Leaving Work at Work: A Meta-Analysis on Employee Recovery From Work

Theoretical Approaches to Recovery

Two complementary theories, the effort-recovery (E-R) model (Meijman & Mulder, 1998) and conservation of resources (COR) theory (Hobfoll, 1989), have served as the predominant frameworks underpinning recovery research and highlight the important role of demands and resources (respectively) for employees. The job demands-resources (JD-R) model (Demerouti et al., 2001) extends these frameworks by incorporating the roles of both resources and demands. Each of these theoretical frameworks provides a unique contribution to the recovery literature, and when viewed together, they provide an overarching understanding of how individuals unwind from the workday.

The E-R model (Meijman & Mulder, 1998) captures the process by which job demands require individuals to expend effort and suggests subsequent deleterious effects on individuals if they lack, or are unable to replenish, energies. This model is physiological in nature and explores the effects of demands and stressors on the bodily systems of the individual. Specifically, employees experience demands on the job that create burdens and “load reactions” (Meijman & Mulder, 1998). These load reactions, which manifest in numerous ways, including fatigue and the release of stress hormones (Ganster & Rosen, 2013), can build up as an individual continually performs work tasks and responds to demands, leading to a need for recovery. During recovery, demands cease, the load is lightened or removed, and the individual is able to subsequently return his or her internal systems to baseline levels absent of demands (Demerouti et al., 2009; Sonnentag, 2001). However, when individuals are unable to recover from these demands, they may be incapable of returning to baseline levels before having these systems activated again. When this happens, individuals are forced to exert additional effort to compensate for their strained state upon beginning their next episode of work, which can be increasingly draining (Geurts & Sonnentag, 2006). The resulting cyclical effect imposes even greater burdens on the employee and his or her psychophysiological systems (Demerouti et al., 2009), which is expected to lead to potentially more harmful or chronic outcomes.

While the E-R model is primarily focused on demands, COR theory (Hobfoll, 1989, 2002) dictates that individuals will seek to build, retain, and maintain resources and further suggests that threats to or loss of these resources can lead to stress-related outcomes. Using this theoretical lens, recovery is viewed as a process facilitating the replenishment of depleted resources. That is, time spent away from work engaging in various recovery activities and experiences should enable employees to restore these threatened or lost resources (Sonnentag & Natter, 2004). In fact, research suggests that this resource regeneration process can be reinforcing (Hobfoll, Halbesleben, Neveu, & Westman, 2018), as resource-rich employees should be better able to gain more resources, which informs both the recovery process and its outcomes.

Building on these two frameworks, the JD-R model (Demerouti et al., 2001) positions demands and resources as parallel mechanisms as well as suggests potential interactive effects between the two. First, job demands tax the individual and can, if unmitigated, lead to strain and impairment for the individual (Bakker & Demerouti, 2007). Second, resources operate as an energizing and replenishing process to enhance well-being and offset job demands. When job demands are chronically high and/or resources low, employees develop load reactions, such as health issues (Bakker, Demerouti, & Verbeke, 2004). Recovery is proposed to then function as an overall mediating mechanism in the relationship between these parallel energizing or straining factors and organizationally relevant outcomes, although these effects are not equivalent across recovery dimensions (Kinnunen et al., 2011). Additionally, the constellation of demands and resources (e.g., high demands–low resources, high demands–high resources, etc.) has been suggested to have implications for whether employees develop strain outcomes, suggesting interactive buffering effects (Bakker & Demerouti, 2007). When individuals have high levels of job resources, they may be better able to cope with high levels of job demands (Bakker, Hakanen, Demerouti, & Xanthopoulou, 2007), and they may not experience the deleterious outcomes that are otherwise emblematic of high job demands when resources are low. That is, recovery experiences or feeling replenished is more likely when resources outstrip demands regardless of the absolute levels of resources/demands.

Taken together, these primary frameworks are used to delineate several important assumptions about recovery. First, the E-R model indicates that for recovery to occur, demands on the individual must cease, removing these wearing burdens on the physical and biological systems (Meijman & Mulder, 1998). Importantly, these demands are not limited to job demands, as any activities that continue to tax these systems (e.g., high-duty activities after work) could impede recuperation for employees (Demerouti et al., 2009). Second, COR theory suggests that during work and nonwork hours, individuals might regenerate key resources that had been depleted by the demands of the workday. This process of replenishment may occur through a variety of activities and experiences, with existing resources potentially enhancing the process by which individuals gain more resources (Hobfoll et al., 2018). Finally, the JD-R model argues that these two processes occur in parallel based on the presence of demands and resources, whereby demands are depleting and resources are energizing, with recovery acting as a potential buffer preventing the development of strain (Kinnunen et al., 2011). Consideration of these theoretical frameworks in concert leads to a more nuanced understanding of the recovery process.

Predictors of Recovery

Given their theoretically paramount roles as drivers of the recovery process, the broad groupings of demands and resources were used to organize predictors of recovery (activities, experiences, and states), which were then further narrowed according to existing frameworks in the literature to more comprehensively examine recovery predictors. In particular, existing research has indicated that demands and resources are not limited to a single domain (i.e., work or home; Peeters, Montgomery, Bakker, & Schaufeli, 2005). Therefore, we consider both work and home demands according to an existing typology that succinctly captures the nature of the demand itself (i.e., overload, cognitive, emotional, and physical demands; ten Brummelhuis & Bakker, 2012). For resources, consistent with recovery research, we further delineate resources in terms of their domain and origin (personal or contextual resources; Hobfoll et al., 2018).

Consistent with the E-R model, demands encountered on or off the job require effort, increase negative activation, and prolong feelings of tension, which prevents or impedes recovery after work (Sonnentag, Arbeus, Mahn, & Fritz, 2014; Sonnentag & Fritz, 2015). In reviewing the interface between work and home and the nuanced effects of demands, ten Brummelhuis and Bakker (2012) suggest that demands be broken down into four subtypes: overload, physical, emotional, and cognitive. Critically, these subtypes include any demand that affects an individual whether the demand is encountered at home or on the job. Overload demands, which consist of stressors such as overly heavy workload or time pressures on the job, occur when an employee has more work than capacity (Peeters et al., 2005). Physical demands include those that require bodily efforts. Emotional demands are those that might be taxing to one’s feelings and influence one on a personal level (e.g., bullying, interpersonal conflicts), and cognitive demands require mental effort and concentration (ten Brummehuis & Bakker, 2012).

Each of these demands, while different in nature, have the potential to impede recovery. Demands limit free time available for recovery (e.g., overload demands may prevent employees from detaching from work by being occupied both during and after work; Sonnentag & Bayer, 2005) as well as increase levels of activation and tension, making it more difficult for employees to unwind (e.g., cognitive or emotional demands inhibit employees’ ability to relax after the workday; Bennett et al., 2016). Although each type of demand is expected to have a negative relationship with recovery due to its taxing nature, recovery’s relationships with physical and cognitive demands should be less negative than emotional and overload demands. While expected to inhibit overall recovery, physical and cognitive demands can also provide employees with opportunities to experience mastery or control (Crawford, LePine, & Rich, 2010; Michel, Turgut, Hoppe, & Sonntag, 2016), which facilitates, or at least reduces the inhibition of, recovery.

The influence of demands on recovery should be considered alongside the positive influence of resources. While the E-R model suggests that a cessation of demands must occur for recovery to begin, COR theory contends that resources serve a regenerative function. Further, as COR theory has developed and matured over time, a more nuanced view of resources and their role in individuals’ lives, including on their recovery from work, has taken shape. As a result, we organize our framework around ten Brummelhuis and Bakker’s (2012) delineation of resources as either contextual or personal.

Contextual resources are those located outside of one’s self (Hobfoll, 2002) and can generally be seen as occurring in two domains, work and home (ten Brummelhuis & Bakker, 2012). Contextual resources in the work domain (or job resources) are the physical, psychological, social, or organizational aspects of the job that help employees achieve goals and stimulate growth, such as job control and supervisor support (Demerouti et al., 2001). Contextual resources at work likely allow employees more flexibility and a sense of independence (Demerouti et al., 2001) and thereby facilitate employees’ ability to schedule their work to allot more time for recovery activities (Rodriguez-Muñoz, Sanz-Vergel, Demerouti, & Bakker, 2012). Similarly, resources such as supervisor support and autonomy likely reduce the extent employees worry about work at home, allowing for individuals to feel more recovered (Bakker, Demerouti, & Euwema, 2005). Still, some research suggests that job resources relate more strongly to specific recovery experiences (e.g., mastery, control, relaxation) than others (e.g., detachment; Kinnunen et al., 2011). Relatedly, contextual resources in the home domain are conceptualized as resources stemming from nonwork factors, such as family or friend support and partner facilitation of recovery (e.g., Park & Fritz, 2015; Shimazu, de Jonge, Kubota, & Kawakami, 2014). Above and beyond contextual work resources, greater levels of home-based contextual resources are likely to fuel recovery activities because the positive home environment should cause recovery activities to be more pleasant. Moreover, given that recovery occurs primarily at home, home-based contextual resources act as a form of situational determinant in the extent to which one is able to experience recovery psychologically and feel recovered.

In addition to the situational context, resources also exist within the individual. Hobfoll (2002) and ten Brummelhuis and Bakker (2012) identify several key personal resources, which are more stable characteristics that cut across situations, such as efficacies and self-esteem. ¹ Personal resources likely open up more opportunities for mastery and learning as well as increase perceptions of control over one’s time (Kinnunen et al., 2011). Moreover, personal resources might also result in greater confidence in the capacity to balance work and life, reducing one’s hesitancy to engage in recovery activities (Kirchmeyer, 2000). Finally, processes known as “gain spirals” (Hobfoll et al., 2018) position personal resources as baseline resources that help individuals obtain additional resources, further aiding recovery.

Outcomes of Recovery

Recovery has been thought to influence a wide variety of outcomes for employees (Sonnentag & Fritz, 2007). In this meta-analysis, we focus specifically on two categories of outcomes suggested by the theoretical frameworks applied to the recovery process. First, the role of recovery on employee well-being is examined in terms of psychological, psychosomatic, and physiological correlates. Inclusion of these variables was driven by the prominent role of recovery on health and well-being as suggested by the E-R model. Second, the relationship between recovery and performance is examined, as tenets of the COR theory and JD-R model suggest resources and demands have, respectively, positive and negative impact on goal attainment and performance episodes (Beal, Weiss, Barros, & MacDermid, 2005; Halbesleben et al., 2014).

Employee well-being has become a popular criterion across a variety of literatures, resulting in various conceptualizations and typologies of the construct. In this study, we examine well-being using three separate, but related, dimensions: psychological, psychosomatic, and physiological well-being. These dimensions are drawn from the allostatic load model (McEwen & Stellar, 1993) and encompass the primary reactions individuals experience from work stress (Ganster & Rosen, 2013). Psychosomatic well-being, which refers to physical conditions that may manifest from mental causes (Shorter, 1994), here includes proximal indicators like employee sleep, fatigue, or general health and somatic complaints (Bono, Glomb, Shen, Kim, & Koch, 2013; Sonnentag, Binnewies, & Mojza, 2010). For psychological well-being outcomes, this meta-analysis considers key outcomes in the recovery literature: mental well-being (e.g., low anxiety, low perceived stress), the experiences of state affect, and life satisfaction. Finally, physiological well-being consists of bodily indicators, such as cortisol, that may change in direct response to stressors (Ganster & Rosen, 2013).

Recovery likely has several important implications for individuals’ psychological well-being. First, recovery acts as a cessation of, and distancing from, demands and stressors, allowing employees’ minds to relieve themselves of taxing activation and improve their mental well-being (Sonnentag & Zijlstra, 2006). Beyond this, experiences of detachment, relaxation, mastery, and/or control allow individuals to replenish and restore lost resources, which can further enhance their return to baseline levels of activation, and encourage positive mental well-being (Mojza et al., 2011). In consideration of life satisfaction, recovery enables individuals to be more pleased with the time they spend outside of work (Sonnentag & Fritz, 2007). Time away from work spent engaging in recovery experiences, such as relaxation, mastery, or low-duty activities, allows employees to broaden their horizons and pursue personal goals and exercise discretion in how their time is spent. In this way, feelings of control and accomplishment associated with certain recovery activities or experiences should lead to a more global sense of subjective well-being as well as an inducement of positive affect and/or reduction of negative affect (Newman, Tay, & Diener, 2014). Research has also shown these types of activities elevate individuals’ satisfaction with their lives (Judge, Bono, Erez, & Locke, 2005) through resource replenishment or fulfillment of other basic psychological needs (Davidson et al., 2010; Newman et al., 2014). Additionally, time spent engaging in recovery experiences can induce positive affect even if only serving as a distraction from demands (Sonnentag & Bayer, 2005).

Generally speaking, and consistent with the predictions of the E-R model (Hahn, Binnewies, Sonnentag, & Mojza, 2011; Meijman & Mulder, 1998), research has suggested that recovery also enhances well-being outcomes with manifestations in the physical body (e.g., psychosomatic and physiological indicators) as the strains on employees’ internal systems (effort expenditure required to meet demands) are reduced. Existing research has shown this process to hold for a number of different types of psychosomatic well-being variables, including general health complaints, fatigue, and sleep (Ganster & Rosen, 2013). For example, smartphone use after work can result in reduced sleep for employees during the evening, likely because of continued activation and lack of detachment from the workday (Lanaj, Johnson, & Barnes, 2014). Engaging in demanding activities (either from work or home) or failing to relax and disconnect may prolong activation and subsequent strain without allowing for resource restoration. If not reduced over time, this increased effort expenditure can result in chronic, long-term health issues (Geurts & Sonnentag, 2006). Further, because the body’s systems are inherently interdependent, the poor sleep, fatigue, and general poor health that are likely to occur as a result of insufficient recovery could exacerbate each other, consistent with the idea of the body’s system developing “load reactions” (Meijman & Muelder, 1998). Like other research areas on employee stress, considerably less attention has been given to recovery’s relationship with physiological well-being indicators (Ganster & Rosen, 2013), such as cortisol or blood pressure, and the scant findings have been inconclusive (e.g., Bono et al., 2013; Dettmers, Vahle-Hinz, Bamberg, Friedrich, & Keller, 2016; Gustafsson, Lindfors, Aronsson, & Lundberg, 2008; von Thiele, Lindfors, & Lundberg, 2006). However, theory generally suggests that when employees are able to engage in recovery activities and experiences or feel refreshed and replenished, they should experience reduced physiological strains (and their respective biomarkers) and prevent the development of chronic health issues (Meijman & Mulder, 1998; Sonnentag et al., 2010). Thus, each of these factors, while distinct, should be positively related to recovery.

Performance concerns are perhaps one of the most influential reasons why organizations and employees are mindful of the recovery process (Fritz & Sonnentag, 2005). Employee performance, conceptualized here as the totality of one’s task performance and organizational citizenship behaviors (OCBs), has had a somewhat ambiguous relationship with recovery in extant literature. For example, researchers have found no effect (Fritz & Sonnentag, 2006), nonlinear effects (Fritz, Yankelevich, Zarubin, & Barger, 2010), and mixed effects depending on the specific recovery experience (Eschleman, Madsen, Alarcon, & Barelka, 2014; Shimazu, Sonnentag, Kubota, & Kawakami, 2012). As a result, Sonnentag and colleagues (2017: 369) conclude in their review of the literature that “taken together, findings remain inconclusive.” Despite this empirical uncertainty in the literature, there is a preponderance of theoretical arguments suggesting that recovery positively predicts performance. For example, consistent with the mechanisms of COR theory, if employees are able to recover and replenish resources, they then should be better positioned to invest these resources into fulfilling their job requirements (Beal et al., 2005; Fritz & Sonnentag, 2005). Empirical evidence further suggests that the resource replenishing and protective effects of engaging in activities and recovery experiences outside of work carry over to the work domain. For example, engaging in nonwork creative activities was positively associated with OCBs on the job (Eschleman et al., 2014). Similarly, a recovered employee may have additional resources to invest in assisting others in the workplace, thus leading to an increase in OCBs (Binnewies et al., 2010). Together, theory suggests that performance as a whole should benefit from the important resource gains that are expected to result from the recovery process as resources are utilized by individuals toward the attainment of goals (Halbesleben et al., 2014).

Literature Search

To identify studies for possible inclusion, an extensive literature search using multiple electronic databases, including EBSCOhost, PsycINFO, and ProQuest Dissertations, was conducted, which returned both published and unpublished studies. When searching for studies, large umbrella terms (including wildcard characters, denoted with an asterisk [*]) were used (i.e., job recovery, employee recovery, and work recovery). Each of the dimensions of recovery experiences in conjunction with the word recovery itself (i.e., recovery mastery, recovery control, recovery detach*, and recovery relaxation) was used as well. To identify other unpublished articles, conference programs from the Academy of Management and Society for Industrial and Organizational Psychology were searched. Requests for studies were sent out on major listservs (e.g., OBWeb, HRDiv). Additionally, and consistent with other meta-analytic work (e.g., Berry, Ones, & Sackett, 2007; Swider & Zimmerman, 2010), the authors also conducted a manual search of all articles that cited the Recovery Experiences Questionnaire (REQ; Sonnentag & Fritz, 2007), the seminal work of Etzion et al. (1998), and articles included in recent reviews (Bennett et al., 2018; Sonnentag et al., 2017; Sonnentag & Fritz, 2015). Results yielded 3,723 studies for possible inclusion in the meta-analysis, including 347 dissertations.

Inclusion Criteria

The identified studies were divided evenly among the four study authors and were reviewed to determine the appropriateness for inclusion. To be included in a meta-analysis, the study needed to meet several criteria. First, the study needed to empirically measure recovery (self-reported) and at least one other variable of interest at the between-person level (or at the within-person level, e.g., daily diary data, with a correlation aggregated to the between-person level provided). Second, the study needed to report empirical results regarding the relationship(s) between the variable(s) of interest and recovery (i.e., was not a pre- and posttest of a variable separated by a recovery period, such as a vacation or weekend). Third, the study had to measure recovery from some type of work rather than general physiological recovery from injury, laboratory tasks, or illness and not occur in a clinical population of human subjects.

The recovery literature features several different types of longitudinal study designs (e.g., daily diary/experience sampling methodology [ESM] studies, panel design studies). However, only between-person correlations from longitudinal studies were included in a meta-analysis due to the limitations inherent to meta-analytic procedures. Within-person correlations, which are typically included in daily diary/ESM studies (e.g., the within-person correlation between all 363 day-level measures of work demands and detachment across the 96 study participants; Park, Fritz, & Jex, 2018), violate the assumptions of independence necessary for these types of analyses (Hunter & Schmidt 2004), while reported between-person correlations also reported in most daily diary/ESM studies do not. While not feasible given the size of this study, estimating meta-analytic within-person effects requires that raw data from every study at the person-observation level be gathered, aggregated, and analyzed to account for the interdependence of observations provided by each individual at each measurement occasion for every study (e.g., Sitzmann & Yeo, 2013). However, to uncover any potential differences or patterns among these within- and between-person correlations, all included daily diary/ESM studies that provided both within- and between-person correlations were qualitatively reviewed. This qualitative review is available in Table S1 of this article’s online supplemental materials. Yet unlike the single within-person correlations typically reported in daily diary/ESM studies, we were able to use the longitudinal data from panel design recovery studies to calculate the stability analyses, because each measurement occasion was reported separately and not collapsed across multiple measurement occasions of each participant. For example, studies that reported recovery experiences at multiple time points (e.g., detachment measured at Time 1 and Time 2; Sonnentag et al., 2014) were included in these analyses. Similar procedures have been used in other meta-analyses that have estimated stability of a given construct over time (e.g., Riketta, 2008).

To determine inclusion, each article was independently reviewed by two authors. Each study had a primary and secondary reviewer (89.2% initial agreement). Any study where authors disagreed about inclusion or exclusion was resolved by a third author to reach a final decision. After reviewing the 3,723 identified articles, 188 articles, with 198 independent samples, met the inclusion criteria and were used in the final meta-analyses. ²

Coding Procedures and Scheme

Prior to coding, the authors met to discuss the coding scheme and criteria, coded one article as a group, and then coded the same five, randomly selected articles independently. The authors met again to discuss discrepancies stemming from the training exercise. After all issues and differences were resolved, the authors divided up all articles for coding. The author team coded multiple characteristics of each study, including sample size, measures of the independent and dependent variables, effect size and reliabilities of each variable, sample characteristics, and study methodologies. ³ Once coding was completed, each coder took another coder’s work to review for any errors, and disagreements were marked (95.5% agreement level). All disagreements were discussed among authors until 100% agreement was reached.

Meta-Analytic Procedures

Random-effects method of meta-analysis (Schmidt & Hunter, 2015) was used to generate effect size estimates for all meta-analyses in this study. Corrections at the individual-study level were used (Schmidt & Le, 2004) as most studies included reliability estimates for measures of recovery and variables of interest. Specifically, reliability estimates (i.e., alphas) reported in each study were used to correct correlations for unreliability in measures of recovery and other variables. ⁴ In instances when the primary study failed to report information on the reliability of a measure, the mean reliability across studies measuring that variable was used. For studies measuring recovery activities with single-item measures of time or count data, a conservative approach was taken and a reliability of 1.00 was assumed and imputed (Wanous & Hudy, 2001).

Only one correlation per sample was used for each separate meta-analysis to maintain statistical independence. When multiple studies utilized the same data, the sample with the largest sample size was retained. The mean was used when sample sizes were reported as ranges. To retain as much useful statistical information as possible, we included an effect size in our meta-analyses for each unique variable. When a study included measures of two or more variables within a broad antecedent or outcome grouping (e.g., overload demands and physical demands), both variables were retained in their respective meta-analyses. If multiple correlations between a given variable and recovery were reported for a single sample (e.g., anxiety measured at Time 1 and Time 2), then an average correlation was used. When a study included multiple constructs within the same variable grouping (e.g., both time pressure and overload as overload demands), these effects were averaged. Relatedly, if multiple studies from the same data provided measures of different constructs within the same variable grouping (e.g., one study reported detachments’ correlation with job control and the other study reported the correlation with social support), then these correlations were averaged to a single correlation. “All recovery experiences” correlations included any single correlation of REQ dimensions as well as the within-sample means from any studies reporting correlations with multiple REQ dimensions and a given variable.

In addition to the average correlation ( r ¯ ) and the corrected correlation (ρ), the 95% confidence interval (CIs) around the mean corrected correlation was calculated using the standard error of the mean corrected correlation. CIs excluding zero suggest that the mean corrected correlation differs from zero. In addition to CIs, 80% credibility intervals (CVs) were also calculated to examine corrected correlation variability across studies. CVs that exclude zero indicate that a relationship is expected to exist across all situations (i.e., generalizable), though the magnitudes of the relationship may vary (Hunter & Schmidt, 2004). Additionally, the overall percentage of variability in observed correlations that could be accounted for by known statistical artifacts (e.g., sampling error, unreliability) was also computed. When less than 75% of the variance in observed correlations can be accounted for by artifacts, moderators may exist (Schmidt & Hunter, 2015).

Temporal Considerations

To provide further insight into the temporality between recovery and its correlates, we conducted additional analyses on a subset of the included studies. Specifically, we examined the relationships between demands, resources, and recovery where relationships were temporally lagged (i.e., demands/resources were measured prior to recovery measures) as well as between recovery and outcomes in a similar fashion (i.e., recovery variables were measured prior to outcome measures) in the primary studies. When there were enough samples to calculate the time-lagged relationships, the results generally supported the hypothesized relationships albeit with slightly smaller effect sizes on average than those generated in the main study meta-analyses. Complete results of these analyses are included in the article’s online supplementary materials in Tables S5a to S5e. These results provide additional support to the temporal ordering of the included variables and also augment prior meta-analytic work examining temporally separated measures of recovery antecedents and outcomes (Bennett et al., 2018).

To further explore the potential differences in effect sizes stemming from temporal factors, we conducted supplemental moderator analyses in which we examined the effect of overall study design (e.g., cross-sectional vs. longitudinal) on the magnitude of effects. Studies categorized as “longitudinal” included both daily diary/ESM designs as well as traditional time-lagged panel designs. Results of these analyses were generally consistent with hypothesized effects, with longitudinal study designs often, but not always, being smaller in magnitude than cross-sectional designs when both had sufficient samples for analysis. Complete results of these analyses are available in the online supplemental materials (Tables S4a to S4t). In addition to these supplemental quantitative analyses, Table S1 contains a qualitative review of all primary studies that included both within- and between-person correlations. These analyses largely suggested that aggregated within- and between-person correlations did not meaningfully differ in any specific or identifiable pattern in terms of the direction and magnitude of the reported effects.

Theoretical Implications

In reviewing and summarizing this rapidly growing literature, we provide clarity around a collection of somewhat inconsistently applied conceptualizations of recovery as a set of activities, as experiences, or as a state. First, our findings test a major assumption of the recovery literature, which is that engaging in recovery activities and experiences leads to a state of being recovered. We found that a state of being recovered is positively related to recovery experiences (i.e., REQ) and low-duty activities as well as negatively related to high-duty activities, albeit at small to moderate magnitudes (ρs of .34, .20, and –.10, respectively). These findings raise an important issue for scholars studying recovery. Specifically, given these relationships, scholars should be mindful of their research question and the conceptualization of recovery suggested by their theory as well as how they interpret existing findings. That is, while (modestly) related, recovery activities, experiences, and state are each distinct elements of the overall recovery process and add value when they are theoretically matched to the research question at hand. They should not be used interchangeably. The weaker effects across both sets of activities, for instance, underscore the need for this specificity, as a given activity (e.g., childcare) may be depleting for one individual but relaxing or energizing for others. The activity could even vary within an individual (e.g., playing quietly with a child vs. dealing with misbehavior). The four recovery experiences also exhibited differential relationships with other conceptualizations of recovery as well as recovery antecedents and outcomes, underscoring the importance of choosing theoretically and situationally relevant operationalizations. Although the state of being recovered tended to manifest the largest effect sizes with recovery outcomes, likely due to more directly assessing the phenomenon in question, conceptualizing recovery as experiences may provide researchers with more actionable how findings compared to simply concluding that the end state of being recovered is important.

Second, we summarize the theoretical approaches to recovery (i.e., the E-R model, COR theory, and JD-R model) and apply these theoretical mechanisms to meta-analytically examine the relationship between demands, resources, and employee recovery. Our analyses produced broad empirical support for these frameworks, as demands and resources were shown to have negative and positive relationships (respectively) with state of being recovered as well as recovery experiences. That said, the magnitudes of effect size estimates suggest that resources, which have generally received less theoretical attention as predictors of recovery in extant literature, warrant more consideration from researchers. Interestingly, personal resources had the largest effect sizes with recovery of any resource or demand examined in this study. Given that personal resources are largely stable characteristics, these findings lend credence to the idea that recovery may have substantial variance explained by individual differences. This may also help explain the high levels of temporal consistency found for recovery experiences, as prior measures of recovery explained up to 49% of the variance in future measures of recovery. For demands, physical demands were the only demands employees experience that did not consistently relate to lower recovery. Although based on only a few samples, the weak physical demands–recovery relationship suggests that increasing the physical demands at work would not be expected to unilaterally inhibit employee recovery (Calderwood, Gabriel, Rosen, Simon, & Koopman, 2016). Overall, as demands and resources encompassed both the home and work domains, our empirical findings indicate that scholars should consider how recovery, an inherently nonwork activity, is impacted by nonwork demands and resources in addition to work-specific demands and resources.

The findings of this study also underscore the critical role that recovery plays in an individual’s psychological and psychosomatic well-being as well as job performance. The meta-analytic results suggest recovery acts as a critical mechanism by which employees can improve overall well-being. Furthermore, it is important to note that recovery was found to be roughly as related to psychosomatic outcomes as psychological outcomes. Simply put, recovery from work is beneficial for the body and mind both in the short term, as evident by the relationships with sleep and state positive and negative affect (respectively), and the long term, as evident by the relationship with life satisfaction. Additionally, findings indicate a modest yet consistent positive relationship between recovery and job performance, which clarifies existing murky results in the literature (Binnewies et al., 2009; Fritz et al., 2010; Fritz & Sonnentag, 2006; Sonnentag et al., 2017). Given the criticality of performance to employees and organizations (Austin & Villanova, 1992), findings of a generalizable positive relationship between recovery and performance highlight the beneficial role and regenerative properties of spending time away from work engaging in nonwork experiences (particularly, mastery and control) and low-duty activities.

Finally, this study also contributes to the examination of recovery at the within-person level. Building upon initial work summarizing within-person designs for detachment specifically (Sonnentag & Fritz, 2015), we sought to detail whether and how the body of studies included in this meta-analysis demonstrated differential effects when examining within- and between-person correlations. Overall, the results of our qualitative review suggest that for the most part, correlations at the within and between levels are similar in direction and magnitude. Correspondingly, a recent quantitative review of within-person research found only a small minority of recovery studies (5%) exhibited differences in sign and significance for between- and within-person correlations (McCormick, Reeves, Downes, Li, & Ilies, 2018). When reviewing these primary studies, it was common to see researchers examining recovery and its correlates within person employing nearly identical arguments as those made by researchers focusing on these relationships exclusively at the between-person level (i.e., homologous relationships). Going forward, studies testing the homology of recovery relationships across levels will likely provide limited novel insights into the phenomenon (Gabriel et al., 2018). Instead, results from our qualitative review suggest that many of the key insights gained from daily diary studies of recovery occur when stable, between-person factors are shown to impact within-person dynamics of recovery (i.e., cross-level moderation). These nonhomologous examinations undoubtedly help inform our understanding of how recovery changes or remains stable for employees both in the short and long term.

Managerial Implications

The findings in this study also provide considerable practical implications for organizations and employees. Our results emphasize that job demands are negatively related to the recovery process and that employees are more likely to recover when they have sufficient resources, particularly, personal resources. Although it is impossible for organizations to shield their employees from job demands entirely, organizations should consider keeping demands in check when they can be controlled (e.g., not repeatedly creating time pressures via tight deadlines) or offering opportunities to build personal resources, such as psychological capital or (recovery-specific) self-efficacy. Additionally, organizations may seek to implement policies and norms that encourage employee recovery, such as segmentation norms that regulate work e-mail response expectations and other after-work activities that may cut into opportunities to engage in recovery activities or experiences. Overall, organizations should seek to promote and facilitate recovery not only for the psychological well-being of their employees but also to capture the benefits of increased employee performance.

Further, as recovery is positively related to the psychosomatic health of employees, organizations may be able to save significant money in employee health care costs when their employees are able to effectively recover. One way organizations can help to ensure they are facilitating recovery is to encourage managers to provide sufficient contextual resources, such as social and instrumental support, to their subordinates. Organizations can also work to design jobs with greater autonomy and flexibility, which may allow employees to better plan and regulate job demands on their own (and subsequently make time to detach) as well as when and how they recover from work (and subsequently experience control or mastery). Organizations have faced skyrocketing health care costs during the past few decades (PWC, 2015), so encouraging employee recovery may provide an avenue for organizations to simultaneously reduce total labor costs while improving employee health, well-being, and performance.

Just as organizations must be cognizant of, and take steps to facilitate, employee recovery, so, too, must employees. If made aware of the deleterious consequences associated with a lack of recovery, employees may be more motivated to prioritize after-work recovery by choosing to make, and actually utilize, time for detachment and relaxation whenever possible. As organizations may not always be successful in providing sufficient resources to employees, or are forced to craft jobs with high overload or cognitive demands, it will be crucial for employees to proactively seek out ways to increase resources in and outside work as well as their own personal resources (e.g., psychological capital; Luthans, Avey, & Patera, 2008). This may require employees to make planned, significant changes in their work or nonwork routines to improve their recovery. For instance, sleep was not only found to be one of the strongest, most consistent outcomes of recovery but may also be seen by employees as a controllable life factor that would help build resources, reduce demands, and possibly prevent chronic issues (e.g., health problems; Geurts & Sonnentag, 2006; Sonnentag et al., 2008), which are all additionally associated with recovery.

Limitations and Future Research Directions

The demands-and-resources organizing framework utilized in this meta-analysis, which also serves as the framework in several primary studies (e.g., Kinnunen et al., 2011), often implies causal relationships. Although these frameworks squarely place demands, resources, and recovery in a given sequential order, our meta-analysis cannot empirically tease out the definitive temporal ordering of these variables. Thus, we are unable to assess causation in the relationships examined in this study. It is possible that recovery reduces demands or that more recovered individuals can accrue more resources (e.g., social support) or even that these relationships both occur but with imbalanced effects. Further, given this resource accrual, individuals with a strong sense of psychosomatic or psychological well-being may be better able to engage in the recovery process. Future research employing longitudinal designs will allow for greater clarity surrounding the “process” elements of recovery. In fact, there is likely value in examining how specific elements in the recovery process unfold over time and impact employee outcomes at both the within- and between-person levels. For instance, research on employee mindfulness and mind wandering (Good et al., 2016) may help explain the causes and consequences of a situation when an employee engaging in a recovery activity during nonwork hours switches back and forth between focusing on the recovery activity and focusing on work-related issues.

Additionally, a closer examination of the primary studies identified through our search efforts revealed a paucity of research on physiological indicators of well-being relative to other conceptualizations of well-being. While this is not inconsistent with other literatures examining stress (see Ganster and Rosen, 2013, for a review), likely because of the inherent difficulty in obtaining these data, it remains an area ripe for investigation. Specifically, physiological indicators of load reactions can help scholars and practitioners understand the biological mechanisms through which recovery activities, experiences, and states influence employees’ bodies. This would answer questions about how much recovery is necessary to reduce the physiological influences of work on individuals, determine whether certain recovery experiences or activities have stronger influences on the biophysiological systems, and provide a more direct test of the E-R model in application to recovery. Moreover, and of particular interest to employers, future research should continue to further document the significant relationships between recovery (especially detachment) and performance (and the various facets of performance), as performance was understudied in comparison to other outcome variables, like fatigue or general health. Additional research regarding moderators to these performance–recovery relationships could also help shed light on previously unclear findings. Greater research on performance would produce actionable, more fine-grained findings regarding the beneficial effects of recovery for employees and employers.

To spotlight the importance of recovery on a number of thematically relevant variables, we focused on main effects. Yet, the results of our efforts also reveal the potential for meaningful moderators (i.e., low percentage of variance accounted for by statistical artifacts; Hunter & Schmidt, 2004) in virtually every relationship between recovery, its antecedents, and its outcomes. Additionally, meta-analytic methodology precluded the examination of cross-product interactions (Aguinis, Gottfredson, & Wright, 2011), so we were unable to fully test the JD-R model, which suggests that the unique combination of both demands and resources best predicts recovery. As detailed in Table S1 in the online supplement, several studies that examined both within- and between-person effects found support for cross-level interactions that could not be examined meta-analytically. Incorporating cross-level moderators, and even same-level moderators, represents an important opportunity for researchers looking to extend the literature by exploring situations in which the recovery process might be experienced differently at the within- versus between-person level (McCormick et al., 2018). Even instances where we found large effects that are expected to generalize across all situations (e.g., mental well-being), substantial variance in observed effect sizes remained. Overall, we resoundingly encourage researchers to add important nuance to the recovery literature by examining models containing moderation, moderated mediation, and cross-level moderation when possible and to explicitly consider the role of time in examining these constructs at both the within- and between-person levels.

Finally, we examined the temporal stability of various recovery conceptualizations, with results indicating that individuals exhibit substantial consistency in their state of being recovered as well as their experiences of detachment, relaxation, and mastery over time. That is, certain individuals may be more or less prone to engage in recovery experiences due to personality factors or routines, regardless of situational or contextual factors. Additionally, it may also be that characteristics of individuals’ work or home demands remain somewhat consistent over time. Future research should address how short-term (e.g., intershift or weekend) recovery processes compound or build up over longer periods of time to result in chronic consequences or sustainable benefits to employees. Alternatively, researchers could use multiple within-person observations of recovery to create aggregate between-person constructs, such as variability of individuals’ detachment, to use as meaningful predictors or outcomes in a study. Taken together, it is possible that individuals have baseline levels of demands and resources, and as such, research is needed to understand how these relatively stable factors can be adjusted or changed over time to help facilitate the recovery process.

Recovery is critical for employees facing the demands of today’s challenging and dynamic business environment. The findings of this meta-analytic investigation establish recovery’s impact on employees’ psychosomatic health, psychological welfare, and job performance. Beyond this, workplaces stand to gain healthier and more productive workers when their employees are able to experience recovery and feel refreshed. Research should continue to build on these findings by exploring the wide array of strategies that employees, leaders, organizations, families, and even governments can employ to capture the benefits of recovery across the domains of life.