home and the workplace, it is likely that the activities, even when observing the same individual, also differ in other dimensions than just the location. Individuals who, for example, experience more stress at work might be more likely to take work home. This would result in the estimates being biased by confounding factors. Additionally, the observation of remote work on a single, randomly chosen day is a highly inaccurate indicator of the general utilization of remote work opportunities because it relies on the random sampling of individuals during a (tele)work activity. That an individual is not observed working from home in ATUS on the randomly chosen day does not mean that she does not work from home often on other days. Hence, there would be statistical noise in this remote work variable, which would cause attenuation bias. Additionally, during most years in our estimation period (2010–2013), remote work was still a relatively rare event, while in 2021, during the COVID pandemic, remote work in many cases did not follow any regularities, which amplifies the concerns regarding measurement error given above. Consistent with these concerns, specifications using an indicator for realized remote work yield imprecisely estimated coefficients and do not provide additional insights for our research question ^[6].

To obtain a reliable indicator of teleworkability, we utilize representative data on occupational characteristics from the Occupational Information Network (O*NET). O*NET is a program of the US Department of Labor’s Employment and Training Administration. The O*NET database includes a large set of occupational characteristics on nearly the entire universe of occupations in the US economy. O*NET has been used in various economic studies to estimate the degree of teleworkability at the occupational level (Dingel and Neiman, 2020; Mongey et al., 2021). We use the information on occupational requirements included in O*NET’s work activities (WA) and work context (WC) modules, which are collected with standardized questionnaires given to random samples of job incumbents ^[7]. In the WA questionnaire, job incumbents are asked to rate how important certain types of work activities are for the job they currently hold. In the WC questionnaire, respondents are asked questions about their working conditions, such as their work setting and its possible hazards, the pace of work, and their dealings with other people. In both questionnaires, all questions about the frequency or importance of certain conditions or activities are answered on a scale from 1 (never/not important) to 5 (every day/extremely important).

Not all surveys in the O*NET database are conducted at the same point in time, so the information for different occupations is gathered at different points in time. New versions of O*NET contain updated information for a subset of occupations. To temporally match O*NET data to the well-being information we have in the ATUS, we use O*NET versions that contain updates for specific occupations that are as close in time as possible to the ATUS waves 2010, 2012, 2013 and 2021, respectively. For more details on the data and merging process, see Supplementary Material Section B.

We construct a measure of the degree of teleworkability of a specific occupation following Dingel and Neiman (2020) ^[8]. Their definition of teleworkability is based on a number of queried activities, which are either less likely to be conducted from home (e.g. outdoor activities, wearing safety equipment, exposure to diseases, dealing with violent people) or which make working from home easier (e.g. frequently using email for communication). In line with their work, we use the same set of items from the O*NET work-activities and work-context modules ^[9]. An overview of the included items can be found in Table A.4 in the Supplementary Material. The teleworkability indicator is then calculated as the arithmetic mean of all items in an occupation in a given year.

Figure 3 shows the distribution of the teleworkability variable and also includes examples for occupations with different teleworkability levels for illustration.

We test the validity of this teleworkability indicator by analyzing its association with the realized shares of remote work of ATUS respondents in an occupation. Table 1 illustrates the share of all work episodes done remotely separately for occupation with high and low teleworkability (as compared to the median). We can see that the share of remote work is, on average, over all years more than twice as high in teleworkable jobs. The difference increases dramatically in 2021 due to the COVID pandemic, when the share of remote work was over 50% in teleworkable jobs while only 16.56% in jobs with low teleworkability. The correlation coefficient (ρ) between remote work and teleworkability of the occupation is ρ = 0.21 (ρ₂₀₂₁ = 0.42 in 2021).

Another important aspect of the descriptive statistics on teleworkability for the validity of our later estimation strategy is the within-occupation variation in teleworkability over time. As we will discuss later, using within-occupation variation would, in principle, be beneficial to avoid omitted variable bias caused by time-invariant occupational characteristics. Nevertheless, from a theoretical point of view, the interpretation of within-occupation variation in teleworkability is questionable. The teleworkability of a job is relatively stable. Within-occupation changes in it over time are quantitatively small and might reflect general changes in working environments rather than actual changes in teleworkability. As an example, a non-teleworkable job, such as being a firefighter, will not become more teleworkable just because using e-mails is more common in 2021 than it was in 2010. In order to also show this descriptively, we calculate the share of between- and within-occupation variation of teleworkability as well as between-year correlation coefficients. We can see that only 13% (0.07) of the overall standard deviation of teleworkability (0.55) can be attributed to within-occupation variation, and the between-year correlation of teleworkability of an occupation varies between 0.94 and 0.997 depending on the year (see Figure A.3 in the Supplementary Material).

Thus, anything we find in the later estimation is mainly driven by between-occupation variation in teleworkability. We still use the full set of information available to us instead of an averaged version of teleworkability but check the robustness of our results using average teleworkability (see Table D.1 in the Supplementary Material). Results are robust and, if anything, increase in magnitude, indicating that our estimates are more conservative and might even be biased downward due to attenuation bias.

We are interested in the relationship between teleworkability and subjective well-being, conditional on other relevant individual and occupational characteristics. To determine this relationship, we regress emotional well-being WB_ij of individual i in episode j on the standardized linear measure of teleworkability ^[10], TW_i, as well as a list of episode-level (A_ij), individual-level (X_i) and occupation-level (O_i) controls. The estimation equation has the following form:

The episode-level control variables A_ij are the duration of the episode, time of the day, an indicator for weekday/weekend and the number of people present during the activity ^[11]. Individual-level control variables X_i are age, education level, immigration status, region of living (urban or rural), marital status, number of children, presence of a small child in the household, weekly earnings, number of weekly working hours, as well as year and month of the survey. Occupation-level controls O_i are occupational prestige, average hourly wage and average weekly working hours, total number of employees, wage dispersion, average required education, average required tenure, as well as the level of cognitive, physical and emotional job demands. Including these controls is particularly important because teleworkability may correlate with broader occupational characteristics; conditioning on them helps distinguish the association of teleworkability from other observable dimensions of job quality and task composition. Details on data sources and variable construction for all control variables are presented in Table A.1 in the Supplementary Material. Observations are weighted with the respondent weights for the WB Module in all estimation models, and standard errors are clustered on the occupation times O*Net data update level.

In a second step, we examine the association between teleworkability and the duration-weighted average AWB over the survey day as well as cognitive well-being (WB_i). This reduces the sample to only one observation per individual (i). Analogously to (1), WB_i is regressed on the teleworkability indicators as well as on the set of control variables:

One should be cautious about interpreting the estimated partial correlations causally. It is conceivable that the estimated relationships are affected by (1) unobserved confounders, for example corporate culture, and (2) reverse causality, for example the self-selection of individuals with a very high or low level of emotional well-being at work into more or less teleworkable jobs.

We hope that (1) is largely addressed by the substantial set of control variables we add to our estimation model. Nevertheless, teleworkability may still be correlated with unobserved occupational characteristics or aspects of workplace organization that are not captured in the available data. To assess the potential importance of such omitted variables, we apply the approach proposed by Oster (2019) that tests coefficient stability with respect to the relative degree of selection on observable and unobservable confounders. In our case, the estimated effect size is not attenuated by the inclusion of controls, but increases substantially (in absolute terms) when moving from the baseline to the fully controlled specification, suggesting that selection on observables biases the uncontrolled estimates towards zero. Consistent with this, the Oster (2019) test yields very high critical values of the relative degree of selection on unobservables (δ > 3), indicating that an implausibly strong selection on unobservables would be needed to explain away the observed effect. Since the coefficient increases with the inclusion of controls, the main concern of Oster’s method – that unobservables explain the effect – is less relevant in our case.

Reverse causality (2) is always a potential problem in cross-sectional well-being studies. For this reason, we focus on the teleworkability of jobs rather than on episodes of actual remote work. While the choice of one’s work location might be highly endogenous and influenced by one’s temporal emotional state, this is, arguably, less the case with respect to general job attributes such as teleworkability. Nevertheless, endogeneity remains a potential problem.

Alternative models for the identification of a causal effect of teleworkability on AWB would either involve some experimental variation in teleworkability [e.g. Choudhury et al., 2026 or Bloom et al. (2015) for telework] or examine the within-person changes in well-being following a variation in the teleworkability of occupations over time. Both are, nevertheless, not feasible in our setting, given that teleworkability varies much less over time. As discussed in Section 2.2, what we observe over time is not an increase in the teleworkability of jobs but an increase in the correlation between teleworkability and telework probability as telework becomes a more and more common practice (within the group of teleworkable jobs). Hence, there is not enough within-occupation variation in teleworkability to allow for estimating regressions with fixed effects.

In Table 2, we present the results of the regressions of AWB on teleworkability (TW) (equation (1)). We estimate the regressions separately for women (columns 1–3) and men (columns 4–6). First, we restrict our attention to the well-being experienced during labor activities. The upper panel summarizes the findings on net affect, while the middle and lower panels show the results for positive and negative affect, respectively. Different sets of control variables are gradually included in the model.

Columns (1) and (4) contain the unconditional associations between teleworkability and net affect. For women, this association is significantly negative, while the association is not statistically significant for men (although of a similar magnitude). A one-standard-deviation higher teleworkability is associated with a 0.142-unit lower net affect during labor activities for women (about 7% of a standard deviation). We find a significant negative association with positive affect for women, but again not for men (middle panel). As can be seen in the lower panel, there is no significant association between teleworkability and negative affect for women, while men seem to experience more negative emotions when they are employed in more teleworkable jobs. When we control for activity and individual characteristics, the association between teleworkability and net affect is (still) significantly negative for women, and the magnitudes of the associations increase slightly (see column (2)). Column (1) of Tables C.1 and C.2 in the Supplementary Material present the results of regressing each emotion separately. The results show that the lower level of net affect is reflected by lower levels of happiness for women. Additionally, teleworkable jobs are associated with higher stress levels, but also lower levels of tiredness. These unconditional correlations, especially the differences in tiredness, are likely caused by certain job demands, in particular physical strain, which are positively correlated with both teleworkability of jobs and the levels of these emotions. In order to control for such confounders, we additionally control for a set of occupational characteristics (columns (3) and (6) of Table 2) ^[12]. When comparing them to the model in which only activity and individual characteristics are controlled for, we can see that especially the coefficients for women increase considerably, which indicates that our fully controlled model captures relevant occupational heterogeneity between teleworkable and non-teleworkable jobs. For women, we find a more negative relationship between teleworkability and well-being. A one-standard-deviation increase in teleworkability is associated with a decrease in net affect during labor activities of 33% of a standard deviation (0.681 units). This indicates that teleworkable jobs have other occupational characteristics that are themselves positively associated with well-being and confound the estimated association ^[13]. The observed changes are driven by a more negative relationship of teleworkability with happiness as well as a more positive association with stress and tiredness. No associations with sadness are observed in the fully controlled setting. In order to assess whether this negative association is driven by women in teleworkable jobs not liking their (more common) remote work episodes or by more structural disadvantages of workplace flexibility (as we argue theoretically), we analyze the association of teleworkability with AWB separately for work episodes performed at the workplace and those done remotely. Columns (1)–(4) of Tables C.5 and C.6 in the Supplementary Material contain the corresponding results and show that the negative association can be found both at the workplace and remotely, indicating a more general structural difference between more or less flexible jobs besides the higher frequency of remote work episodes.

For men, the estimated association between teleworkability and net affect increases in absolute magnitude but remains statistically insignificant when adding occupational controls. When looking into workplace episodes and remote work episodes separately, we can see that general teleworkability is negatively associated with AWB during remote activities but not during episodes conducted from the workplace. Overall, these findings provide tentative evidence that an occupation’s teleworkability might not be as universally beneficial for the AWB of workers as theoretically predicted and publicly discussed.

A higher degree of flexibility in the determination of the place of work could allow for reducing time conflicts with non-work activities, which could then increase the AWB experienced outside of work. In the following, we examine this theoretical extension into other activities by analyzing the association of teleworkability with AWB during non-work activities. Tables 3 (women) and 4 (men) contain the results of estimating AWB in unpaid work, leisure and other activities, applying the regression model (1) ^[14].

In line with the findings for labor activities, we can see a negative association of teleworkability with AWB during unpaid work for women (driven by weaker positive emotions) in the uncontrolled setting (column (1) of Table 3). We also observe a significant negative association between teleworkability and positive affect during unpaid work for men (Table 4, column (1)). However, these associations become statistically insignificant when we control for individual and occupational characteristics.

Additionally, we also do not find significant evidence that teleworkability is associated with the net affect experienced during leisure activities by men and women (columns (3) and (4) of Tables 3 and 4). When looking at positive and negative emotions separately, we find a significant negative association between teleworkability and positive emotions for women in the regressions without controlling for individual and occupational characteristics. With these controls, the association with positive affect disappears, but we find a significant negative association between teleworkability and negative emotions for women.

During other activities (i.e. activities that are neither unpaid work nor core leisure activities), we see a negative association between teleworkability and negative affect for women in the regression without controls (but no significant effects with controls). For men, significant associations are found in the regressions with controls between teleworkability and net affect (negative) and negative affect (positive). We examine two of these activities more closely (eating and drinking, and work-related traveling) ^[15]. In the fully controlled model, we cannot find any significant associations between the teleworkability of jobs and positive or negative emotions experienced during eating and drinking or work-related traveling.

Based on the findings discussed in the previous section, we can now analyze how the observed differences in net affect during certain activities translate into differences in overall emotional well-being. Accordingly, we analyze the association between teleworkability and day-average AWB.

Table 5 contains estimates of the association between teleworkability and the duration-weighted average AWB that individuals experience over the survey day as well as their cognitive well-being. We estimate these associations, applying regression equation (2) and using individual-level data for the 7,871 women and 7,827 men in our sample. As the measure for cognitive well-being was not surveyed in 2010, the sample further reduces to 5,407 women and 5,406 men in the estimations using this outcome variable. The results are shown in Table 5.

We do not find significant associations between teleworkability and cognitive well-being. We also see that neither the raw differences in well-being during paid work activities between men and women with or without teleworkable jobs nor the negative association for women in the fully controlled model translate into lower levels of day-average well-being. Since we found a negative association between teleworkability and well-being during labor activities and no clear evidence for extensions into non-work activities, we would expect to see (weakly) negative associations between teleworkability and day-average well-being. While our estimates all have negative signs, none of them is statistically significant. It is conceivable that differences in day-average AWB are influenced not only by the differences in the level of emotional well-being experienced during specific activities but also by differences in the composition of the various activities, that is differences in time use in general (as day-average emotional well-being is duration-weighted). Individuals in teleworkable jobs might, on average, spend more or less time in enjoyable activities. An example is the difference in the composition of work-related traveling, as one of the least enjoyable activities. While individuals with non-teleworkable jobs might need to commute on a daily basis, individuals with teleworkable jobs might commute less but longer distances because they are able to take up jobs in more distant locations.

To disentangle the role of differences in experienced emotions and differences in the composition of activities, we take a closer look at the time spent in different activities when women or men hold (non)teleworkable jobs ^[16]. Our findings suggest that all differences in time use between individuals with and without teleworkable jobs can be explained by other individual or job characteristics, such as the lower level of sleep for men, higher levels of sports and eating/drinking for both men and women as well as lower levels of housework for women. This is even true for the average duration of labor activities conducted at the workplace and remotely. While controlling for education, a job’s physical demands and other characteristics removes differences in average time spent in remote work, this does not necessarily imply that teleworkability is irrelevant for workplace flexibility. The absence of an independent association with remote work duration simply indicates that much of the variation in actual remote work is driven by other, correlated factors.

However, the persistent differences in AWB – even after adjusting for the same controls – suggest that teleworkability captures latent aspects of flexibility, such as the option to work remotely when needed, greater autonomy over work arrangements or reduced constraints in adapting to personal circumstances. Such flexibility may be positively related to well-being regardless of whether it systematically translates into more remote work.

To test whether the association between telework and emotional well-being is stable across survey years – particularly given the onset of COVID – we estimate the relationship separately for 2010–2013 (pre-COVID) and 2021 (during COVID). Results are shown in Table E.1 in the Supplementary Material.

We see that the negative relationship between teleworkability and AWB for women during labor activities is particularly evident prior to the COVID pandemic. The estimates for 2021 are smaller and less precisely estimated due to the smaller sample size. In contrast, for men we see a large and statistically significant negative association during COVID, but no significant association before COVID. This could indicate that while for men the actual realization of remote work is important for their well-being (as has already been shown in the heterogeneity analysis by work location above), for women other aspects of teleworkability are detrimental to their AWB (which are nevertheless overshadowed by the advantages of actual remote work during the COVID pandemic).

In line with this, we do find significant evidence for a positive relationship between teleworkability and the net affect of women during unpaid work in 2021. This could for example suggest that the teleworkability of a job has gained importance for the compatibility of family and work, caused by an overall increase in the importance of telework and hybrid work in teleworkable jobs during the pandemic.

In addition to locational flexibility, another dimension of flexible work arrangements concerns the timing of work. Many employment relationships have defined schedules that follow from technical constraints or the need to coordinate work in teams. A reduction of schedule constraints could make it easier for employees to coordinate their professional and private obligations, which would reduce time conflicts and improve their work–life balance (Herrera-Ballesteros et al., 2025; Goldin, 2021). To complement our previous findings, we use O*NET information to generate a measure of working-time flexibility and add it as an additional explanatory variable into our model as well as an interaction term between time flexibility and teleworkability. Details on the variable construction and estimation model can be found in Supplementary Material Section F.

Table F.1 in the Supplementary Material contains the results of this analysis. We do not find significant evidence that time flexibility is directly associated with net affect or moderates the relationship between teleworkability and well-being during labor activities. For women, we find significant evidence that time flexibility is positively associated with negative emotions during paid work. For men, we observe a significant interaction term for negative emotions, indicating that men in jobs that are both teleworkable and time-flexible experience significantly more negative emotions. During non-labor activities, we find significant evidence for a negative interaction between teleworkability and time flexibility for women during leisure, which suggests that time flexibility is associated with less enjoyment of leisure time, especially when jobs are teleworkable. Our findings also suggest that men with more time-flexible jobs, on average, also have a higher enjoyment of leisure activities. Since all other estimates are not statistically significant, we conclude that these findings do not provide enough evidence to claim that time flexibility plays a crucial role in moderating the association between teleworkability and AWB.

This study examines the relationship between teleworkability – the structural potential to perform a job remotely – and workers’ AWB. By focusing on this occupational characteristic rather than on actual episodes of remote work, we capture the latent flexibility embedded in a job’s task profile. Well-being may vary with this potential even on days when employees work on-site or when they are off work.

Using well-being data from the ATUS and information on the feasibility of remote work at the occupation level from O*NET, we show that this structural flexibility is not uniformly beneficial: for women, higher teleworkability is associated with lower AWB during labor activities, while for men we find no consistent pattern ^[17].

Beyond paid work, we find little evidence that teleworkability exhibits systematic associations with well-being in non-labor activities such as unpaid work or leisure, day-average AWB or cognitive well-being. Overall, the observed associations appear to be concentrated in the work domain, with limited extensions into other parts of daily life.

Importantly, when we examine differences in time use, any apparent disparities in remote-work duration between high- and low-teleworkability jobs disappear after controlling for job characteristics. This suggests that the well-being differences are unlikely to be driven simply by a higher prevalence of remote work among employees in more teleworkable jobs.

Our findings complement the mixed evidence in the existing literature on telework and well-being and highlight the importance of distinguishing between the availability of remote work and its actual use. While previous studies focus primarily on realized telework, our results indicate that the broader context of occupational flexibility is relevant for understanding workers’ emotional experiences. This suggests that teleworkability captures latent aspects of workplace flexibility – such as autonomy, expectations of availability or the option to work remotely when needed – rather than the frequency of working from home per se. For women, these latent aspects may include heightened role conflict or blurred boundaries between work and family; for men, the potential benefits may depend more on whether actual remote work opportunities are exercised.

On the other hand, the absence of a positive association between workplace flexibility and workers’ AWB, combined with evidence that employees nonetheless value remote work (Mas and Pallais, 2017), suggests that workers are not mere hedonists focused solely on maximizing momentary emotional well-being but also pursue other goals such as the well-being of their children.

Additionally, the lack of consistent and statistically significant effects for men in our study could indicate that the advantages of telework are heterogeneous and context-dependent (see Oakman et al., 2020), varying across different work environments and job characteristics. This is also in line with Song and Gao (2020), who emphasize the difference between working from home and taking work home.

Overall, our findings underline that the availability of remote work is not equivalent to its use and that the potential for flexibility can carry both benefits and hidden costs. Distinguishing between teleworkability and realized remote work is therefore essential for understanding the complex, gendered association between workplace flexibility and workers’ well-being. This calls for a more tailored approach to the design and implementation of flexible work policies, considering the diverse needs and experiences of different groups of employees.

From a managerial and policy perspective, our results suggest that flexible and hybrid work arrangements should not be treated as universally beneficial but instead as context-dependent forms of structural flexibility. The implications of remote and hybrid work policies may differ systematically across workers, particularly by gender. Gender-sensitive job design therefore requires attention not only to formal access to flexible work options but also to how these options interact with job demands, expectations of availability and the unequal distribution of paid and unpaid work. In particular, organizations may need to ensure that hybrid arrangements are implemented in ways that avoid reinforcing gendered patterns of task allocation or career penalties, for example by maintaining transparency in performance evaluation, limiting implicit expectations of constant availability and ensuring equal access to high-visibility tasks. Designing flexible work arrangements that are predictable, clearly communicated and adaptable to individual circumstances can help ensure that increased structural flexibility translates into equitable experiences and outcomes across employee groups.

Future research needs to be based on more comprehensive and detailed data on locational and temporal flexibility at the level of individual jobs (rather than using occupation-level information or restricting attention to actual remote work activities only). This will support our understanding of the consequences of both employee- and employer-sided flexibility for individual well-being. Additionally, more recent (post-COVID) and regularly collected longitudinal data on both the prevalence of realized telework and the telework potential of jobs would substantially benefit the literature. If linked to surveys on time use and AWB, such data would enable identification strategies exploiting within-person variation over time, for example through individual fixed effects models that account for time-invariant unobserved heterogeneity. Ideally, researchers could analyze experimental or quasi-experimental variation in teleworkability to convincingly identify causal effects. This could be achieved, for example, by randomly granting access to remote-work options to some workers but not others or by exploiting policy or organizational changes that differentially expand telework eligibility across occupations or teams. This would allow the literature to move beyond descriptive correlations and better understand how remote-work options shape well-being, time allocation and the broader dynamics of work–life balance.

The authors are grateful for valuable comments made by two anonymous referees as well as Philipp Biermann, Jeannette Brosig-Koch, Andrew Clark, Jan Delhey, Clemens Hetschko, Carina Keldenich, Michael Kvasnicka, Ulrike Vollstädt and participants at ESPE 2025 in Naples, SEHO 2025 in Zaragoza, ISQOLS 2025 in Luxembourg, the Department of Economics research seminar at University of Crete (2025), the BeWell meeting in Magdeburg (2024) and the FIS-Forum in Berlin (2024).

The supplementary material for this article can be found online.