Work-family conflict is a common and widely studied stressor that has been negatively associated with employee health and well-being.^14–20 Work-family initiatives are defined as deliberate organizational changes in policies, practices, or culture to reduce work-family conflict.¹¹ Although they can be used to promote wellness and productivity, these programs can have mixed consequences.¹² Some U.S. public policies (e.g., the Family Medical Leave Act) promote basic work-life initiatives, but the majority of initiatives in the United States are firm-driven. Examples of these initiatives include telecommuting, flexible work arrangements, co-working, and work schedule and/or workload redesigns.

A comprehensive review conducted by Kelly and colleagues¹¹ highlights the gaps in the scientific literature examining the effects of work-family initiatives on work-family conflict and performance. The direct effects of work-family initiatives on work-family conflict and performance are mixed. Recent research has instead focused on the effects of work-family initiatives on employee attitudes and behaviors, such as perceived management support, perceived work-life culture, perceived flexibility, schedule control, and burnout as an antecedent to work-family conflict. Kelly and colleagues¹¹ note that much evidence in support of work-family initiatives affecting these attitudes and behaviors is positive but mainly correlational and cross-sectional. Similarly, studies that have associated work-family conflict and employee performance have also been cross-sectional or observational studies.^{21, 22} The current analysis uses a longitudinal, group-randomized field experiment, the WFHN study,²³ to rigorously examine the effects of a work-family initiative on employee performance. The WFHN intervention reduced work-family conflict¹³ and had a positive ROI²⁴ at the IT firm where it took place.

In this study, employee performance is measured across three domains: absenteeism, presenteeism, and total hours worked. A recent report found that access to work-family supports, supervisor supports, and flexible working arrangements are related to decreased absenteeism, increased engagement, and increased hours worked.²⁵ Absenteeism refers to a worker missing work or not fulfilling expected work responsibilities due to illness or other non-work commitments. Stressful family relationships^{20, 26} and the need for eldercare²⁷ are common reasons for greater absenteeism. Having supportive management and flexible child care arrangements has been found to decrease absenteeism.²⁸

Presenteeism refers to the extent to which an employee is at work but disengaged with work responsibilities. A worker may be physically at work, but may be ill, distracted, or unmotivated and not performing at an expected level. Studies have shown a positive relationship between work-family conflict and presenteeism.²⁹

The relationship between total hours worked and performance can be somewhat distorted. Workers with flexible schedules to accommodate family care needs may work fewer hours on average but may work harder during those hours, increasing performance. Total hours worked may not be a precise measure of performance in itself but is directly related to wages for non-salaried employees and may have other implications for salaried workers.

STAR is expected to increase employee perceptions of control and support, thereby reducing work-family conflict and improving employee performance in the long-term.³⁰ In the short-term, the implementation of an intervention in an active workplace setting could be burdensome to workers and may consequently offset employee performance gains from reduced work-family conflict. Likewise, it could take time for the work-family conflict reductions to impact performance. Consequently, we expect a priori that performance will be relatively unchanged for 6 months immediately following the implementation of STAR. At 12 and 18 months, we expect a priori that the intervention will have a positive effect on performance. Although negative effects of the intervention are possible, given the nature of work-family conflict and its specific impacts on performance as demonstrated in previous literature, we hypothesized a positive effect of the intervention on performance when the study was initially designed.

STAR is designed to increase employees’ control over their work time and work schedule, to increase supervisors’ support for this change, and to increase employees’ and supervisors’ focus on results.^{25, 31} STAR involved participatory activities with employees and supervisors together and a separate supervisor training. Supervisor/employee participatory training sessions occurred over the course of 6 weeks. Sessions were highly scripted with structured messages presented to all study groups, but also highly interactive in allowing participant responses and questions to change the discussion. Thus, every group received the same content, but the resulting outputs varied slightly to accommodate each study group’s needs. Manager expectations, informal work practices, and formal company policies were common discussion topics across study groups. Common proposed changes included a reduction in required “face-time” for performance evaluation, increased use of conference call lines and instant messaging as opposed to in-person meetings and drop-ins, and reduced negativity around non-work responsibilities during business hours (e.g., running an errand, taking a walk). The participatory sessions also strengthened the formal telecommuting policy and clarified work, communication, and scheduling expectations for telecommuting.

A separate, scripted, computerized training was provided to intervention supervisors that described the impacts of work-family conflict on business outcomes and reviewed behaviors and dialogue that supported both work and non-work support. Managers used an iPod touch to set and track goals of supportive interactions with coworkers over 2 weeks and received a personalized feedback report.

A comprehensive process evaluation was conducted alongside the implementation of the intervention to better tailor the intervention to the company’s needs and to track intervention fidelity. All study groups completed the planned training sessions, which were delivered by trained STAR facilitators. The average employee attended 74% of STAR sessions, while 10% of employees attended less than half of sessions and 3.9% attended no sessions.

A full description of the randomization procedure, data collection methods and measures, and pre-defined analysis protocols is provided in Bray et al.²³ STAR was implemented as part of a longitudinal, group-randomized trial using 56 study groups. Study groups were determined in collaboration with company management. We aggregated smaller teams up to a common management level. In some groups, a large team reported to a single manager. In other groups, multiple smaller teams reported to a senior executive. Study groups were spread across senior executives in the IT division and included a variety of job function and were of various sizes. These three criterion were used as conditions for the randomization - job function, organizational hierarchy, and size.

Study groups were randomized using an adaptive, biased-coin technique.²³ Four study groups were initially assigned to the intervention and control conditions using simple randomization. Next, each unassigned study group was assigned to the intervention condition and a balance assessment was made across randomization criterion. We then repeated this exercise by assigning the remaining groups to the control condition. To minimize the chance of imbalance, we used the lowest p-values from this simulation to adaptively alter the bias parameters of the randomization routine.

Employees were eligible for the baseline survey if they had non-contractor status and were located in the two participating locations. Employees who completed a baseline survey were eligible for follow-up unless they left the company; no employees were added to the sample after baseline. Self-report survey data were collected from employees using computer-assisted personal interviews at baseline and at 6, 12, and 18 months post-baseline by trained field interviewers who were blinded to employees’ study assignment.

Employees received $20 for completing the survey at each time point. All study procedures were reviewed by the RTI International Institutional Review Board and informed consent was obtained by a trained field interviewer. Employees who refused to participate in the survey still participated in the intervention if assigned to the intervention condition. The decision to implement the intervention was made by company management as a part of normal business operations and was not presented as an elective choice for employees.

Analyses were conducted on employees who completed baseline and at least one of the follow-up surveys. Of the 1,171 employees who were invited to participate in the baseline survey, 823 employees completed the baseline survey (70% response rate). In addition, 716 employees completed at least one follow-up interview. Fifteen employees who were randomized to the intervention condition but were never invited to participate in the STAR intervention due to a staff error were excluded. Furthermore, eight employees were excluded because they shifted reporting structures due to a business reorganization. Therefore, the final analytic sample included 693 individuals (61% of the randomized sample). For the intervention condition, the final sample included 348 employees at baseline and 6-month follow-up, 324 employees at 12-month follow-up, and 302 employees at 18-month follow-up. For the control condition, the final sample included 345 employees at baseline, 329 employees at 6-month follow-up, 334 employees at 12-month follow-up, and 345 employees at 18-month follow-up.

Kelly et al.¹³ examined nonresponse bias in the study sample, and we summarize their findings here. Women and non-white employees were more likely to complete the baseline survey, and the baseline participants were younger on average than the non-participants. Across time, younger employees were less likely to complete the survey. There was no evidence that employees leaving the company after baseline were different from employees that stayed.

The WFHN survey included four self-reported employee performance items from the World Health Organization Health and Work Performance Questionnaire (HPQ): (a) hours worked in the past 7 days, (b) expected hours per week, (c) personal job performance on a 10-point scale, and (d) coworker job performance on a 10-point scale.³² From these four questions, we calculated 5 additional performance measures as recommended by Kessler et al³²: absolute absenteeism ([4*b]-[4*a]), relative absenteeism ([4*b]-[4*a]/ [4*b]), relative hours of work (a/b), relative presenteeism (c/d), and total productivity (relative hours of work*relative presenteeism).³² We assessed the effect of the intervention on 9 outcomes: the 4 HPQ questions and the 5 measures calculated from the HPQ questions.

A pre-determined intent-to-treat (ITT), difference-in-differences approach was used to estimate the effect of the intervention on employee performance.²³ ITT classifies employees as intervention or control strictly based on the randomization process and is a standard approach used in randomized field experiments.

The models were estimated using generalized linear mixed models with random effects for individuals and for study groups.²³ Binary indicators for study assignment to the intervention group, for each data collection time point (6 months, 12 months, and 18 months post-baseline), and the interaction between the assignment and time indicators are the primary effects of interest in a difference-in-differences design, with the interaction terms capturing the intervention effect. Additional control variables include demographic characteristics, an indicator for having dependents (either children or dependent elders), and an indicator for knowledge of an impending merger that was announced during the study. We tested for moderation of the intervention effect by gender, age, race/ethnicity, and dependents using a three-way interaction of the moderator, time indicators, and study assignment indicators. Stata 14.0 was used to conduct all study analyses.³³