ASRXXX10.1177/0003122415622391American Sociological ReviewMoen et al. 6223912015

American Sociological Review 2016, Vol. 81(1) 134­–164 Does a Flexibility/Support © American Sociological Association 2015 DOI: 10.1177/0003122415622391 Organizational Initiative http://asr.sagepub.com Improve High-Tech Employees’ Well-Being? Evidence from the Work, Family, and Health Network

Phyllis Moen,a Erin L. Kelly,b Wen Fan,c Shi-Rong Lee,a David Almeida,d Ellen Ernst Kossek,e and Orfeu M. Buxtond

Abstract This study tests a central theoretical assumption of stress process and job strain models, namely that increases in employees’ control and support at work should promote well-being. To do so, we use a group-randomized field trial with longitudinal data from 867 information technology (IT) workers to investigate the well-being effects of STAR, an organizational intervention designed to promote greater employee control over work time and greater supervisor support for workers’ personal lives. We also offer a unique analysis of an unexpected field effect— a company merger—among workers surveyed earlier versus later in the study period, before or after the merger announcement. We find few STAR effects for the latter group, but over 12 months, STAR reduced burnout, perceived stress, and psychological distress, and increased job satisfaction, for the early survey group. STAR effects are partially mediated by increases in schedule control and declines in family-to-work conflict and burnout (an outcome and mediator) by six months. Moderating effects show that STAR benefits women in reducing psychological distress and perceived stress, and increases non-supervisory employees’ job satisfaction. This study demonstrates, with a rigorous design, that organizational-level initiatives can promote employee well-being.

Keywords subjective well-being, flexibility, organizational intervention, work-family, gender

Scholars have long theorized social structures, aUniversity of Minnesota b contexts, and policies as central to physical and Massachusetts Institute of Technology cBoston College emotional health (cf. Berkman, Kawachi, and dPennsylvania State University Glymour 2014; House 2002; Schoeni et al. ePurdue University 2008; Tausig and Fenwick 2011). This “social determinants of health” framing has generated a Corresponding Author: lively field of sociological research on the social Phyllis Moen, Sociology Department, University of Minnesota, 1123 Social Science Building, 267 causes of illness, health, and subjective well- 19th Avenue South, Minneapolis, MN 55455 being (Aneshensel, Rutter, and Lachenbruch E-mail: [email protected] Moen et al. 135

1991; Link and Phelan 1995; Turner, Wheaton, Third, fundamental cause theory is used to and Lloyd 1995; Wheaton 2001; Wheaton and explain the ongoing relationship between Clarke 2003; Wickrama et al. 1997). This work socioeconomic status (SES) and health/well- emphasizes the embeddedness of individuals in being outcomes over time (Link and Phelan particular social structures, with corresponding 1995; Link et al. 2008; Phelan et al. 2004; risks, rules, claims (or needs/demands), and Phelan, Link, and Tehranifar 2010). People resources that shape their subjective well-being with higher SES more often possess and have (Keyes 1998; Kohn and Schooler 1982; access to resources—such as economic, cul- Mirowsky and Ross 2003a, 2003b; Ross and tural, and social capital—that protect them Mirowsky 1992, 2003; Ryff and Keyes 1995). from multiple deleterious health outcomes. The social determinants of health perspec- The fourth thread emphasizes the social gra- tive proposes that changing the social context dient, arguing that occupational status pre- should, therefore, promote or reduce health dicts health, well-being, and mortality outcomes. This is the thesis underpinning our (Marmot et al. 1997; Marmot et al. 1991). study, in which we sought to implement a Because we are testing whether a deliber- change in the pivotal social environment of ate change in the work environment promotes paid work. Specifically, we draw on a group- subjective well-being, and not examining the randomized field trial of an organizational reproduction of advantages/disadvantages intervention designed to promote control over among people with different status locations, work time and supervisor support for employ- we draw on stress process and job strain theo- ees’ personal and family life. We collected ries. We control for location in the status hier- longitudinal data from 867 information tech- archy by sampling workers in one industry nology (IT) workers in a Fortune 500 corpo- (information technology) and one organiza- ration to assess whether such changes do, in tion (a Fortune 500 firm). This professional fact, improve four measures of subjective sample was a deliberate choice because we well-being: burnout, job satisfaction, per- were seeking employees in a demanding work ceived stress, and psychological distress. environment but without a raft of other stress- Studying the well-being effects of increasing ors associated with low income that might work-time control and support also informs override any work initiative we could design. scholarship on work-family conflict, gender, In other words, this sample allows us to inves- and work-time mismatches (Angrave and tigate the effects of changes in the work envi- Charlwood 2015; Reynolds and Aletraris ronment among workers who face high work 2006; Schieman, Milkie, and Glavin 2009). demands, but who are unlikely to face addi- tional neighborhood and family stressors tied to poverty and financial struggles that would Theoretical not be addressed by this intervention. Underpinnings Four theoretical threads guide much of the research on the social determinants of health. Theorizing Change: The Stress Sociologists have drawn on stress process Process Approach theory, which contends that structural condi- Stress has been defined as the mismatch tions produce stress that affects subjective between claims and resources, or the misfit well-being (Pearlin 1999; Pearlin et al. 1981). between a person and her environment (Kaplan A second thread is the considerable interdisci- 1983). A long tradition of theory and research plinary scholarship on the job strain theoreti- on family stress (e.g., Hill 1949; Hochschild cal model, where high strain is the result of 1989), life course processes (Elder, George, high demands combined with low control and and Shanahan 1996; Moen and Roehling low support (Johnson and Hall 1988; Karasek 2005), and stress more generally (e.g., Lazarus 1979; Karasek and Theorell 1990, 2000). and Folkman 1984; Mirowsky and Ross 2003a, 136 American Sociological Review 81(1)

2003b; Pearlin et al. 1981; Pearlin et al. 2005; demanding jobs with low job control and low Turner et al. 1995) depicts stress as occurring support—are the most at risk of occupational when a gap between resources and claims (i.e., hazards tied to poor health. Karasek the demands, needs, and expectations that indi- (1979:290) describes job control as an viduals face) reduces people’s sense of control. employee’s “potential control over his tasks Scholars underscore that shifts in both and his conduct during the working day,” resources and claims alter the social environ- operationalizing job control as “decision ment over the life course (e.g., Elder 1998; authority” and “intellectual [or skill] discre- Pearlin et al. 1981; Pearlin et al. 2005). Pearlin tion.” Job control has been empirically linked, (1989:241) points out the importance of in cross-sectional and longitudinal studies, to observing “how deeply well-being is affected exhaustion and depressive symptoms by the structured arrangements of people’s (­Mausner-Dorsch and Eaton 2000), psycho- lives and by the repeated experiences that stem physiological stress responses, alcohol use, from these arrangements.” high blood pressure, heart disease (e.g., The structural arrangements of paid work Bosma, Stansfeld, and Marmot 1998), mental are fundamental to well-being, given that and physical health (D’Souza et al. 2003; work is pivotal to identity, meaning, routine, Stansfeld and Candy 2006), and work-family and status, as well as income, and occupies conflict and strain (Thomas and Ganster many waking hours of adults in the work- 1995). Ample evidence thus links job control force. For contemporary workers, role strain with subjective well-being as well as physical is often implicitly or explicitly about time health (see also de Lange et al. 2004; Van Der (e.g., gaps between the time needed for occu- Doef and Maes 1999). pational and family demands, and how to Research using the job strain model typi- schedule and arrange work and family tasks cally compares workers with considerable so multiple responsibilities can be met) (see resources (like high levels of control and sup- also Clawson and Gerstel 2014; Fenwick and port) to other workers whose job conditions Tausig 2007; Jacobs and Gerson 2004; Moen offer few such resources (little control or sup- 2003; Schieman, Whitestone, and Van Gundy port). These research designs raise issues of 2006). This was the rationale for the develop- selection, because different types of workers ment of our intervention aimed at providing choose or are allocated to particular job envi- workers with more control over their time and ronments that provide more or less control training supervisors to recognize the conflict- and more or less support. In other words, it is ing work and home pressures workers face. difficult to disentangle the well-being effects Note that we are examining workers’ control of stress-reducing organizational resources over their work time, not reductions in work (such as control and support) from the well- time such as part-time hours. Research shows being effects of other, unmeasured variables that people with more ability to manage their that may lead workers to jobs with different time and more support from their supervisors resources. report greater well-being. The challenge is to devise and test whether an organizational Combining these Approaches and change giving workers greater control over Key Contributions their time and more supervisor support improves their well-being. What is missing in the job strain literature are the health and well-being effects of employ- ees’ control over work time and supervisors’ Theorizing Stressful Work support of family and personal lives. The job Conditions: The Job Strain Model strain model focuses on employees’ control The job strain model proposes that people over how they perform their work, but many with the greatest strain—workers in highly employees are stressed because they do not Moen et al. 137 have control over their work time (see Kelly, within-person change in well-being over time Moen, and Tranby 2011; Lyness et al. 2012; in light of changes in their work-related Moen, Kelly, and Hill 2011; Moen, Kelly, and resources. We do so through a rigorous field Lam 2013; Moen, Kelly, Tranby, and Huang trial design, randomizing work teams to 2011). Similarly, the job strain model ties “STAR” (treatment) and “usual practice” (con- support at work to health but does not attend trol) groups. This permits an assessment of specifically to workplace support for non- whether a deliberate shift in the social environ- work domains. ment produces corresponding shifts in well- What is missing in the stress process litera- being outcomes. This is arguably one of the ture is recognition of the importance of delib- strongest tests of stress process theory to date. erately changing conditions and the importance A substantial literature explores the rela- of the meso-level work context, that is, the tionship between work conditions and subjec- ways shifting policies and practices of a spe- tive well-being (e.g., Karasek 1979; Marklund, cific work organization affect individual Bolin, and von Essen 2008; Mausner-Dorsch workers’ (micro-level) mental health. The and Eaton 2000; Thomas and Ganster 1995), stress process approach recognizes that but relatively few longitudinal studies exam- changing resources and claims or needs ine changes in these relationships over time enhance or reduce people’s sense of control (but see Kleiner and Pavalko 2010; Scott- over their lives and affect their well-being Marshall 2010). Occupational health scholars (Gotlib and Wheaton 1997; Pearlin et al. have long emphasized workplace health pro- 2005). For example, becoming unemployed motion through individual behavioral change (Pearlin 1989) or taking on caregiving of an (e.g., reducing smoking [see Cahill and Lan- aging parent (Aneshensel et al. 1995) caster 2014; Kristensen 2000]). These schol- increases stress and reduces feelings of well- ars are increasingly investigating deliberate being. Longitudinal studies capturing chang- interventions to change work conditions more ing environments are thus key to capturing broadly to improve employees’ well-being, the stress process. However, researchers can- but this literature is largely divorced from not randomly assign respondents to unem- sociologists’ analysis of the social determi- ployment or caregiving, for instance, and nants of health and well-being. For example, various selection effects or corollary stressors a review of (non-experimental) organizational affecting the odds of those changes may also intervention studies “found some evidence of affect well-being. This study thus extends the health benefits (especially beneficial effects stress process literature to investigate whether on mental health, including reduction in anxi- and how deliberate randomized organiza- ety and depression) when employee control tional changes may promote well-being. improved or (less consistently) demands The importance of our study to both stress decreased or support increased” (Egan et al. process and job strain theories is that we 2007:945). But even in this field, randomized move beyond comparisons of different types controlled trials are rare. None were included of workers (status differences) or those in dif- in systematic reviews of the health effects of ferent environments (e.g., high versus low organizational interventions (Egan et al. control jobs) to study organizational changes 2007), flexible work interventions and health that may reduce workers’ stress and promote (Joyce et al. 2012), or schedule control and their well-being. We are thus testing the stress health (Nijp et al. 2012).1 Yet randomized process approach and a variant of the job field studies of the effects of organizational strain model (looking at time control rather interventions on changes in subjective well- than job control) in a deliberate experiment, being over time are key to establishing causal not simply comparing people in different cir- relationships (Morgan and Winship 2007; see cumstances. Specifically, we investigate also Bianchi and Milkie 2010). 138 American Sociological Review 81(1)

Design Previous analyses show that employees This is a randomized field trial assessing the randomized to STAR subsequently report effect of an organizational intervention on a more schedules they describe as “variable” range of measures of subjective well-being rather than a set shift, work more at home, among employees of a large U.S. firm we call feel more control over their work time, and TOMO. The workplace initiative, STAR see their supervisors as more supportive of (Support. Transform. Achieve. Results.), was their personal and family commitments, com- developed by the interdisciplinary Work Fam- pared to employees in the control group ily and Health Network (WFHN), a national (Kelly et al. 2014). We analyze STAR’s consortium of researchers working with orga- impact on subjective well-being outcomes to nizational development experts to create and begin investigating whether and how these implement an evidence-based intervention to workplace changes affect employees’ health. improve the work-family interface and health Specifically, we identify the effects of STAR (King et al. 2012). on changes in four measures of well-being Both stress process and job strain theories over a 12-month period: burnout, job satisfac- emphasize the importance of work resources, tion, perceived stress, and psychological dis- and the implicit assumption in both is that tress. Such subjective well-being measures increasing those resources should promote are important in terms of intrinsic life quality, well-being. We fashioned the STAR inter- but they are also related to other health out- vention based on stress process and job strain comes and have the potential to decrease theories, together with insights garnered stress-related symptoms and illnesses over a from previous pilot studies (Hammer et al. longer period of time (Melamed et al. 2006; 2011; Kelly et al. 2011; Kossek et al. 2011; Stansfeld et al. 2002). We focus on this set of Moen, Kelly, and Hill 2011; Moen et al. well-being outcomes to assess whether out- 2013; Moen, Kelly, Tranby, and Huang comes proximal to work (burnout and job 2011). Details on the STAR intervention and satisfaction) may be more sensitive to STAR its rollout are provided below, but we address effects than more global measures (perceived the underlying logic here. Schedule control stress and psychological distress). appears to be related to but distinct from tra- We also consider how the timing of an ditional measures of job control (Moen, unexpected stressful organizational event—a Kelly, and Huang 2008; Moen et al. 2013; merger—may have moderated any STAR Schieman et al. 2009). Similarly, having a effects. Because a merger was announced dur- supervisor supportive of family and personal ing the course of data collection, we are able life has been shown to promote well-being to investigate whether the deliberate changes (Hammer et al. 2011; Kossek et al. 2011). tied to the STAR intervention, including STAR aims to (1) increase employees’ con- increased support and control over work time, trol over their work hours and schedules and had different effects depending on whether (2) increase employee perceptions of super- they were experienced prior to the shock of visor support for employees’ work and fam- the merger announcement or after the stress of ily/personal lives. STAR’s third goal, tied to the impending merger was introduced. research on the organizational rewards asso- ciated with “face time” and long hours, is to reorient the work culture toward results, Hypotheses rather than hours at the workplace (Kossek Taken together, the job strain and stress pro- et al. 2014). We address the question: Does cess approaches lead us to theorize that delib- providing workers real flexibility in the form erate organizational changes targeting of greater control over their working time employees’ control over their work time and and more supportive supervisors improve the support they receive should produce cor- their subjective well-being? responding changes in employees’ subjective Moen et al. 139 well-being. Unlike worksite health promotion increasing workers’ sense that their supervi- interventions that target stress or well-being sors are supportive, reducing workers’ work- directly as a private individual trouble (e.g., family conflict, encouraging workers to shift yoga classes or mindfulness training [see their schedules and work more at home, and Hartfiel et al. 2011; Kabat-Zinn 2003]), STAR reducing workers’ feelings of burnout. We aims to change the rules of the game at work treat burnout—specifically emotional exhaus- (including the policies, practices, interac- tion—as an outcome, because it is an impor- tions, and expectations that structure employ- tant measure of psychological health, but we ees’ lives on the job) and affect how work and also investigate whether changes in burnout personal life can be managed effectively and by six months mediate the effects of STAR on with less strain. STAR facilitates new prac- the other well-being outcomes. tices, such as increased options to work at Increasing workers’ schedule control and home and flexibility to shift one’s schedule, sense of having a supportive supervisor are but also emphasizes that employees have con- key goals of the STAR intervention and have trol over when, where, and how they do their been shown to change in the expected direc- work and that managers and co-workers sup- tions by the six-month follow-up (Kelly et al. port their efforts to address family and per- 2014). Given that these two resources are sonal responsibilities (Kelly et al. 2014). strongly correlated with subjective well-being Stress process theory suggests that this work outcomes (Grzywacz, Carlson, and Shulkin redesign intervention may increase energy 2008; Moen, Kelly, Tranby, and Huang 2011), resources and decrease exposure to stresses we expect changes in them to mediate the by giving employees control to “fit” the salutary well-being effects of STAR. Work- pieces of their lives together more easily and family conflicts (from family to work as well feel supported in those efforts, thereby as from work to family) are a common form improving subjective well-being. Similarly, of stress in the lives of contemporary workers the job strain model suggests that increasing (Bianchi, Casper, and King 2005; Bianchi and employees’ control over where and when they Milkie 2010; Major, Klein, and Ehrhart 2002), do their work and providing greater support as is burnout (Halbesleben and Buckley 2004; for family and personal life will enhance Maslach, Schaufeli, and Leiter 2001; well-being. Schaufeli and Bakker 2004). We test whether the STAR intervention promotes subjective Hypothesis 1: Employees randomized to the well-being through the reduced work-to-family STAR initiative will experience positive and family-to-work conflict observed previ- changes in indicators of well-being (an in- ously (Kelly et al. 2014). Shifting work crease in job satisfaction and decreases in schedules and working at home may also burnout, perceived stress, and psychologi- enable workers to adjust their work arrange- cal distress) compared to employees work- ments to better fit their individual and family ing under usual practices (control group). needs, thereby improving well-being.

We also investigate possible mediators of Hypothesis 2: Changes in workers’ percep- the relationship between STAR and subjec- tions of schedule control, family support- tive well-being to help identify mechanisms ive supervisors, work-family conflict, and through which workplace interventions may burnout, along with increases in working at benefit employees. We theorize a mediated home and shifts in schedules, will mediate model in that we anticipate STAR will have STAR effects on subjective well-being. direct effects in enhancing workers’ subjec- tive well-being, but will also operate in part The stress process approach emphasizes through five theorized mechanisms: increas- potential differences in well-being outcomes ing workers’ sense of schedule control, based on status locations (Pearlin 1999; 140 American Sociological Review 81(1)

Pearlin et al. 1981). This theory underscores the IT division commonly piloted new initia- the inequality in stress and well-being based tives developed in-house or brought in by on mostly invariant status locations like gen- consultants. der, race, and social class. Additionally, role STAR involved eight hours of participatory strain, or “the felt difficulty in fulfilling role training sessions that encouraged teams and obligations” (Goode 1960:483), is a chronic managers to identify new work practices and stressor found more often among people in processes that would increase employees’ con- different status locations. Accordingly, we trol over their work time and focus on key examine variations in STAR effects for par- results (rather than simply face time at work). ticular subgroups, theorizing that workers in For example, there were extensive discussions this professional sample most at risk of concerning communicating by instant messen- chronic stress at home and on the job, and ger, coordinating more efficiently with off- facing the greatest work-family strains, will shore staff, and better anticipating and handling benefit most from STAR. (Naturally, STAR periods of high demands, such as around soft- may be too small a change to address the ware releases. Additionally, training activities many multifaceted stressors—including eco- proposed that being “always on”—quickly nomic insecurity—confronting less privi- responsive via e-mail, instant messenger, or in leged workers; hence our focus on person—did not necessarily mean workers professionals.) were productive or efficient in accomplishing the most important tasks. Teams varied in the Hypothesis 3: Workers with more family-care changes they implemented regarding work obligations (i.e., women generally, or women processes and coordination, but all workers or men with children at home) or fewer job- randomized to the STAR initiative experienced related resources (employees compared to considerable flexibility as to where and when managers) will benefit more from STAR. they worked, including the ability to work at home without requiring a supervisor’s permis- We also investigate age-cohort differences in sion. Previously, work at home was sometimes the effects of STAR, given the different family- allowed with managerial approval, but such a care obligations and different expectations case-by-case “accommodation” does not shift regarding work and family across age-cohorts control to employees (Kelly et al. 2014; Moen, (Galinsky, Aumann, and Bond 2009). Kelly, Tranby, and Huang 2011). The STAR intervention also involved supervisor training to encourage supervisor Organizational Changes: support for family/personal life and profes- Star and an Unexpected sional development; this involved four hours Merger of additional training sessions as well as cus- The STAR intervention was delivered as a tomized computer-based training for managers pilot initiative in the information technology that included a video message from a senior (IT) division of a Fortune 500 company and executive endorsing STAR. Given their cen- was announced by senior executives. STAR trality to STAR (and the theories that informed was not presented as part of the larger study the intervention), we analyze changes in work- investigating how work affects health and ers’ sense of control over work time and loca- personal life broadly, but instead as a com- tion, and their supervisor being supportive of pany pilot that might well be implemented family and personal life, as likely mechanisms throughout the IT division eventually. linking the STAR initiative with improvements Although STAR was developed jointly by in workers’ well-being. For more information WFHN researchers and outside consultants, on STAR, its development, and its customiza- the training was delivered by consultants. tion for this organizational culture, see Kelly This pilot process was not seen as unusual; and colleagues (2014), Kossek and colleagues Moen et al. 141

(2014), and STAR materials at http://www loss. The limited research, though, suggests .workfamilyhealthnetwork.org. that organizational interventions aimed at Randomized field trials offer the strongest increasing employees’ control have no positive causal evidence for the effects of interventions effects in the context of downsizing (Egan et al. (Hannan 2006; Oakes and Kaufman 2006). 2007). Yet, any field study is vulnerable to real-world We are able to use this conjunction of the events. In this case, a merger was announced planned and unplanned organizational during our data collection, while STAR train- changes to suggest scope conditions for the ing was occurring for some groups randomized benefits of initiatives like STAR. The timing to the intervention. This unexpected event, of the baseline survey and the introduction of portending significant organizational changes, the STAR intervention in relation to the exog- directly affected the study, because the survey enous shock of the merger announcement and STAR were rolled out to different units of adds complexity to the study, but it also ena- this IT workforce over an extended period of bles us to assess whether an increase in about 12 months.2 Such external shocks cannot resources (supervisor support and schedule simply be ignored in research (although they control) prior to an exogenous stressor often are). Moreover, this surprise event (merger announcement) operates differently offered the opportunity to further capture the than introduction of an intervention following stress process (Pearlin 1989; Pearlin et al. such an exogenous shock. We expect the 1981; Pearlin et al. 2005): suddenly learning Early Survey Group (who experienced STAR one’s employer is to be taken over by another before learning about the merger) may have corporation generates a pervasive sense of been more open to this work redesign and uncertainty about the future (Lam et al. 2015). thus better positioned to benefit from it. This Everyone in the study was exposed to the group is the focus of our analysis. merger announcement and process, but the Although implementation of STAR was timing vis-à-vis our data collection suggested identical for the Early and Late Survey Groups, different outcomes by survey timing. The the ways employees interpreted the interven- Early Survey Group was interviewed first and, tion may not be the same. Workers undergoing for those randomized to the experimental con- the baseline survey and STAR training after dition, began the STAR intervention prior to the merger announcement (Late Survey Group) the merger announcement.3 The Late Survey may not have taken the STAR intervention Group was exposed to the merger announce- seriously, given the uncertainties and disloca- ment before the baseline survey and before tions fueled by this announcement. Indeed, STAR. Because of this difference in timing, observations of STAR training sessions occur- we consider these two groups separately. As ring after the merger announcement revealed we will describe, we propose that STAR will that employees routinely and explicitly ques- have stronger effects on well-being outcomes tioned whether STAR would “survive” the for respondents in the Early Survey Group, merger and whether it fit with the emerging who had completed or initiated STAR prior to culture of the combined firm. Moreover, the the merger announcement. Late Survey Group’s baseline responses In effect, the merger provided an opportu- already reflect the effects of knowing about the nity to assess the influence of an unexpected merger, leading to some differences from the organizational stressor at about the same time baseline responses of the Early Survey Group an intervention to improve well-being was (see Lam et al. 2015). Differences between the introduced. Mergers and restructuring are part two groups may arise from different perspec- and parcel of corporate life, but few studies tives because the Late Survey Group knew of examine organizational interventions in the the upcoming takeover. Accordingly, we pre- context of other ongoing organizational changes sent the results separately for these two groups that may involve new roles, relocation, or job and focus on the Early Survey Group. 142 American Sociological Review 81(1)

Methods eligible employees participated (N = 823); 87 percent (N = 717) and 85 percent (N = 701) of Randomization baseline employee participants were retained The randomization process began by group- in the 6-month and 12-month follow-ups, ing employees and managers in the IT divi- respectively. The response rate at baseline sion into 56 study groups. Some groups among managers was 86 percent (N = 221), consisted of large teams reporting to the same with 89 percent (N = 196) and 85 percent manager; other study groups included combi- (N = 188) of baseline managers completing nations of several smaller teams who either the 6-month and 12-month follow-ups. We reported to the same senior leadership or conducted analyses among respondents who worked closely together on the same applica- completed all three survey waves.5 tion. We developed a randomization design Study groups began the study on a rolling that would ensure balance on job function, basis over a year to accommodate limited division (reporting to a particular VP), and staff and space for the in-person interviews. size of study group, modifying a biased-coin When the Late Survey Group went through randomization technique for use with group STAR training, they already knew about the randomization (Bray et al. 2013; Frane 1998). impending takeover by another organization. All eligible work units within the IT division In contrast, the Early Survey Group rand- participated in and were randomized to STAR omized to STAR went through the initiative or to usual practice. early in the roll-out, before learning of the looming merger. Study Recruitment and The Early Survey Group analytic sample Data Collection consisted of 455 respondents nested in 32 study groups. The Late Survey Group’s We use data collected at three time points— answers at baseline reflect the fact that they baseline, six months, and 12 months—from already knew about the upcoming merger. employees and managers in two cities who However, for comparison, we show results were part of the Information Technology (IT) for both survey groups in Tables 1 and 2. division of this major U.S. firm.4 Recruitment Results for the pooled sample (867 respond- materials emphasized the value of a study ents in 56 study groups) are shown in Table S2 investigating the connections between of the online supplement (http://asr.sagepub employees’ work, family, and health for the .com/supplemental); these models include a employees (who received some personalized variable merger announcement timing distin- health information), the employing organiza- guishing the Early Survey Group (= 1) from tion, and scientific knowledge more broadly; the Late Survey Group (= 0). there was no reference to STAR. Recruitment materials also emphasized the independence of the research team from TOMO and the Measures confidentiality of individual data. Computer- assisted personal interviews lasting approxi- Our principal explanatory variable is expo- mately 60 minutes were conducted at the sure to the STAR intervention, with respon- workplace on company time. dents randomized to the intervention (STAR) coded 1 and the controls (usual practice) coded 0. This specification reflects an intent- Sample to-treat analysis, in which all respondents The sample combines responses from employ- randomized to the intervention are assumed ees and managers; they answered the same to be “exposed” to the treatment of STAR. In survey questions, except managers were not this analytic sample, mean attendance rate asked about perceived stress (saving time for was 76 percent of sessions. Among this other questions). At baseline, 70 percent of group, 9 percent (n = 20) attended fewer than Moen et al. 143 half of the STAR sessions, and only 3 percent conflict, suggesting it is not, in fact, a media- (n = 7) of those randomized to STAR attended tor. Variable schedule distinguishes work none of the sessions. Intent-to-treat is the schedules that may change from a set day- more conservative approach for evaluating time, evening, or night schedule. We con- effects of interventions, preserving the identi- structed hours working at home based on the fication strategy of the experimental design question, “About how many hours/week do (Begg et al. 1996; Friedman, Furberg, and you work or take calls from home on this DeMets 2010). job?” We analyze burnout as a measure of Details of measures, including items and subjective well-being (when we assess change alphas, are available in Table S1 in the online by 12 months) and a potential intervening supplement. Emotional exhaustion (burnout) mechanism (change by six months in burnout consists of a three-item subscale of the may mediate other 12-month outcomes). Maslach Burnout Inventory (Maslach and Subgroup analyses investigate whether Jackson 1986). Job satisfaction is measured respondents with greater vulnerability or by a three-item scale developed by Cammann resources relevant to the work-family inter- and colleagues (1983) on global job satisfac- face benefit more from the STAR interven- tion. Perceived stress is evaluated with a four- tion. We defined three key subgroups on the item scale validated by Cohen, Kamarck, and basis of gender (women coded as 1), age- Mermelstein (1983) that has been found to be cohort (respondents born before 1955 coded predictive of physical and mental health out- as Leading-Edge Boomers, respondents born comes.6 This measure was only on the between 1955 and 1964 coded as Trailing- employee survey; managers are therefore not Edge Boomers, and respondents born in 1965 included in the analysis of perceived stress. and later coded as GenXers), and managerial Psychological distress is captured by the K6, status (official supervisors coded as 1). See a six-item scale validated by Kessler and col- Table 1 for descriptive statistics on these leagues (2003) that is widely used in the moderators. We also investigate four sub- United States as an assessment of nonspecific groups that combine gender and the presence psychological distress. of a child under age 18 at home: women with To preserve temporal order, we constructed children at home, women with no children at mediators as six-month changes, subtracting home, men with children at home, and men Wave 1 from Wave 2, in the following varia- with no children at home. bles. Schedule control reflects the degree to which respondents felt they had control over their work time, with eight items adapted Analysis from Thomas and Ganster (1995). Family To examine the effects of STAR on subjective supportive supervisor behaviors (FSSB) well-being, we estimate mixed-effects models assesses employee perceptions of supervi- with respondents nested within study groups, sors’ behavioral support for integrating work the unit randomized to STAR (treatment) or and family (Hammer et al. 2013). Family-to- usual practice (control). Well-being outcomes work conflict reflects the degree to which role were measured at Wave 3, 12 months after responsibilities from family or personal life baseline. We regressed these outcomes on are incompatible with participation in the their lagged measures at baseline and condi- work role; we use a five-item subscale devel- tion (STAR or control). We additionally con- oped by Netemeyer, Boles, and McMurrian trol for gender, age-cohort, and manager (1996). We also tested the effects of changes status to improve precision of estimates, but in work-to-family conflict (Netemeyer et al. the pattern of findings does not change when 1996) as a potential mediator. For the Early these are omitted. Coefficients for STAR cap- Survey Group, however, STAR does not sig- ture effects of the intervention on outcomes nificantly predict changes in work-to-family (Hypothesis 1). Results for the Early Survey 144 American Sociological Review 81(1)

Group are our main models, presented in We present moderation results (Hypothesis Table 2 and compared side-by-side with the 3) with figures, showing STAR effects for dif- Late Survey Group. ferent subgroups. We conducted the media- To test mediation (Hypothesis 2), we exam- tion and moderation analyses using the Early ine the effects of changes in potential media- Survey Group, who took the baseline survey tors over a six-month period on subsequent and experienced STAR before learning about changes (by 12 months) in subjective well- the merger. being measures. We adopt two approaches to test mediation. One is the classic Baron and Results Kenny (1986) approach, which requires satis- faction of three conditions: (1) the key inde- Descriptive Statistics pendent variable (i.e., STAR) is a significant Table 1 shows descriptive statistics by experi- predictor of the dependent variable (i.e., well- mental condition for the Early Survey Group being outcomes); (2) the key independent vari- (Panel A) and Late Survey Group (Panel B). able is a significant predictor of the mediator; For our focal Early Survey Group sample, and (3) the coefficient for the key independent subjective well-being and demographic vari- variable is greatly reduced when adding the ables at baseline are balanced by experimen- potential mediator. Condition 1 is established tal condition, as would be expected due to in Table 2, and Condition 2 is established in randomization. The only exception is age- Table 1 (based on t-tests, but results remain cohort membership (by chance). even with mixed-effects models adjusting for We find no differences in baseline well- covariates). In Table 3 (Section 1 for each being outcomes among respondents in the panel), we assess Condition 3 by evaluating Early Survey Group (Panel A); but as hypoth- the extent to which the STAR coefficients are esized, by Wave 3 the Early Survey respond- weakened after mediators are added. ents in STAR had significantly lower levels of A second approach for mediation analysis burnout ( p < .001), perceived stress ( p < .05), builds on Sobel’s test (Sobel 1982) and and psychological distress ( p < .05), as well addresses concerns that Baron and Kenny’s as noticeably higher job satisfaction ( p < .01). approach does not directly estimate and quan- This provides initial support for Hypothesis 1. tify the magnitude and significance of the By contrast, the Late Survey Group (Panel B) mediation effect (i.e., to what extent the showed no difference in well-being outcomes mediator mediates the predictor’s effect on at Wave 3 by STAR condition. the outcome). The canonic Sobel’s test was We calculated changes in perceptions of developed for single-level models and suffers schedule control and family supportive supervi- low statistical power (MacKinnon, Warsi, and sor behavior, reductions in family-to-work con- Dwyer 1995), which is problematic given our flict and burnout, and increased adoption of multilevel data structure and relatively small variable schedules and working from home as sample of groups. We therefore use a method the first step in testing potential mediators developed by Krull and MacKinnon (2001) between STAR and changes in well-being out- for multilevel models that combines the comes by 12 months. Previous research demon- Sobel’s test with bootstrapping to obtain the strates that over a six-month period, STAR sampling distribution of the mediation effect improved employees’ sense of schedule control non-parametrically. We conducted this set of and family supportive supervisor behaviors, analyses using the ml_mediation program in while reducing family-to-work conflict (Kelly Stata; the mediation effects were obtained by et al. 2014), findings replicated in Table 1 for bootstrapping the ml_mediation command the Early Survey Group (Panel A). Over six with 1,000 replications with a seed value of 1. months among the Early Survey Group sample, Corresponding results are shown in Section 2 schedule control increased by .31 for STAR for each panel in Table 3. respondents, whereas the increase was merely *

** ** *** *** *** t -test .33 .3 .31 .47 .59 .58 .45 .44 .5 .49 .38 .47 .75 .8 SD 6.9 1.09 3.02 2.67 2.68 1.46 2.39 1.41 ( N = 210) .06 Usual Practice 1.06 –.24 –.05 –.06 8.23 4.05 9.87 3.9 7.94 0% 4.01 3.71 12% 10% 11% 67% 26% 44% 39% 17% 32% 10.38 Mean/% .36 .24 .41 .5 .53 .59 .47 .43 .49 .48 .42 .5 .77 .75 SD 1.15 3.24 2.77 3.19 1.58 2.8 1.54 Panel B: Late Survey Group 12.17 .23 9.32 6% –.26 –.03 –.01 8.46 4.03 4.41 8.16 4.1 3.98 15% 22% 57% 24% 41% 37% 22% 46% 10.55 10.03 STAR ( N = 202) STAR 100% Mean/% * * * * * ** ** ** ** *** *** *** *** t -test .32 .36 .32 .49 .55 .63 .52 .37 .5 .47 .3 .49 .75 .77 SD 7.97 1.19 3.24 2.68 3.58 1.45 2.85 1.41 ( N = 219) .08 .07 .02 .04 2.21 8.81 3.95 0% 4.52 8.8 3.89 4.54 Usual Practice 12% 15% 12% 61% 16% 57% 33% 10% 38% 11.09 10.83 Mean/% .36 .25 .43 .5 .57 .77 .62 .41 .5 .5 .35 .48 .79 .71 SD 1.41 3.04 2.7 3 1.53 2.56 1.52 11.67 Panel A: Early Survey Group .16 .31 7.6 7% –.26 –.04 8.63 3.95 4.37 8.18 4.12 3.92 STAR ( N = 234) STAR 15% 24% 54% 21% 42% 44% 14% 36% 10.78 10.12 Mean/% 100% p < .001 (two-tailed). *** p < .01; ** GenXers (b.1965 to 1980) Trailing-Edge Boomers (b. 1955 to 1964) Trailing-Edge Leading-Edge Boomers (b. 1946 to 1954)

Yes both waves Yes Yes –> No Yes No –> Yes No both waves Managerial Status Women (= 1) Women Baseline Psychological Distress Baseline Perceived Stress Baseline Job Satisfaction Experimental Condition (STAR = 1) Experimental Condition (STAR

Δ Work Hours at Home (W2–W1) Δ Work Δ Variable Schedule Δ Variable Δ Burnout (W2–W1) Δ Family-to-Work Conflict (W2–W1) Δ Family-to-Work Δ FSSB (W2–W1) Δ Schedule Control (W2–W1) Study Group Level Psychological Distress by Wave 3 Psychological Distress by Wave Individual Level Perceived Stress by Wave 3 Perceived Stress by Wave Job Satisfaction by Wave 3 Job Satisfaction by Wave Burnout by Wave 3 Burnout by Wave p < .05;

Potential Mediators Age Cohort Age

Independent Variables

Burnout Baseline

Table 1 . Descriptive Statistics of the Analytic Sample Table Dependent Variables Early Survey Group refers to respondents who did not know about the merger until after their baseline survey. Late Survey Group refers to respondents Note: Early Survey Group refers to respondents who did not know about the merger until after their baseline survey. and usual practice (control) groups are tested using Mean differences between STAR who were aware of the upcoming merger prior to baseline survey. t -tests. *

145 146 American Sociological Review 81(1)

.04 for the control group ( p < .001). A similar Cohen’s (1988) rule of thumb, but in general pattern occurred for family supportive supervi- are higher for burnout (.36) and job satisfac- sor behaviors (.16 for STAR and .02 for control, tion (.27) than for perceived stress (.18) and p < .05). Family-to-work conflict decreased by psychological distress (.18). By contrast, for .04 for respondents randomized to STAR, respondents in the Late Survey Group (taking whereas it increased by .07 for the control group baseline survey after merger announcement), ( p < .05). About one in four STAR respondents STAR had no well-being effects. This sug- reported their schedules changed from “not var- gests STAR benefited the subjective well- iable” to “variable,” compared with 12 percent being of workers who were not worried about in the control group ( p < .001). Over the six the merger at the time they began or received months, STAR respondents increased their time the STAR training. Given the null STAR working at home by 7.6 hours a week, signifi- effects for the Late Survey Group, we con- cantly higher than the two-hour increase among ducted all subsequent analyses on the Early control respondents ( p < .001). Finally, we Survey Group. observe a reduction in burnout over six months As a sensitivity check, we also conducted for STAR respondents but not for the control analyses for the full sample (see Table S2 in group (–.26 versus .08, p < .01). These findings the online supplement). For this combined lay the groundwork for testing their effects as sample, we find that STAR significantly mediators between STAR and changes in well- decreased burnout (–.289, p < .01) and being. Also note that, except for perceived and increased job satisfaction (.160, p < .01) by actual schedule flexibility, respondents in the Wave 3. These effects are maintained even Late Survey Group did not experience the desir- after controlling for the effect of being in the able changes brought about by the STAR inter- Early or Late Survey Group. Given that the vention (Panel B). Late Survey Group’s baseline responses are confounded by their prior exposure to the stress of the merger, results from the Early Effects of STAR on Subjective Survey Group represent the most valid tests Well-Being of STAR’s effects on well-being. Table 2 shows the effects of STAR on burnout, job satisfaction, perceived stress, and psycho- logical distress at 12 months for Early (Panel Mediation Analysis of STAR on A) and Late (Panel B) Survey Groups. We Subjective Well-Being present the outcomes in this order because the To more precisely assess how STAR improves first two are distinctively job-related, whereas subjective well-being in the Early Survey the latter capture general states of subjective Group, we examine specific mechanisms well-being. Results from the multilevel analy- theorized to account for the relationship sis show that by Wave 3, STAR respondents in between the organizational intervention and the Early Survey Group (not exposed to well-being: increasing schedule control and merger announcement when randomized to family supportive supervisor behaviors STAR) experienced significantly reduced (FSSB), decreasing family-to-work conflict emotional exhaustion/burnout (–.534, p < and burnout, and increasing schedule flexibil- .001) and increased job satisfaction (.208, p < ity and working at home. .01), as well as lower perceived stress (–.478, We use two methods to test mediation, p < .05) and psychological distress (–.570, p < Baron and Kenny’s approach (Sections 1 of .05) than similarly situated respondents in the Table 3) and a modified Sobel’s test, tailored control group. We further calculated effect for multilevel models with bootstrapped stand- sizes by dividing STAR coefficients (from ard errors (Sections 2 of Table 3). These two Table 2) by the standard deviation of the out- methods do not always coalesce; our discus- come at baseline. STAR effects are mostly sion focuses on similar patterns found in both small (.2) to moderate (.5) as judged by approaches. First, psychological distress has *** 35 412 1899 .092 .599 .402 .119 .00371 .019

.403 (.233) (.036) (.260) (.235) (.306) (.251) –.093 –.066 5.132 Wave 3 Wave Distress, Psychological *** 35 309 1361 .126 .594 .0153 .063 .777 .371 (.255) (.043) (.242) (.308) (.263) –.137 Stress, –.177 –.152 4.072 Wave 3 Wave Perceived *** 35 Job 412 .100 .584 .091 .009 .021 .018 .0457 .381 .352 .333 840.3 (.081) (.041) (.072) (.065) (.085) (.069) –.066 Wave 3 Wave Panel B: Late Survey Group Satisfaction, *** 35 412 1346 .584 .078 .193 .090 .038 .0278 .603 .369 (.139) (.038) (.133) (.120) (.157) (.128) –.060 –.083 1.316 Wave 3 Wave Burnout, * *** 0 32 455 2239 .608 .004 .604 .515 .000 .348 (.254) (.040) (.322) (.263) (.409) (.275) –.570 –.528 7.115 Wave 3 Wave Distress, Psychological * *** 32 370 1676 .593 .047 .102 .073 .006 .029 .922 .326 (.238) (.043) (.238) (.362) (.248) –.478 Stress, 4.756 Wave 3 Wave Perceived ** *** * 32 Job 869 455 .208 .567 .179 .071 .046 .022 .008 .344 .489 .364 (.066) (.036) (.071) (.058) (.090) (.061) –.021 Wave 3 Wave Satisfaction, Panel A: Early Survey Group *** *** 0 32 453 .575 .146 .040 .000 .338 1497 (.114) (.038) (.144) (.117) (.182) (.122) –.534 –.165 –.046 1.414 1.000 Wave 3 Wave Burnout, p < .001 (two-tailed). *** p < .01; ** Leading-Edge Boomers (b. 1946 to 1954) Trailing-Edge Boomers (b. 1955 to 1964) Trailing-Edge Explained Explained p < .05;

STAR (= 1) STAR . Linear Mixed-Effects Models Predicting Subjective Well-Being by Wave 3, by Early/Late Survey Group by Wave 2 . Linear Mixed-Effects Models Predicting Subjective Well-Being Table

Lagged Dependent Variables at Baseline Lagged Dependent Variables

Managers (= 1) (= 1) Women

Age Cohort (Ref. = GenXers, b. 1965 to 1980)

ICC Study-Group Variance Number of groups BIC Individual Variance Observations Proportion of Study-Group-Level Variance Proportion of Study-Group-Level Variance Proportion of Individual-Level Variance Proportion of Individual-Level Variance Early Survey Group refers to respondents who did not know about the merger until after their baseline survey. Late Survey Group refers to respondents Note: Early Survey Group refers to respondents who did not know about the merger until after their baseline survey. Proportion of study-group-level variance and proportion individual-level who were aware of the upcoming merger prior to baseline survey. explained are calculated in comparison with corresponding null models (models no covariates) that not shown here. Proportion of study-group-level variance explained is undefined for psychological distress, because the study group in null model zero (and hence denominator zero). *

147

(continued) : ** * * .235 .096 (.123) (.105) (.084) (.102) (.156) (.006) –.374 –.262 –.149 –.002 *** – (.122) (.006) (.029) –.492 –.006 –.030 *** – .177 .017 (.118) (.157) (.019) –.506 *** ** * .311 (.116) (.100) (.021) –.477 –.041 7.89% *** *** – (.116) (.080) (.024) –.495 –.267 –.036 *** *** * (.119) (.100) (.039) –.418 –.342 –.098 18.95% *** (.114) –.534 STAR (= 1) STAR Δ Schedule Control (W2–W1) Δ FSSB (W2–W1) Conflict (W2–W1) Δ Family-to-Work Δ Variable Schedule No (W1) to Yes (W2) Schedule No (W1) to Yes Δ Variable Hours at Home (W2–W1) Δ Work Indirect Effects (bootstrapped standard errors in second row, proportion of total effects mediated [“–” if mediators not significant] in third row) Indirect Effects (bootstrapped standard errors in second row, Schedule Control (W2–W1)–>Outcome STAR–>Δ FSSB (W2–W1)–>Outcome STAR–>Δ STAR–>Δ Family-to-Work Conflict (W2–W1)–>Outcome Family-to-Work STAR–>Δ (W2)–>Outcome Schedule No (W1) to Yes Variable STAR–>Δ Hours at Home (W2–W1)–>Outcome Work STAR–>Δ . Multilevel Mediation Analysis Predicting Subjective Well-Being by Wave 3, Early Survey Group by Wave 3 . Multilevel Mediation Analysis Predicting Subjective Well-Being Table Panel A: Burnout, Wave 3 Panel A: Burnout, Wave 1. Baron and Kenny's Approach: Estimates from Mixed-Effects Models with Mediators Added

2. Sobel's Test with Bootstrapped Standard Errors: Estimates of Indirect Effects from Multilevel Mediation Analysis 2. Sobel's Test

148 ** *** .167 .022 .041 .001

.141 (.042) (.050) (.022) (.076) (.063) (.052) (.003) –.039 –.039 (continued) : *** .216 .002 (.065) (.003) ** – .198 (.065) (.078) (.009) –.011 –.001 ** *** * .176 .027 (.063) (.021) (.012) –.077 13.33% ** * – .192 .014 (.063) (.049) (.010) –.110 ** *** – .190 .176 .022 (.061) (.039) (.013) ** ** * .171 .128 .035 (.064) (.049) (.015) 17.11% ** .208 (.066) STAR (= 1) STAR Δ Family-to-Work Conflict (W2–W1) Δ Family-to-Work Δ Burnout (W2–W1) (W2) Schedule No (W1) to Yes Δ Variable Δ Work Hours at Home (W2–W1) Δ Work Δ Schedule Control (W2–W1) Δ FSSB (W2–W1) Indirect Effects (bootstrapped standard errors in second row, proportion of total effects mediated [“–” if mediators not significant] in third row) Indirect Effects (bootstrapped standard errors in second row, Schedule Control (W2–W1)–>Outcome STAR–>Δ STAR–>Δ FSSB (W2–W1)–>Outcome STAR–>Δ STAR–>Δ Family-to-Work Conflict (W2–W1)–>Outcome Family-to-Work STAR–>Δ Burnout (W2–W1)–>Outcome STAR–>Δ (W2)–>Outcome Schedule No (W1) to Yes Variable STAR–>Δ Panel B: Job Satisfaction, Wave 3 Panel B: Job Satisfaction, Wave 1. Baron and Kenny's Approach: Estimates from Mixed-Effects Models with Mediators Added

2. Sobel's Test with Bootstrapped Standard Errors: Estimates of Indirect Effects from Multilevel Mediation Analysis 2. Sobel's Test

Table 3 . (continued) Table

149 *

.016 .164 .298 .205

(.248) (.222) (.176) (.218) (.090) (.320) (.012) –.417 –.349 –.005 (continued) : * .011 – (.014) (.243) (.012) –.571 –.009 * (.231) (.310) –.479 –.354 * ** .233 (.227) (.081) –.495 * – .338 (.229) (.203) (.035) –.504 –.045 * – .006 .001 (.231) (.158) (.025) –.526 * – (.233) (.198) (.047) –.513 –.144 –.036 * (.238) –.478 STAR–>Δ Work Hours at Home (W2–W1)–>Outcome Work STAR–>Δ STAR (= 1) STAR Δ Schedule Control (W2–W1) Δ FSSB (W2–W1) Conflict (W2–W1) Δ Family-to-Work Δ Burnout (W2–W1) (W2) Schedule No (W1) to Yes Δ Variable Δ Work Hours at Home (W2–W1) Δ Work Indirect Effects (bootstrapped standard errors in second row, proportion of total effects mediated [“–” if mediators not significant] in third row) Indirect Effects (bootstrapped standard errors in second row, Schedule Control (W2–W1)–>Outcome STAR–>Δ FSSB (W2–W1)–>Outcome STAR–>Δ STAR–>Δ Family-to-Work Conflict (W2–W1)–>Outcome Family-to-Work STAR–>Δ Table 3 . (continued) Table

Panel C: Perceived Stress, Wave 3 Panel C: Perceived Stress, Wave 1. Baron and Kenny's Approach: Estimates from Mixed-Effects Models with Mediators Added

2. Sobel's Test with Bootstrapped Standard Errors: Estimates of Indirect Effects from Multilevel Mediation Analysis 2. Sobel's Test

150 * ***

.103 .229 .076

.771 (.269) (.233) (.186) (.225) (.101) (.341) (.013) –.205 –.135 –.014 (continued) : – (.074) (.264) (.013) –.049 –.420 –.021 * – .064 (.036) (.254) (.340) –.036 –.502 *** – .315 (.045) (.250) (.092) –.064 –.389 *** .911 (.247) (.212) –.384 (.252) (.173) –.453 –.186 * * (.256) (.214) (.059) –.383 –.428 –.118 23.47% * (.254) –.570 STAR–>Δ Variable Schedule No (W1) to Yes (W2)–>Outcome Schedule No (W1) to Yes Variable STAR–>Δ Hours at Home (W2–W1)–>Outcome Work STAR–>Δ STAR–>Δ Burnout (W2–W1)–>Outcome STAR–>Δ STAR (= 1) STAR Δ Schedule Control (W2–W1) Δ FSSB (W2–W1) Conflict (W2–W1) Δ Family-to-Work Δ Burnout (W2–W1) (W2) Schedule No (W1) to Yes Δ Variable Δ Work Hours at Home (W2–W1) Δ Work Indirect Effects (bootstrapped standard errors in second row, proportion of total effects mediated [“–” if mediators not significant] in third row) Indirect Effects (bootstrapped standard errors in second row, Schedule Control (W2–W1)–>Outcome STAR–>Δ Table 3 . (continued) Table

3 Panel D: Psychological Distress, Wave 1. Baron and Kenny's Approach: Estimates from Mixed-Effects Models with Mediators Added

2. Sobel's Test with Bootstrapped Standard Errors: Estimates of Indirect Effects from Multilevel Mediation Analysis 2. Sobel's Test

151

– (.080) –.110 – .006 (.033) * (.058) –.114 22.72% * (.052) –.118 23.57% – (.027) –.023 p < .001 (two-tailed). *** p < .01; ** STAR–>Δ FSSB (W2–W1)–>Outcome STAR–>Δ STAR–>Δ Family-to-Work Conflict (W2–W1)–>Outcome Family-to-Work STAR–>Δ Burnout (W2–W1)–>Outcome STAR–>Δ (W2)–>Outcome Schedule No (W1) to Yes Variable STAR–>Δ Hours at Home (W2–W1)–>Outcome Work STAR–>Δ p < .05; Table 3 . (continued) Table

All models control for lagged dependent Note: Early Survey Group refers to respondents who did not know about the merger until after their baseline survey. and age cohort. Results for covariates are available upon request. Multilevel mediation analysis was done using the variables at baseline, manager status, gender, obtained indirect effects by bootstrapping the ml_mediation ml_mediation program in Stata, which was developed based on Krull and MacKinnon (2001). We command with 1,000 replications a seed value of 1. Coefficients under sections 1 and 2 are not directly comparable. Those section (Baron the mediator of burnout for job satisfaction (Panel B) as an Take Kenny's approach) denote how the outcome changes with a one-unit increase in mediator. example, –.077 means a one-unit increase in burnout decreases job satisfaction by .077. Coefficients under section 2 (Sobel's test with bootstrapped standard effect on the mediator and mediator's outcome, so signs could differ from corresponding errors) represent the product of two values: STAR's Again use the mediator of burnout for job satisfaction (Panel B) as an example, .027 has a affects the mediator. coefficients in section 1 depending on how STAR burnout, and because burnout decreases job satisfaction. The product of these two negative values therefore decreases the mediator, positive sign because STAR satisfaction relationship is .027. is positive, .027, which means the mediation effect of reduced burnout on STAR-job *

152 Moen et al. 153 most of its STAR effects mediated (Panel D). the social environment and well-being (Pearlin Baron and Kenny’s approach suggests a full 1999). Accordingly, we examine whether mediation by three mediators (schedule con- STAR effects differ among various subgroups trol, family-to-work conflict, and burnout). of respondents, again focusing on the Early Similarly, the modified Sobel’s tests show that Survey Group. To do so, we analyze the inter- these three indirect paths are all statistically action of STAR and four potential moderators: significant, each accounting for over 20 per- gender, gendered parental status, age-cohort, cent of the total STAR effect on psychological and managerial status. We find no statistically distress. Second, perceived stress is the only significant difference between STAR and the outcome with little mediation (Panel C), sug- usual practice groups in subjective well-being gesting STAR has a direct, rather than indirect, outcomes by age-cohort, suggesting that effect on employees’ stress. Third, changes in STAR promoted well-being across various schedule control (from baseline to six months) age groups of workers. are a strong mediator, accounting for 17 per- However, we find that STAR reduced psy- cent of the STAR effect in increasing job satis- chological distress more for women than for faction, 23 percent of the STAR effect in men. To facilitate understanding of this moder- decreasing psychological distress, and 19 per- ating effect, we present a figure showing models cent of the STAR effect in decreasing burnout with significant interaction terms. In Figure 1a, between baseline and Wave 3. Changes in women randomized to STAR have significantly family-to-work conflict also mediate the rela- lower levels of psychological distress by Wave 3 tionship, but not as strongly; 8 percent and 24 than women in the control group. No such dif- percent of the STAR effects on decreased ference is found for men (gender interaction, burnout and decreased psychological distress, p < .05). We also see a gender interaction for respectively, are mediated by reductions in perceived stress, although it is only marginally family-to-work conflict. We also considered significant ( p < .06, see Figure 1b). Note that burnout itself as a possible mediator, finding both men and women benefit from STAR in the that changes in burnout account for 13 percent two well-being measures most closely tied to of the STAR effect on increased job satisfac- work, with similar declines in burnout and tion and 23 percent of the STAR effect on increases in job satisfaction regardless of gen- decreased psychological distress. der. The gender component of Hypothesis 3 is Fourth, of the six mediators, FSSB and the thus only partially supported. two flexibility practice measures (schedule Figure 2a further examines gender differ- changes and working at home) seem to be the ences by presenting predicted psychological weakest; none of the mediation effects are distress at Wave 3 for four subgroups: women defined as statistically significant according to with children at home, women without children the modified Sobel test. Taken as a whole, at home, men with children at home, and men Hypothesis 2 is partially supported in that without children at home. We do not find sup- some (but not all) of the STAR effects on well- port for our hypothesis that workers engaged in being outcomes occur through STAR’s effects active parenting (especially mothers) will ben- in increasing workers’ schedule control and efit most from STAR; the STAR effect seems to reducing their levels of family-to-work con- be largest among women without children at flict and burnout. home ( p < .05). Figure 2b shows means of psychological distress at Wave 1 and Wave 3 for women with and without children at home Moderation Analysis of STAR Effects and by STAR. We see that mothers and women on Subjective Well-Being without children at home randomized to STAR The stress process theoretical model empha- reported comparable declines in psychological sizes the importance of status locations as distress over 12 months (the two solid lines). potential moderators of relationships between Women without children in the control group, 154 American Sociological Review 81(1)

Figure 1a. Predicted Psychological Distress at Wave 3 with 95% CIs by STAR and Gender, Early Survey Group

Figure 1b. Predicted Perceived Stress at Wave 3 by STAR and Gender, Early Survey Group by contrast, experienced increased psychologi- differences in effects of STAR by gender/active cal distress over this period (black dashed line), parenting status. which appears to be the main reason why we We hypothesized that managers would find a particularly large STAR effect for women benefit less than employees in terms of without children at home. In further analysis, improved subjective well-being, which is par- we find women without children at home were tially supported for job satisfaction. Figure 3 older (average age of 50 versus 44 to 48 for the shows that employees (but not managers) other groups), much more likely to be single randomized to STAR had higher levels of job (41 percent versus 3 to 23 percent), and more satisfaction than their peers in the usual prac- apt to have adult care responsibilities (31 per- tice groups ( p < .05). cent versus 20 to 25 percent). We speculate that the increased distress of women in the control group may reflect the challenges and vulnera- Discussion and bilities of older women in highly demanding Conclusions jobs with less control over their time, but this is This study contributes to the growing body of the case for only one well-being measure; theory and research on the social determi- analyses of the three other outcomes found no nants of health and well-being (Berkman Moen et al. 155

Figure 2a. Predicted Psychological Distress at Wave 3 by STAR, Gender, and Family Caregiving Context, Early Survey Group

Figure 2b. Means of Psychological Distress at Wave 1 and Wave 3 among Women, by STAR and Parental Status, Early Survey Group et al. 2014; House 2002; Tausig and Fenwick We advance stress process and job strain 2011) and specifically to stress process and theoretical models by moving beyond examin- job strain theories. Its primary contribution is ing workers in different occupations or circum- in putting these approaches in motion, exam- stances to demonstrate that changes in work ining the implications of changing the social conditions enhance subjective well-being. environment of work for corresponding Although the potential for change in social changes in well-being. Some social structures environments is an explicit component of stress consequential for health are seemingly intrac- process and job strain theories, most research table, such as one’s gender, race, age, and involves cross-sectional correlational studies parents’ socioeconomic status (Berkman et al. or, to a lesser extent, longitudinal studies of 2014; Tilly 1998). Others are institutional- events, such as job loss or caregiving, to exam- ized, taken-for-granted regimes—including ine potential declines in well-being. We dem- temporal regimes of policies, practices, rules, onstrate, with a rigorous randomized field trial, and routines regarding work time—that can that changes in work conditions arising from an be changed through deliberate efforts like the organizational intervention (providing workers STAR organizational intervention. with greater control over the time and timing of 156 American Sociological Review 81(1)

Figure 3. Predicted Job Satisfaction at Wave 3 with 95% CIs by STAR and Managerial Status, Early Survey Group their work, and encouraging supervisor support “clean” research designs such as field experi- for family and personal life) produce corre- ments. One major complexity of our study sponding salutary changes over a 12-month was the sudden announcement of a corporate period in four components of subjective well- merger in the midst of data collection. This being: burnout, job satisfaction, perceived event allowed us to examine how work condi- stress, and psychological distress. tions are affected by both deliberate organiza- Bronfenbrenner (2005) makes the claim tional initiatives and the challenging, that to really understand something, one changing social context. In retrospect, we should try to change it—and yet few scholars now see the IT world as a turbulent environ- have studied deliberate efforts to improve ment full of mergers, acquisitions, and con- subjective well-being by modifying the social stant change. Fortunately, given the study environments of work. By contrast, major timeline, we were able to parse out interven- strides have been made in removing or pro- tion effects on the Early Survey Group (who tecting workers against toxic physical envi- were not exposed to the merger shock before ronments, even though too many exposure to STAR) from the Late Survey disadvantaged workers still face hazards Group. today. There has been a real lag in recogniz- We found pronounced salutary effects of ing the toxicity of social environments, the intervention only for the Early Survey including the ways work is constrained by Group. One important implication of our rigid work-time rules and unsupportive findings is that workers (like those in the Late supervisors. Some employers offer individual- Survey Group) already faced with major level stress reduction techniques, suggesting organizational dislocations (e.g., mergers, methods to cope with difficulties rather than acquisitions, or dramatic rounds of downsiz- promoting well-being by improving the work ing) are unlikely to buy into and benefit from environment. This is understandable, given more positive organizational interventions. the time and financial costs of developing Other research has also found that workplace and testing a theoretically grounded interven- interventions have null effects (and, in some tion that is acceptable to organizational cases, negative impacts) on health and well- stakeholders while simultaneously substan- being when organizational restructuring and tial enough to make a difference in workers’ downsizing are also occurring (Egan et al. lives. 2007). There was no significant downsizing Our study does just that. It also confirms in the study period we report on here, but we there are inevitable complexities with even nevertheless find no STAR effects on Moen et al. 157 well-being among the Late Survey Group. clearly affected by work roles and resources The uncertainty of the upcoming merger off- as well as normative family caregiving set—and perhaps overwhelmed—any poten- responsibilities (Simon 2014), and the link tial benefits of STAR for workers facing the between work-family strains and well-being impending organizational changes. is clearer for mothers than for fathers (Noma- The positive effects of this organizational guchi, Milkie, and Bianchi 2005). However, intervention on the Early Survey Group raise the increase in psychological distress among additional questions relevant to stress process women without children in the control group and job strain theoretical models. In particular, suggests that vulnerability extends beyond how does STAR influence subjective well- simply caring for young children. Future being? We find evidence of three partial medi- research needs to consider the stresses of ators for different components of well-being: employed older women who are more apt to increases in respondents’ sense of schedule be single and caring for parents or other control, reductions in their family-to-work adults, in particular. Our findings suggest that conflict, and reductions in burnout. But much women without children at home may be dis- of STAR’s effects remain unexplained by tressed due to their high workloads (including these mediators, suggesting the need for paid work and caregiving) and lower likeli- greater attention to the “how” of organiza- hood of having a spouse to share home tasks, tional innovations like STAR and, indeed, for caregiving, and breadwinning, and that continued attention to the mechanisms under- organizational changes, like STAR, can lying inequality in health and well-being, as improve their well-being. well as in other resources (Reskin 2003). Mothers benefit from these organizational Another important question raised by both changes too. Mothers in STAR reported sig- stress process and job strain models is whether nificantly more time adequacy with their fam- this type of workplace intervention benefits ily (Kelly et al. 2014), and mothers in STAR particular sets of workers more than others. increased their time with adolescent children We theorized that workers potentially vulner- over a one-year period, whereas mothers in able to stress at work, at home, and due to the the control group decreased their time with juxtaposition of the two would be most apt to children (Davis et al. 2015). There is also see improvements in subjective well-being. evidence that STAR “crossed over” from par- We find no moderators of burnout; in other ents’ experiences to positively affect adoles- words, STAR reduced workers’ burnout cents’ well-being, as measured in greater regardless of their gender, their parenting positive affect and less negative reactivity to circumstances, their age-cohort, or whether daily stressors for children who had a parent they were managers or non-supervisory in STAR, compared to children whose parent employees. Similarly, STAR improved the was in the usual practice group (Lawson et al. job satisfaction of all these subgroups with forthcoming). Adolescents with a parent in only one exception: employees were more apt STAR also reported less variation in sleep than managers randomized to STAR to expe- duration from night to night, less time to fall rience an increase in job satisfaction. The asleep, and better sleep quality, although no broad effects of STAR across subgroups difference was found in sleep duration likely reflect that these outcomes are more (McHale et al. 2015). While research on explicitly tied to work than the more global mothers, fathers, and children is certainly measures of perceived stress and psychologi- important, it should be balanced by research cal distress. on the growing number of older women Our findings that STAR reduced psycho- workers to fully understand the nexus of logical distress and perceived stress more for work, family, and health. women is consistent with our expectation that What are the implications for studying more vulnerable workers would benefit more workers’ mental health and for devising from these changes. Women’s well-being is organizational policies and practices to reduce 158 American Sociological Review 81(1) risk factors? First, drawing on stress process well-being should move beyond IT workers and job strain approaches, we provide clear to consider the service sector, blue-collar evidence that specific conditions matter for jobs, and other white-collar work environ- well-being and can be changed. This neces- ments and how economic hardship may mod- sitates more than “helping” workers muddle erate any intervention effects. through their stressful lives by teaching them Fourth, issues of sustainability are impor- individual coping strategies. Instead, this tant. We were unable to test for sustainability study demonstrates the value of an organiza- given subsequent dislocations in the merging tional initiative promoting greater flexibility organization after the 12-month survey. This and control for workers as well as greater is a key topic for future research. supervisor support. Moreover, organizational Finally, this study demonstrates the value changes that lower workers’ burnout, per- of real-world, randomized field trial research ceived stress, and psychological distress bring designs for teasing out causal relationships benefits to employers as well, because such and processes, despite the costs and com- outcomes are apt to reduce productivity and plexities they invariably engender. If rand- increase absenteeism, presenteeism (i.e., omized field trials are not feasible, natural showing up, but not being engaged at work), experimental designs studying the effects of turnover, health problems, and poor life qual- policy, economic, or other changes happening ity at work and at home. “on the ground” but not deliberately imple- Second, we show that organizational inter- mented by researchers are another way of ventions can promote facets of subjective capturing and studying change. well-being partly because they increase con- Social scientists are now equipped with the trol and reduce family-to-work conflict and methodological tools, and more often able to burnout, but there are also other unidentified collect the necessary longitudinal data, to bet- mechanisms that need to be investigated. ter inform both theory and policy decision- Moreover, different mechanisms may operate making related to the social determinants of for people in working-class or other nonpro- physical and mental health as well as other fessional or less technical jobs. For example, resources. Scholars should begin to recognize employees represented by unions may appre- and strive to improve the quality of interven- ciate this type of intervention (if pursued with tion and organizational research to showcase the union’s support) because it increases their the structural forces and changes that can voice outside of the formal bargaining con- impede or promote workers’ well-being. text and legitimates addressing work-life con- cerns as well as wages and benefits in negotiations and on the job (Berg et al. 2014). Acknowledgments Third, interventions targeting all workers, Special acknowledgments go to Extramural Staff Science Collaborator, Rosalind Berkowitz King, PhD and Lynne not just those most at risk, seem to have broad Casper, PhD for design of the original Work, Family, impacts and reach beyond the most vulnera- Health and Well-Being Research Initiative. Our thanks to ble subgroups of workers. But IT workers are the TOMO managers and employees who participated in all less “at risk” than workers with less educa- the study and facilitated our research; to Rachel Magen- tion, skills, and professional status; future nis, Kimberly Fox, Laurie Pasricha, Jonathon Vaughn, and Holly Whitesides for facilitating data collection and studies need to test similar interventions and conducting field research; to Jane Peterson for editorial specify mechanisms in different workforce assistance; to Michael Oakes and other members of the populations. A real limitation, though a delib- Work, Family and Health Network for research design erate one, of this study is the focus on IT decisions and helpful comments; to the Minnesota Popu- workers who have demanding, stressful jobs lation Center (NICHD R24HD041023) for research sup- port; to GoRowe.com for collaboration on the intervention; but who are not simultaneously income- to the ASR reviewers and editors; and to audiences at the strapped. Researchers seeking to develop and University of Minnesota, University of Maryland, UCLA, test interventions advancing subjective the American Sociological Association, the Work and Moen et al. 159

Family Researchers Network, and the Population Associ- in the Early Survey Group in age, cohort member- ation of America meetings for their comments. Apprecia- ship, gender, managerial status, baseline well-being tion also to the Center for Advanced Study in the outcomes, or baseline job demands (psychological Behavioral Sciences (CASBS) at Stanford University for job demands and work hours). Given these similari- providing the first author valuable writing time. This ties and given our rationale in distinguishing early manuscript does not represent an official position of the versus late respondents (i.e., whether their baseline NICHD (NIH, USDHHS), nor does it present a prescrip- well-being measures were contaminated by the tive policy statement from the NICHD on the conduct of merger announcement), we report the simpler com- research. parison of early compared to later respondents. 4. Workers were eligible to participate in the study if they were employees (not contractors) or manag- Funding ers located in the two cities where data collection This research was conducted as part of the Work, Family occurred. One study group, whose employees were and Health Network (http://www.WorkFamilyHealthNet- represented by collective bargaining agreements, work.org), which is funded by a cooperative agreement was excluded because of concerns that the inter- through the National Institutes of Health and the Centers vention might conflict with some contractual work for Disease Control and Prevention: Eunice Kennedy rules; union leadership was consulted on the deci- Shriver National Institute of Child Health and Human sion and suggested that course of action. Development (Grant # U01HD051217, U01HD051218, 5. We excluded 15 employees who were randomized U01HD051256, U01HD051276), National Institute on to the STAR intervention but were not invited to Aging (Grant # U01AG027669), Office of Behavioral participate in STAR because of a staff error. We and Social Sciences Research, and National Institute for also excluded eight employees initially randomized Occupational Safety and Health (Grant # U01OH008788, to the control group, because they were shifted to U01HD059773). Grants from the University of Minne- new teams already going through STAR. sota’s College of Liberal Arts, McKnight Foundation, 6. If a respondent skipped a specific item but com- William T. Grant Foundation, Alfred P. Sloan Founda- pleted at least 75 percent of the scale (e.g., 3 out of tion, and the Administration for Children and Families 4 items for perceived stress), we assigned the mean provided additional funding. The contents of this publi- from that respondent’s responses to other questions cation are solely the responsibility of the authors and do in that scale. If respondents did not complete at least not necessarily represent the official views of these 75 percent of the scale, they were omitted from that institutes and offices. model.

Notes References 1. We found one study in which nursing units were Aneshensel, Carol S., Leonard I. Pearlin, Joseph T. Mul- randomly assigned to self-managed scheduling or lan, Steven H. Zarit, and Carol J. Whitlatch. 1995. the control condition (Pryce, Albertsen, and Nielsen Profiles in Caregiving: The Unexpected Career. New 2006); this design is closest to our group-random- York: Academic Press. ized trial. Pryce and colleagues (2006) found no Aneshensel, Carol S., Carolyn M. Rutter, and Peter A. significant effects of self-scheduling on nurses’ Lachenbruch. 1991. “Social Structure, Stress, and health and well-being, although the intervention Mental Health: Competing Conceptual and Analytic improved work-life balance and job satisfaction. Models.” American Sociological Review 56(2):166– 2. A corporate merger is always highly confidential 78. because it significantly affects a firm’s stock price. Angrave, David, and Andy Charlwood. 2015. “What Our contacts in TOMO had a sense the company Is the Relationship between Long Working Hours, was not doing well, but neither HR managers, man- Over-employment, Under-employment and the Sub- agers, nor employees had specific clues about a jective Well-being of Workers?” Human Relations forthcoming merger announcement. 68(9):1491–1515. 3. Within the Early Survey Group, 76 percent of the Baron, Reuben M., and David A. Kenny. 1986. “The STAR sample had completed the STAR training Moderator-Mediator Variable Distinction in Social before the merger was announced. The remainder Psychological Research: Conceptual, Strategic, and of the Early Survey Group randomized to STAR Statistical Considerations.” Journal of Personality had completed the baseline survey and had begun and Social Psychology 51(6):1173–82. the STAR training sessions, or their upper manage- Begg, Colin, Mildred Cho, Susan Eastwood, Richard ment had introduced STAR, before the merger was Horton, David Moher, Ingram Olkin, Roy Pitkin, announced. In supplemental analyses, we found that Drummond Rennie, Kenneth F. Schulz, David Simel, this latter group—who learned of the merger during and Donna F. Stroup. 1996. “Improving the Quality the STAR training period, rather than after it was of Reporting of Randomized Controlled Trials: The completed—did not differ significantly from others CONSORT Statement.” JAMA 276(8):637–39. 160 American Sociological Review 81(1)

Berg, Peter, Ellen Ernst Kossek, Kaumudi Misra, and and Reciprocal Relationships in a 4-Wave Study.” Dale Belman. 2014. “Do Unions Matter for Work- Work & Stress 18(2):149–66. Life Flexibility Policy Access and Use?” Industrial D’Souza, Rennie M., Lyndall Strazdins, Lynette L. Lim, and Labor Relations Review 67(1):111–36. Dorothy H. Broom, and Bryan Rodgers. 2003. “Work Berkman, Lisa F., Ichiro Kawachi, and Maria Glymour. and Health in a Contemporary Society: Demands, 2014. Social Epidemiology, 2nd ed. New York: Control, and Insecurity.” Journal of Epidemiological Oxford University Press. and Community Health 57(11):849–954. Bianchi, Suzanne M., Lynne M. Casper, and Rosalind B. Egan, Matt, Clare Bambra, Sian Thomas, Mark Petticrew, King. 2005. Work, Family, Health, and Well-Being. Margaret Whitehead, and Hilary Thomson. 2007. Mahwah, NJ: Lawrence Erlbaum Associates. “The Psychosocial and Health Effects of Workplace Bianchi, Suzanne M., and Melissa A. Milkie. 2010. Reorganisation. 1. A Systematic Review of Organ- “Work and Family Research in the First Decade of isational-Level Interventions That Aim to Increase the 21st Century.” Journal of Marriage and Family Employee Control.” Journal of Epidemiology and 72(3):705–725. Community Health 61(11):945–54. Bosma, Hans, Stephen A. Stansfeld, and Michael G. Mar- Elder Jr., Glen H. 1998. “The Life Course as Develop- mot. 1998. “Job Control, Personal Characteristics, mental Theory.” Child Development 69(1):1–12. and Heart Disease.” Journal of Occupational Health Elder Jr., Glen H., Linda K. George, and Michael J. Psychology 3(4):402–409. Shanahan. 1996. “Psychosocial Stress over the Life Bray, Jeremy W., Erin L. Kelly, Leslie B. Hammer, David Course.” Pp. 247–91 in Psychosocial Stress: Per- M. Almeida, James W. Dearing, and Rosalind B. King. spectives on Structure, Theory, Life Course and Meth- 2013. “An Integrative, Multilevel, and Transdisciplinary ods, edited by H. B. Kaplan. Orlando, FL: Academic Research Approach to Challenges of Work, Family, and Press. Health.” RTI Press publication No. MR-0024-1302. Fenwick, Rudy, and Mark Tausig. 2007. “A Political Research Triangle Park, NC: RTI Press. Economy of Stress: Recontextualizing the Study of Bronfenbrenner, Urie. 2005. Making Human Beings Mental Health/Illness in Sociology.” Pp. 143–67 in Human: Bioecological Perspectives on Human Mental Health, Social Mirror, edited by W. Avison, Development. Thousand Oaks, CA: Sage. J. D. McLeod, and B. A. Pescosolido. New York: Cahill, Kate, and Tim Lancaster. 2014. “Workplace Inter- Springer. ventions for Smoking Cessation.” Cochrane Data- Frane, James W. 1998. “A Method of Biased Coin Ran- base of Systematic Revision, Issue 2: CD 003440. domization, Its Implementation, and Its Validation.” Retrieved May 18, 2015 (http://www.cochranelibrary Drug Information Journal 32(2):423–32. .com). Friedman, Lawrence M., Curt D. Furberg, and David L. Cammann, Cortlandt, Mark Fichman, G. Douglas Jen- DeMets. 2010. Fundamentals of Clinical Trials. New kins, and John Klesh. 1983. “Michigan Organiza- York: Springer. tional Assessment Questionnaire.” Pp. 71–138 in Galinsky, Ellen, Kerstin Aumann, and James T. Bond. Assessing Organizational Change: A Guide to Meth- 2009. Times Are Changing: Gender and Generation ods, Measures, and Practices, edited by S. E. Sea- at Work and at Home. New York: Families and Work shore, E. E. Lawler, P. H. Mirvis, and C. Cammann. Institute. New York: Wiley-Interscience. Goode, William J. 1960. “A Theory of Role Strain.” Clawson, Dan, and Naomi Gerstel. 2014. Unequal Time: American Sociological Review 25(4):483–96. Gender, Class, and Family in Employment Schedules. Gotlib, Ian H., and Blair Wheaton. 1997. Stress and Thousand Oaks, CA: Russell Sage Foundation. Adversity over the Life Course: Trajectories and Cohen, Jacob. 1988. Statistical Power Analysis for the Turning Points. New York: Cambridge University Behavioral Sciences, 2nd ed. Hillsdale, NJ: Lawrence Press. Erlbaum Associates. Grzywacz, Joseph G., Dawn S. Carlson, and Sandee Cohen, Sheldon, Tom Kamarck, and Robin Mermelstein. Shulkin. 2008. “Schedule Flexibility and Stress: 1983. “A Global Measure of Perceived Stress.” Jour- Linking Formal Flexible Arrangements and Per- nal of Health and Social Behavior 24(4):385–96. ceived Flexibility to Employee Health.” Community, Davis, Kelly D., Katie M. Lawson, David M. Almeida, Work & Family 11(2):199–214. Erin L. Kelly, Rosalind B. King, Leslie B. Hammer, Halbesleben, Jonathon R. B., and M. Ronald Buckley. Lynne M. Casper, Cassandra A. Okechukwu, Ginger 2004. “Burnout in Organizational Life.” Journal of C. Hanson, and Susan M. McHale. 2015. “Parents’ Management 30(6):859–79. Daily Time with Their Children: A Workplace Inter- Hammer, Leslie B., Ellen E. Kossek, Todd E. Bodner, vention.” Pediatrics 135(5):875–82. W. Kent Anger, and Kristi L. Zimmerman. 2011. de Lange, Annet H., Toon W. Taris, Michiel A. J. Kom- “Clarifying Work–Family Intervention Processes: pier, Irene L. D. Houtman, and Paulien M. Bongers. The Roles of Work–Family Conflict and Family- 2004. “The Relationships between Work Characteris- Supportive Supervisor Behaviors.” Journal of tics and Mental Health: Examining Normal, Reversed Applied Psychology 96(1):134–50. Moen et al. 161

Hammer, Leslie B., Ellen E. Kossek, Todd E. Bodner, Evidence from the Work, Family, and Health Net- and Tori L. Crain. 2013. “Measurement Development work.” American Sociological Review 79(3):485–516. and Validation of the Family Supportive Supervisor Kelly, Erin L., Phyllis Moen, and Eric P. Tranby. 2011. Behavior Short-Form (FSSB-SF).” Journal of Occu- “Changing Workplaces to Reduce Work-Family Con- pational Health Psychology 18(3):285–96. flict Schedule Control in a White-Collar Organiza- Hannan, Peter J. 2006. “Experimental Social Epidemi- tion.” American Sociological Review 76(2):265–90. ology: Controlled Community Trials.” Pp. 335–64 Kessler, Robert C., Catherine Barber, Arne Beck, Patri- in Methods in Social Epidemiology, edited by J. M. cia Berglund, Paul D. Cleary, David McKenas, Nico Oakes and J. S. Kaufman. San Francisco: Jossey- Pronk, Gregory Simon, Paul Stang, T. Bedirhan Bass/Wiley. Üstün, and Phillip Wang. 2003. “The World Health Hartfiel, Ned, Jon Havenhand, Sat Bir Khalsa, Graham Organization Health and Work Performance Ques- Clarke, and Anne Krayer. 2011. “The Effectiveness tionnaire (HPQ).” Journal of Occupational and Envi- of Yoga for the Improvement of Well-Being and ronmental Medicine 45(2):156–74. Resilience to Stress in the Workplace.” Scandinavian Keyes, Corey L. M. 1998. “Social Well-Being.” Social Journal of Work, Environment & Health 37(1):70–76. Psychology Quarterly 61(2):121–40. Hill, Rueben. 1949. Families under Stress: Adjustment King, Rosalind B., Georgia Karuntzos, Lynne M. Casper, to the Crises of War Separation and Reunion. New Phyllis Moen, Kelly D. Davis, Lisa F. Berkman, Mary York: Harper & Brothers. Durham, and Ellen E. Kossek. 2012. “Work-Family Hochschild, Arlie R. 1989. The Second Shift: Working Balance Issues and Work-Leave Policies.” Pp. 323– Families and the Revolution at Home. New York: 39 in Handbook of Occupational Health and Well- Avon Books. ness, edited by R. J. Gatchel and I. Z. Schultz. New House, James S. 2002. “Understanding Social Factors York: Springer. and Inequalities in Health: 20th Century Progress Kleiner, Sibyl, and Eliza K. Pavalko. 2010. “Clocking and 21st Century Prospects.” Journal of Health and In: The Organization of Work Time and Health in the Social Behavior 43(2):125–42. United States.” Social Forces 88(3):1463–86. Jacobs, Jerry A., and Kathleen Gerson. 2004. The Time Kohn, Melvin L., and Carmi Schooler. 1982. “Job Condi- Divide: Work, Family, and Gender Inequality. Cam- tions and Personality: A Longitudinal Assessment of bridge, MA: Harvard University Press. Their Reciprocal Effects.” American Journal of Soci- Johnson, Jeffrey V., and Ellen M. Hall. 1988. “Job Strain, ology 87(6):1257–86. Work Place Social Support, and Cardiovascular Dis- Kossek, Ellen E., Leslie B. Hammer, Erin L. Kelly, and ease: A Cross-Sectional Study of a Random Sample Phyllis Moen. 2014. “Designing Work, Family & of the Swedish Working Population.” American Jour- Health Organizational Change Initiatives.” Organiza- nal of Public Health 78(10):1336–42. tional Dynamics 43(1):53–63. Joyce, Kerry, Roman Pabayo, Julia A. Critchley, and Kossek, Ellen E., Shaun Pichler, Todd E. Bodner, and Clare Bambra. 2012. “Flexible Working Conditions Leslie B. Hammer. 2011. “Workplace Social Sup- and their Effects on Employee Health and Wellbe- port and Work-Family Conflict: A Meta-Analysis ing.” Cochrane Database of Systematic Reviews, Clarifying the Influence of General and Work-Family Issue 2, Art. No.: CD008009. Retrieved May 18, 2015 Specific Supervisor and Organizational Support.” (http://www.cochranelibrary.com). Personnel Psychology 64(2):289–313. Kabat-Zinn, Jon. 2003. “Mindfulness-Based Interven- Kristensen, Tage S. 2000. “Workplace Intervention Stud- tions in Context: Past, Present, and Future.” Clinical ies.” Occupational Medicine 15:293–306. Psychology: Science and Practice 10(2):144–56. Krull, Jennifer L., and David P. MacKinnon. 2001. Kaplan, Howard B. 1983. Psychosocial Stress: Trends in “Multilevel Modeling of Individual and Group Level Theory and Research. New York: Academic Press. Mediated Effects.” Multivariate Behavioral Research Karasek, Robert A., Jr. 1979. “Job Demands, Job Deci- 36(2):249–77. sion Latitude, and Mental Strain: Implications for Lam, Jack, Kimberly E. Fox, Wen Fan, Phyllis Moen, Job Redesign.” Administrative Science Quarterly Erin L. Kelly, Leslie B. Hammer, and Ellen E. Kossek. 24(2):285–308. 2015. “Manager Characteristics and Employee Job Karasek, Robert A., Jr., and Töres Theorell. 1990. Healthy Insecurity around a Merger Announcement: The Role Work: Stress, Productivity, and the Reconstruction of of Status and Crossover.” Sociological Quarterly Working Life. New York: Basic Books. 56(3):558–80. Karasek, Robert A., Jr., and Töres Theorell. 2000. “The Lawson, Katie M., Susan M. McCale, Kelly D. Davis, Demand-Control-Support Model and CVD.” Occupa- David M. Almeida, Erin L. Kelly, and Rosalind B. tional Medicine: State of the Art Reviews 15(1):78–83. King. Forthcoming. “Effects of Workplace Interven- Kelly, Erin L., Phyllis Moen, J. Michael Oakes, Wen Fan, tion on Affective Well-Being in Employees’ Chil- Cassandra A. Okechukwu, Kelly D. Davis, Leslie B. dren.” Developmental Psychology. Hammer, Ellen E. Kossek, Rosalind B. King, Ginger Lazarus, Richard S., and Susan Folkman. 1984. Stress, C. Hanson, Frank J. Mierzwa, and Lynne M. Casper. Appraisal and Coping. New York: Springer Publish- 2014. “Changing Work and Work-Family Conflict: ing Company. 162 American Sociological Review 81(1)

Link, Bruce G., and Jo C. Phelan. 1995. “Social Condi- Mirowsky, John, and Catherine E. Ross. 2003b. Social tions as Fundamental Causes of Disease.” Journal of Causes of Psychological Distress, 2nd ed. New Health and Social Behavior 35(extra issue):80–94. Brunswick, NJ: Aldine Transaction. Link, Bruce G., Jo C. Phelan, Richard A. Miech, and Moen, Phyllis. 2003. It’s about Time: Couples and Emily Leckman Westin. 2008. “The Resources That Careers. New York: Cornell University Press. Matter: Fundamental Social Causes of Health Dis- Moen, Phyllis, Erin L. Kelly, and Rachelle Hill. 2011. parities and the Challenge of Intelligence.” Journal of “Does Enhancing Work-Time Control and Flexibility Health and Social Behavior 49(1):72–91. Reduce Turnover? A Naturally-Occurring Experi- Lyness, Karen S., Janet C. Gornick, Pamela Stone, and ment.” Social Problems 58(1):69–98. Angela R. Grotto. 2012. “It’s All about Control: Worker Moen, Phyllis, Erin L. Kelly, and Qinlei Huang. 2008. Control over Schedule and Hours in Cross-National Con- “Work, Family and Life-Course Fit: Does Control text.” American Sociological Review 77(6):1023–49. Over Work Time Matter?” Journal of Vocational MacKinnon, David P., Ghulam Warsi, and James H. Behavior 73(3):414–25. Dwyer. 1995. “A Simulation Study of Mediated Moen, Phyllis, Erin L. Kelly, and Jack Lam. 2013. Effect Measures.” Multivariate Behavioral Research “Healthy Work Revisited: Do Changes in Time Strain 30(1):41–62. Predict Well-being?” Journal of Occupational Health Major, Virginia S., Katherine J. Klein, and Mark G. Psychology 18(2):157–72. Ehrhart. 2002. “Work Time, Work Interference with Moen, Phyllis, Erin L. Kelly, Eric P. Tranby, and Qinlei Family, and Psychological Distress.” Journal of Huang. 2011. “Changing Work, Changing Health: Applied Psychology 87(3):427–36. Can Real Work-Time Flexibility Promote Health Marklund, Staffan, Malin Bolin, and Jan von Essen. 2008. Behaviors and Well-Being?” Journal of Health and “Can Individual Health Differences Be Explained by Social Behavior 52(4):404–429. Workplace Characteristics? A Multilevel Analysis.” Moen, Phyllis, and Patricia V. Roehling. 2005. The Social Science & Medicine 66(3):650–62. Career Mystique: Cracks in the American Dream. Marmot, Michael G., Hans Bosma, Harry Hemingway, Boulder, CO: Rowman & Littlefield. Eric Brunner, and Stephen A. Stansfeld. 1997. “Con- Morgan, Stephen L., and Christopher Winship. 2007. tribution of Job Control and Other Risk Factors to Counterfactuals and Causal Inference: Methods and Social Variations in Coronary Heart Disease Inci- Principles for Social Research. New York: Cam- dence.” The Lancet 350(9073):235–39. bridge University Press. Marmot, Michael G., Stephen A. Stansfeld, Chandra Netemeyer, Richard G., James S. Boles, and Robert Patel, Fiona North, Jenny Head, Ian White, Eric McMurrian. 1996. “Development and Validation of Brunner, Amanda Feeney, and G. Davey Smith. Work-Family Conflict and Family-Work Conflict 1991. “Health Inequalities among British Civil Scales.” Journal of Applied Psychology 81(4):400– Servants: The Whitehall II Study.” The Lancet 410. 337(8754):1387–93. Nijp, Hylco H., Debby G. J. Beckers, Sabine A. E. Geurts, Maslach, Christina, and Susan E. Jackson. 1986. Maslach Philip Tucker, and Michiel A. J. Kompier. 2012. Burnout Inventory Manual, 2nd ed. Palo Alto, CA: “Systematic Review on the Association between Consulting Psychologists Press. Employee Worktime Control and Work-Non-work Maslach, Christina, Wilmar B. Schaufeli, and Michael P. Balance, Health and Well-Being, and Job-Related Leiter. 2001. “Job Burnout.” Annual Review of Psy- Outcomes.” Scandinavian Journal of Work and Envi- chology 52(1):397–422. ronmental Health 38(4):299–313. Mausner-Dorsch, Hilde, and William W. Eaton. 2000. Nomaguchi, Kei M., Melissa A. Milkie, and Suzanne “Psychosocial Work Environment and Depres- M. Bianchi. 2005. “Time Strains and Psychological sion: Epidemiologic Assessment of the Demand- Well-Being: Do Dual-Earner Mothers and Fathers Control Model.” American Journal of Public Health Differ?” Journal of Family Issues 26(6):756–92. 90(11):1765–70. Oakes, J. Michael, and Jay S. Kaufman. 2006. Methods in McHale, Susan M., Katie M. Lawson, Kelly D. Davis, Social Epidemiology. San Francisco: Jossey-Bass/Wiley. Lynne M. Casper, Erin L. Kelly, and Orfeu M. Bux- Pearlin, Leonard I. 1989. “The Sociological Study of ton. 2015. “Effects of a Workplace Intervention on Stress.” Journal of Health and Social Behavior Sleep in Employees’ Children.” Journal of Adoles- 30(3):241–56. cent Health 56(6):672–77. Pearlin, Leonard I. 1999. “The Stress Process Revisited: Melamed, Samuel, Arie Shirom, Sharon Toker, Shlomo Reflections on Concepts and their Interrelationships.” Berliner, and Itzhak Shapira. 2006. “Burnout and Pp. 395–415 in The Handbook of the Sociology of Risk of Cardiovascular Disease: Evidence, Possible Mental Health, edited by C. S. Aneshensel and J. C. Causal Paths, and Promising Research Directions.” Phelan. New York: Kluwer. Psychological Bulletin 132(3):327–53. Pearlin, Leonard I., Elizabeth G. Menaghan, Morton A. Mirowsky, John, and Catherine E. Ross. 2003a. Educa- Lieberman, and Joseph T. Mullan. 1981. “The Stress tion, Social Status and Health. New Brunswick, NJ: Process.” Journal of Health and Social Behavior Aldine Transaction. 22(4):337–56. Moen et al. 163

Pearlin, Leonard I., Scott Schieman, Elena M. Fazio, and 21–51 in The Sociology of Mental Health: Selected Stephen C. Meersman. 2005. “Stress, Health, and the Topics from Forty Years, edited by R. J. Johnson, R. Life Course: Some Conceptual Perspectives.” Jour- J. Turner, and B. G. Link. New York: Springer Books. nal of Health and Social Behavior 46(2):205–219. Sobel, Michale E. 1982. “Asymptotic Confidence Inter- Phelan, Jo C., Bruce G. Link, Ana Diez-Roux, Ichiro vals for Indirect Effects in Structural Equation Mod- Kawachi, and Bruce Levin. 2004. “'Fundamental els.” Sociological Methodology 13:290–312. Causes of Social Inequalities in Mortality: A Test of Stansfeld, Stephen A., and Bridget Candy. 2006. “Psy- the Theory.” Journal of Health and Social Behavior chosocial Work Environment and Mental Health: A 45(3):265–85. Meta-Analytic Review.” Scandinavian Journal of Phelan, Jo C., Bruce G. Link, and Parusa Tehranifar. Work and Environmental Health 32(6):443–62. 2010. “Social Conditions as Fundamental Causes of Stansfeld, Stephen A., Rebecca Fuhrer, Martin J. Shipley, Health Inequalities: Theory, Evidence, and Policy and Michael G. Marmot. 2002. “Psychological Dis- Implications.” Journal of Health and Social Behavior tress as a Risk Factor for Coronary Heart Disease in 51(S):S28–S40. the Whitehall II Study.” International Journal of Epi- Pryce, Joanna, Karen Albertsen, and Karina Nielsen. demiology 31(1):248–55. 2006. “Evaluation of an Open-Rota System in a Dan- Tausig, Mark, and Rudy Fenwick. 2011. Work and Men- ish Psychiatric Hospital: A Mechanism for Improving tal Health in Social Context. New York: Springer. Job Satisfaction and Work–Life Balance.” Journal of Thomas, Linda Thiede, and Daniel C. Ganster. 1995. Nursing Management 14(4):282–88. “Impact of Family-Supportive Work Variables on Reskin, Barbara F. 2003. “Including Mechanisms in our Work-Family Conflict and Strain: A Control Perspec- Ascriptive Models of Inequality.” American Socio- tive.” Journal of Applied Psychology 80(1):6–15. logical Review 68(1):1–21. Tilly, Charles. 1998. Durable Inequalities. Berkeley, CA: Reynolds, Jeremy, and Lydia Aletraris. 2006. “Pursuing University of California Press. Preferences: The Creation and Resolution of Work Turner, R. Jay, Blair Wheaton, and Donald A. Lloyd. Hour Mismatches.” American Sociological Review 1995. “The Epidemiology of Social Stress.” Ameri- 71(4):618–38. can Sociological Review 60(1):104–125. Ross, Catherine E., and John Mirowsky. 1992. “House- Van Der Doef, Margot, and Stan Maes. 1999. “The Job holds, Employment, and the Sense of Control.” Demand-Control(-Support) Model and Psychologi- Social Psychology Quarterly 55(3):217–35. cal Well-Being: A Review of 20 Years of Empirical Ross, Catherine E., and John Mirowsky. 2003. “Social Research.” Work & Stress 13(2):87–114. Structure and Psychological Functioning: The Sense Wheaton, Blair. 2001. “The Role of Sociology in the of Personal Control, Distress, and Trust.” Pp. 411–47 Study of Mental Health . . . and the Role of Mental in The Handbook of Social Psychology, edited by J. Health in the Study of Sociology.” Journal of Health DeLamater and A. Ward. New York: Kluwer-Plenum. and Social Behavior 42(3):221–34. Ryff, Carol D., and Corey L. M. Keyes. 1995. “The Struc- Wheaton, Blair, and Philippa J. Clarke. 2003. “Space ture of Psychological Well-Being Revisited.” Journal Meets Time: Integrating Temporal and Contextual of Personality and Social Psychology 69(4):719–27. Influences on Mental Health in Early Adulthood.” Schaufeli, Wilmar B., and Arnold B. Bakker. 2004. “Job American Sociological Review 68(5):680–706. Demands, Job Resources, and Their Relationship with Wickrama, Kandauda A. S., Frederick O. Lorenz, Rand Burnout and Engagement: A Multi-Sample Study.” D. Conger, Lisa Matthews, and Glen H. Elder Jr. Journal of Organizational Behavior 25(3):293–315. 1997. “Linking Occupational Conditions to Physi- Schieman, Scott, Melissa A. Milkie, and Paul Glavin. cal Health through Marital, Social, and Intrapersonal 2009. “When Work Interferes with Life: Work-Non- Processes.” Journal of Health and Social Behavior work Interference and the Influence of Work-Related 38(4):363–75. Demands and Resources.” American Sociological Review 74(6):966–88. Schieman, Scott, Yuko Kurashina Whitestone, and Karen Phyllis Moen is Professor of Sociology and McKnight Van Gundy. 2006. “The Nature of Work and the Stress Presidential Chair at the University of Minnesota, fol- of Higher Status.” Journal of Health and Social lowing 25 years at Cornell University. She studies the Behavior 47(3):242–57. dynamic intersections of work, time, health, family, pol- Schoeni, Robert F., James S. House, George A. Kaplan, icy, and gender. Her book Encore Adulthood: Boomers and Harold Pollack. 2008. Making Americans Health- on the Edge of Risk, Renewal and Purpose is forthcom- ier: Social and Economic Policy as Health Policy. ing (Oxford University Press). New York: Russell Sage Foundation. Scott-Marshall, Heather. 2010. “The Social Patterning Erin L. Kelly is Professor of Work and Organization of Work-Related Insecurity and its Health Conse- Studies at the MIT Sloan School of Management and an quences.” Social Indicators Research 96(2):313–37. affiliate of the Institute for Work and Employment Simon, Robin. 2014. “Twenty Years of the Sociology of Research. Her research examines changing workplace Mental Health: The Continued Significance of Gender policies and employment relations and consequences for and Marital Status for Emotional Well-Being. “ Pp. U.S. workers, families, and organizations. 164 American Sociological Review 81(1)

Wen Fan is Assistant Professor of Sociology at Boston daily stress on health. He is the principal investigator of College. Engaged with questions of how health inequal- the National Study of Daily Experiences. ity is produced and perpetuated, her current research examines how demographic processes shape health dis- Ellen Ernst Kossek is the Basil S. Turner Professor at parities, and how individuals’ and couples’ health behav- Purdue University’s Krannert School of Management and iors are facilitated or constrained by the work and family research director of the Butler Center for Leadership context. Excellence. She was the first elected president of the Work-Family Researchers Network, and works globally to Shi-Rong Lee is a PhD candidate in the Sociology advance knowledge and practice on gender and diversity. Department at the University of Minnesota. Her research interests include finance, work and organizations, health, Orfeu M. Buxton is an Associate Professor and directs and life course studies. the Sleep, Health & Society program in the Department of Biobehavioral Health at The Pennsylvania State Uni- David Almeida is Professor of Human Development and versity. Secondary appointments include Harvard Uni- Family Studies at the Center for Healthy Aging at The versity and Brigham and Women’s Hospital, Boston. His Pennsylvania State University. His research focuses on research focuses on the causes and consequences of sleep the effects of biological and self-reported indicators of deficiency in the workplace, home, and society.