RESEARCH DOSSIER
Employee Turnover, Absenteeism, and Retention
Abstract. Employee turnover, absenteeism, and retention are interconnected employment outcomes shaped by worker health, job demands and resources, organizational relationships, labor-market opportunities, and decisions unfolding over time. This fundamental research monograph defines the principal constructs and competing estimands, traces the historical development of turnover and attendance theories, and evaluates evidence from primary longitudinal investigations, systematic syntheses, meta-analyses, administrative panels, and an organizational randomized trial. A critical narrative synthesis examines job satisfaction, commitment, embeddedness, unfolding decision pathways, sickness presenteeism, collective turnover, and the effects of work design and hybrid arrangements. It identifies the limitations of equating intentions with actual exits, treating all absence as discretionary withdrawal, or interpreting observational retention associations causally. The resulting framework connects person-level transitions, health-related attendance processes, and organization-level workforce stability while specifying analytic, intervention, and equity questions for future research. This work is a critical narrative synthesis of published literature; it reports no original experiment, preregistered systematic search, or new meta-analysis.
Keywords: employee turnover; voluntary turnover; employee retention; absenteeism; sickness absence; organizational withdrawal; job embeddedness; organizational commitment; turnover intention; unfolding model; job satisfaction; job demands–resources; presenteeism; survival analysis; collective turnover; hybrid work; retention interventions; organizational psychology
Scientific Overview and Research Questions
Employee turnover is a transition in an employment relationship, absenteeism is a recorded failure to attend scheduled work, and retention is the persistence of employment across an explicitly defined observation period. These three outcomes arise within a system of occupational opportunities, organizational obligations, health, personal circumstances, and changing preferences. A worker can remain employed while frequently absent; another can attend consistently until an abrupt resignation; a third can intend to quit yet stay because available alternatives are limited. A scientific account therefore requires several distinguishable outcomes and their joint dynamics rather than a single measure labeled employee withdrawal. The historical synthesis of Porter & Steers (1973) and the centennial analysis of Hom et al. (2017) establish the scholarly continuity of the problem, while Johns (2011) shows why attendance and productivity must be considered together. The central object of this monograph is the observed organization of staying, absence, and leaving, with attitudes treated as potential antecedents, mediators, and sometimes consequences.
The research questions are organized around transitions, mechanisms, and intervention. How do employees move from active membership to consideration of alternatives, preparation to leave, and actual exit? Under which circumstances is an absence episode a response to illness, care obligations, scheduling incompatibility, workload, or dissatisfying work? Which organizational practices strengthen continued employment without trapping employees in harmful conditions? How do team-level turnover, skill losses, and replacement practices change the consequences of individual decisions? The analysis also asks how evidence changes when one uses verified personnel records rather than recalled absence, measures actual quitting rather than intention, or follows workers longitudinally instead of studying a single cross-section. Griffeth et al. (2000), Rubenstein et al. (2018), and Harrison & Martocchio (1998) provide quantitative and historical foundations for these questions.
The method is a critical narrative synthesis of identified original research, theoretical contributions, meta-analyses, and systematic reviews. It uses an explicit distinction between descriptive regularity, longitudinal prediction, causal identification, and organizational recommendation. Original papers are interpreted at the level supported by the publisher text or abstract actually accessible; the randomized organizational trial in Bloom et al. (2024) additionally permits inspection of open full-text methods and outcomes. The synthesis is oriented toward theoretical integration and the quality of evidence, without claiming to have conducted a preregistered systematic search, an exhaustive evidence census, an independent meta-analysis, or a new study. Its source selection spans foundational industrial and organizational psychology, human-resource research, occupational health, applied decision science, and organizational economics. The scientific contribution of this form is disciplined integration of established findings with clear statements of what remains uncertain.
A useful unit of explanation is the employment spell, defined by a person, a particular employing organization, a start date, and an observable end or censoring date. Within the spell occur work attendance episodes, changes in attitudes, health trajectories, job offers, supervisory events, and changes in organizational structure. Some processes unfold over days; others require years of exposure, training, or accumulated relationship-specific investment. This temporal arrangement gives one subject a coherent structure: event history explains turnover timing, recurrent-event analysis explains absence, and organizational panels explain retention and collective workforce change. The account of absence time scales in Harrison & Martocchio (1998) is particularly important because a correlation measured over one period does not transfer automatically to a different interval.
Definitions, Outcome Boundaries, and Denominators
Voluntary turnover denotes employee-initiated separation; involuntary turnover denotes employer-initiated termination; retirement, death, contract completion, and transfers may require additional codes. An employee's movement from one position to another inside the same firm is internal mobility rather than necessarily organizational departure. The gross separation rate is conventionally calculated with separations in a defined interval as numerator and an appropriate average employment stock as denominator, but this rate suppresses tenure differences and treats unequal exposure as equivalent. Cohort-specific cumulative incidence or a hazard model instead defines the probability of leaving among those still at risk. Organizations should identify whether their operational concern is all separations, preventable voluntary exits, regretted departures, or the loss of particular competencies. The distinctions matter because an intervention that reduces layoffs changes an all-turnover statistic without necessarily changing employees' willingness to stay Hom et al. (2017), Dalton & Todor (1979).
Retention can be measured as the proportion of an initial cohort remaining employed after a given interval, with explicit attention to cohort formation, censoring, and transfers. Measuring retention among employees hired in January is fundamentally different from measuring the proportion of a company's workforce that has at least one year of tenure: the latter conditions on survival and changes with hiring volume. An apparently high retention rate can reflect satisfaction, strong attachments, constrained alternatives, scarce external vacancies, or a difficult local labor market. Mitchell et al. (2001) demonstrates that links, fit, and anticipated sacrifice explain additional variation in leaving beyond traditional attitude constructs. This leads to a crucial interpretive rule: observed staying identifies a behavioral outcome, while its psychological meaning must be investigated separately.
Absenteeism commonly measures spells or days of nonattendance relative to scheduled work. Sickness absence, certified leave, family-care leave, unauthorized absence, planned vacation, industrial action, and nonattendance caused by an employer scheduling error have different origins and normative status. Analyses must specify whether formally authorized leave is included and whether the denominator is scheduled days, scheduled hours, employee-time, or persons. A single count is affected by the number of workdays, job design, flexible scheduling and the employer's recording system. Johns (1994) found that self-reported absence is a measurement topic in its own right; Johns (2011) further distinguishes attendance from presenteeism, the decision to attend while ill. The term absenteeism should consequently operate as an empirical classification with transparent coding rules, not as a ready-made moral judgment about workers.
Turnover intention, search behavior, job dissatisfaction, disengagement, burnout, and absenteeism are predictors or associated states rather than interchangeable substitutes for employee exit. Intentions are often easier to survey and are theoretically meaningful because they may precede action, but timing, changing opportunities, shocks, and financial constraints can interrupt the intention–behavior sequence. Job dissatisfaction is likewise a heterogeneous appraisal of work rather than an observed separation. The process model in Mobley (1977) specifically elaborated intermediate psychological linkages between satisfaction and eventual turnover, and later synthesis in Steel & Lounsbury (2009) recognized common roles for morale, labor-market mechanisms, and quit/stay intentions. A valid study reports each construct distinctly and avoids recoding a correlation with turnover intention as a demonstrated effect on actual retention.
Historical Development of Turnover and Attendance Research
Early industrial psychology examined labor replacement, employee characteristics, workplace conditions, and the costs of lost organizational participation. The intellectual problem was framed simultaneously by personnel administration and by social-psychological explanations of why people enter and remain in organizations. Historical accounts in Hom et al. (2017) trace a movement from practical control problems toward the theory-driven investigation of employee attitudes, job alternatives, exit processes, and collective turnover. Mid-century organizational models associated participation with the perceived desirability and ease of movement; these ideas later became central to turnover theory. The historical relevance of a model should be distinguished from contemporary evidence for its predictions, because workforce composition, employment law, mobility channels, and measurement capabilities have changed substantially across the century.
By the early 1970s, Porter & Steers (1973) offered a broad scientific review linking organizational, work, and personal conditions to both turnover and absenteeism. This shared conceptual origin is important: both outcomes were often interpreted through a general logic of dissatisfaction and withdrawal. During the following decade, Mobley (1977) proposed intermediate thought and action sequences between dissatisfaction and resignation, thereby changing the explanatory question from who is dissatisfied to how departure occurs. By 1979, Dalton & Todor (1979) had established a complementary argument that separations can carry both organizational losses and potential benefits, depending on circumstances. A single turnover rate thus could never fully represent organizational effectiveness without identifying who left, how they left, and what followed.
Research on attendance developed a partly separate methodological trajectory. Repeated, intermittent nonattendance is statistically different from a one-time exit and can reflect illness patterns, recovery, ongoing obligations, and institutional leave rules. Harrison & Martocchio (1998) surveyed two decades of absenteeism research and emphasized time scale, temporal carryover, and the limited understanding of absence consequences across different horizons. Measurement scrutiny deepened when Johns (1994) examined what could reasonably be inferred from employees' recollections of absence. Subsequently, the relationship between sickness absence and being present while ill became a central occupational-health question Johns (2011), Skagen & Collins (2016). This shift reframed attendance as a health-related allocation decision with potential long-term effects, and it made an exclusively disciplinary interpretation scientifically inadequate.
The turn of the twenty-first century added pathways initiated by shocks, job embeddedness, attitudinal trajectories, group-level departures, and targeted retention mechanisms. In Lee & Mitchell (1994), a resignation can follow paths in which an external event or perceived violation plays a decisive role without a long period of growing dissatisfaction. In Mitchell et al. (2001), remaining becomes a subject of inquiry through links, fit, and sacrifice. Hausknecht et al. (2009) then documented heterogeneity in why workers say they stay. Hom et al. (2017) places these developments within the larger history of the discipline, while Rubenstein et al. (2018) revisited turnover antecedents through quantitative synthesis. The resulting research program studies staying and leaving as decisions embedded in social systems, career opportunities, and time.
Classical Sequential and Unfolding Models of Exit
Sequential turnover models begin with a perceived discrepancy between expected and experienced work, followed by evaluation of alternatives, intention formation, search, and eventual departure. The explanatory strength of the approach lies in its insistence that job satisfaction alone is temporally and psychologically insufficient: more proximal cognitions and practical options intervene. Mobley (1977) provided a concise early articulation, and Steel & Lounsbury (2009) found substantial conceptual convergence across models on morale, labor-market conditions, and intentions. Empirical testing nevertheless faces a measurement problem when the proposed stages are surveyed simultaneously. A cross-sectional association among dissatisfaction, search intention, and exit intention may fit the sequence but does not show that each stage occurred in that order. Diary observations, linked survey histories, and dated employment records offer stronger tests.
An alternative process is departure following a shock, defined as an event that prompts consideration of continued membership. Lee & Mitchell (1994) articulated an unfolding framework with distinct pathways, some involving strongly held personal or occupational images and others involving more deliberate job alternatives. The explanatory insight is that an employee can make a rapid decision after a managerial conflict, unexpected opportunity, personal life event, or organizational restructuring even when prior satisfaction was comparatively high. This does not imply that every surprise leads to resignation; the significance of the event depends on interpretation, available options, and the worker's existing attachments. Lee et al. (1999) reported a replication and extension involving former accounting employees and helped demonstrate both the appeal and the difficulty of coding heterogeneous decision histories.
The sequential and unfolding families address different temporal questions. Sequential frameworks describe accumulation and appraisal; unfolding models emphasize decision pathways and precipitating events. They can be combined in a state-transition description in which slowly changing attitudes alter vulnerability to shocks, while a shock changes the urgency or perceived feasibility of exit. This integration is a conceptual synthesis, not a claim that the cited studies have already estimated a single universal causal model. Hom et al. (2017) notes the growing importance of attitudinal trajectories, and Rubenstein et al. (2018) shows why differences among predictors and contexts require explicit moderation. Research should test whether observed quitting is better explained by pre-shock trends, post-shock changes, or the interaction of both.
An adequate model also permits the reverse decision: a person may search, receive an attractive offer, and then remain after a role adjustment, supportive intervention, or reassessment of relationships. Retention is not merely the absence of a quit event during a short window; it can include repeated decisions to renew an employment relationship under changing conditions. Some workers tolerate an unfavorable workplace because of healthcare access, housing ties, immigration constraints, financial commitments, or limited jobs nearby, whereas others stay because the work is meaningful and relationships are strong. Mitchell et al. (2001) and Hausknecht et al. (2009) demonstrate the importance of studying actual staying reasons. This distinction protects both theoretical validity and the ethical interpretation of organizational retention statistics.
Job Satisfaction, Commitment, and Withdrawal Cognitions
Job satisfaction is an evaluative orientation toward work conditions, tasks, relationships, and rewards; organizational commitment concerns the worker's relationship with the employing organization. Both have established associations with voluntary turnover, but neither provides a complete explanation of when or why separation occurs. The quantitative integrations of Griffeth et al. (2000) and Rubenstein et al. (2018) document that dissatisfaction, commitment, and intentions belong among recurring turnover antecedents, while also showing substantial heterogeneity among estimates and contexts. Their value lies in identifying robust correlates across studies, not in proving that an isolated increase in a questionnaire score will cause an equal reduction in departures. Workers continually compare their current circumstances with aspirations, alternatives, family obligations, and career timing. These comparisons can change after hiring, promotion, policy changes, and crises, making turnover a moving target for explanatory research.
The commitment framework distinguishes affective commitment, associated with emotional attachment; continuance commitment, associated with the perceived cost of departure; and normative commitment, associated with felt obligation to remain. Meyer et al. (2002) synthesized antecedents, correlates, and outcomes for these components and found that their implications for work attitudes and behaviors are not identical. A person who stays because the organization is valued differs psychologically from one who stays because departure entails severe financial or professional loss. Treating both as successful retention may be reasonable for a short-term headcount report, but it conceals different welfare, performance, and future exit implications. This distinction is especially important when retention policy changes benefit eligibility, restrictive agreements, relocation costs, or the availability of advancement.
Withdrawal cognitions include thoughts of quitting, perceived desirability of leaving, consideration of alternative employment, and stated plans to search or resign. In Mobley (1977), these cognitions occupy intermediate positions in an exit process; Steel & Lounsbury (2009) shows how later models preserve the importance of morale, intention, and market constraints while developing different process explanations. A prospective study should date the intention measurement, the search interval, the external offer, and the actual termination. Contemporaneous questionnaires can record coherent narratives without observing the sequence they imply. Intention can also fluctuate rapidly after a job offer disappears, a supervisor changes, or care demands increase. For this reason, survey indicators are most informative when anchored to a specific behavioral window and coupled with observed human-resource records.
Employee attitudes are often measured with multi-item self-report instruments, while voluntary separation is identified in organizational systems. Such designs limit same-source measurement bias for the behavioral endpoint, although selection into survey response remains consequential. The literature summarized by Hom et al. (2017) and Rubenstein et al. (2018) gives strong reasons to measure satisfaction and commitment separately, test nonlinear or interactive patterns, and account for job alternatives. A useful prospective model might include baseline satisfaction, change in satisfaction, embeddedness, health, supervisor turnover, and external opportunities, with exit observed subsequently. Even such a model cannot establish causal effects without defensible assumptions: unmeasured personal events may shift both attitudes and departure. The conceptual distinction among feeling dissatisfied, deciding to leave, and leaving remains a primary requirement of sound inference.
Embeddedness, Social Ties, and the Psychology of Staying
Job embeddedness explains continued employment through the network of links connecting workers to people and activities, perceived fit between person and setting, and anticipated sacrifices associated with leaving. It redirects inquiry from factors pushing someone out to connections making continued participation valuable or costly. Mitchell et al. (2001) introduced and tested this construct in relation to voluntary turnover, showing that embeddedness contributes to prediction beyond familiar attitudinal measures in the settings studied. The construct extends outside the workplace to community attachments, housing, family circumstances, and local ties. It therefore places the employment decision within a relational ecology: an apparently equivalent job offer may be attractive to a worker with few local commitments and difficult to accept for a worker caring for relatives or embedded in neighborhood networks.
Links, fit, and sacrifice require separate observation because the same retention pattern may arise from different configurations. Extensive coworker ties can provide support and knowledge sharing; strong fit can make daily tasks meaningful; substantial sacrifice can include lost benefits, professional status, or disrupted family arrangements. Yet links also entail obligations, fit may change as jobs evolve, and sacrifice can be experienced as pressure. The original embeddedness proposal Mitchell et al. (2001) invites hypotheses about interaction, but the importance of each dimension is likely to vary by labor market, occupation, age, and employment arrangements. In a multi-location organization, measures must distinguish connections to the team, the employing firm, the profession, and the locality. Otherwise, researchers may confuse attachment to coworkers with attachment to a particular employer.
The reasons employees themselves report for staying offer a complementary perspective. In a large organizational field investigation, Hausknecht et al. (2009) examined stated retention reasons and assessed differences associated with performance and job characteristics. Such accounts can illuminate satisfaction, rewards, commitment, development opportunities, and workplace relationships in the population studied. They are valuable for understanding perceived meaning and managerial relevance, while retrospective or contemporaneous reasons are not equivalent to manipulated causes of retention. Employees may rationalize a continuing employment choice, and those who already left are absent from a survey of current workers. Combining stated reasons with longitudinal turnover records and matched observations from leavers helps reduce survivor selection and connects explanations to actual behavior.
Retention strategies grounded in embeddedness should distinguish strengthening chosen connections from increasing costs of departure. A mentoring program or improved team climate can raise the value of staying and support well-being; a restrictive arrangement may also increase tenure but have different ethical and functional implications. Allen et al. (2010) emphasizes moving retention practice toward evidence-based rather than intuitive policies, and Hausknecht et al. (2009) cautions against assuming every worker stays for the same reason. An informative evaluation records whom the intervention reaches, changes in perceived fit and support, eventual exits, and employee welfare. It should also check whether fewer separations coexist with increased absenteeism, psychological distress, or suppressed mobility. Retention is a favorable outcome when it is aligned with sustainable work and the employee's informed choice.
Job Demands, Resources, and Occupational Health
Job demands–resources theory organizes working conditions into demands requiring sustained effort and resources that enable goal achievement, learning, support, and recovery. Demands may include workload, time pressure, role conflict, emotional labor, and physical hazards; resources can include autonomy, helpful supervision, feedback, reliable equipment, and cooperative colleagues. The contemporary synthesis in Bakker & Demerouti (2024) clarifies the theory's central motivational and health-impairment pathways, while the earlier meta-analytic test by Crawford et al. (2010) connects demands and resources to engagement and burnout. The resulting framework is relevant to turnover and absence because occupational demands can influence both the desire to quit and the capacity to attend. Its explanatory contribution is conditional: the significance of demands depends on their form, available resources, and occupational context, rather than on a universal assumption that all demands function alike.
Health mechanisms connect work conditions to attendance with particular immediacy. High exposure to physical hazards or sustained stress can produce illness, fatigue, sleep disruption, and recovery needs. A worker's absence may then protect health and prevent transmission of infection, while organizational pressure to attend can generate sickness presenteeism. Johns (2011) examines attendance dynamics across both absenteeism and presenteeism, and the systematic longitudinal review Skagen & Collins (2016) addresses possible later health consequences of attending while ill. These processes should be represented through measured exposure, symptoms, leave rules, and later outcomes. Concluding from absence frequency alone that a worker lacks motivation would omit the health process that may be producing the observation.
Resources may affect retention through several overlapping channels: improved job satisfaction, greater professional efficacy, psychological safety, meaningful work, stronger social integration, and decreased exhaustion. Crawford et al. (2010) concerns engagement and burnout rather than proving an intervention-specific effect on quits; Rubenstein et al. (2018) addresses turnover correlates but not every proposed mediation. Integrating these lines of evidence thus produces a testable explanatory pathway rather than an already established chain in all organizations. A longitudinal study should measure changes in demands and resources before changes in health or satisfaction, and then observe subsequent attendance and employment transitions. Where a program randomly modifies scheduling autonomy or supervisor practices, its effects on later exits and absence can be assessed with substantially stronger causal leverage.
Occupational health and withdrawal may also form feedback loops. Following repeated absence, lost learning opportunities, strained coworker relationships, and changing managerial expectations may make continued employment less sustainable. Conversely, working while unwell can temporarily keep absence rates low while increasing later health risk. Harrison & Martocchio (1998) supports analyzing absence over multiple time windows, and Johns (2011) emphasizes the interaction among sickness behavior, work demands, and productivity loss. The organizational unit of observation matters: a team with inadequate staffing may impose heavier demands on remaining members after a resignation, leading to subsequent absences and departures. A multilevel approach should therefore treat individual absence and exit as processes potentially coupled through workload redistribution, not simply as independent personnel indicators.
Labor Markets, Careers, and the Feasibility of Exit
The same subjective desire to leave can yield different observed outcomes in a growing labor market and a contracting one. Available vacancies, geographic mobility, credentials, wages, family obligations, and institutional rules shape the feasibility and expected consequences of moving. In the synthesis of turnover-process models Steel & Lounsbury (2009), labor-market opportunities operate alongside morale and intentions. The broad meta-analysis Griffeth et al. (2000) also places perceived alternatives among turnover predictors. The distinction between objective and perceived opportunity is important: workers may perceive few suitable vacancies despite a high regional vacancy rate, or may have substantial professional networks that create opportunities not visible in public statistics. Research should therefore measure relevant occupational alternatives rather than substitute a single national unemployment rate for each worker's actual choice set.
Career stage changes what staying means. An early-career employee may value training and portable credentials, a midcareer employee may value predictable care-compatible schedules, and an older employee may face retirement options or pension-related incentives. These are plausible mechanisms requiring population-specific tests; they should not be converted into stereotypes about age or presumed motivation. Hausknecht et al. (2009) found heterogeneity in employees' stated retention reasons, supporting differentiated inquiry. A study seeking causal explanation can follow cohorts across career stages while controlling for tenure, job level, organizational change, and broader labor-market cycles. It must also distinguish age effects from historical cohort experiences and period effects, which cannot be automatically identified from a single cross-section.
Internal transfers, promotions, and lateral career moves affect external turnover by changing the range of opportunities available without requiring departure from the employer. They can also shift the personal meaning of organizational attachment: an employee may value the profession or the team more than the current role. The turnover histories reviewed by Hom et al. (2017) show why contemporary research includes trajectories and alternative employment relationships. A useful empirical contrast compares workers offered an internal move with otherwise comparable workers who were not, recording subsequent external exits and satisfaction. Because those selected for mobility differ on performance, preferences, and managerial support, naive comparisons are vulnerable to selection bias. Employee-level panel designs and carefully identified policy changes provide stronger evidence than comparisons of promotion recipients with nonrecipients.
The broader institutional environment affects both the observed number of departures and their interpretation. Income insurance, paid sickness leave, access to medical care, disability accommodations, collective bargaining, temporary-contract rules, and immigration-related employment restrictions alter the costs and permissions attached to absence and exit. These institutions vary across jurisdictions and over time; an organizational sample from one country cannot automatically support worldwide retention prescriptions. Porter & Steers (1973) already recognized personal and organizational determinants, and subsequent synthesis in Hom et al. (2017) documents the progressive expansion of turnover theory. A scientifically useful comparative agenda specifies policy exposure, worker eligibility, cultural context, and measurement invariance before inferring that cross-national differences in absence or quitting reveal fixed psychological traits.
Absenteeism as a Recurrent, Time-Structured Outcome
Employee absence is observed repeatedly across a continuing employment spell. A worker can have several short episodes, one prolonged sickness spell, or an extended sequence linked to a single condition. These patterns have different probable causes and different implications for work organization. Harrison & Martocchio (1998) proposed that time scale should organize attendance research and reviewed evidence concerning the origins, offshoots, and consequences of absence. For event-level analysis, the relevant information includes the beginning and end of each episode, scheduled hours missed, reason categories, whether leave was paid or certified, and the person-time available for an episode to occur. For organizational comparison, coverage rules, shift systems, and employee eligibility should also be documented. Without a common definition, cross-firm differences in absence rates can reflect administrative accounting rather than underlying differences in health or motivation.
A frequency measure counts spells, a duration measure sums missed days or hours, and an incidence measure identifies whether a person experienced any qualifying absence during an interval. Recurrence depends on a worker's prior episodes and changing exposure, so statistical models assuming independent, equal-risk observations may be unsuitable. An occupational-health cohort can estimate the first long sickness absence separately from recurrent short absences; a human-resource dashboard can present both frequency and severity. The research surveyed by Harrison & Martocchio (1998) indicates why longer and shorter causal windows should not be collapsed. It also suggests that prior absence may matter because health problems persist, work conditions remain unchanged, or the individual experiences consequences of earlier episodes. These rival explanations require measurement of the underlying conditions, not simply an autoregressive association.
An observed absence is always conditional on the ability and obligation to attend a scheduled work period. Employees on annual leave, nonworking shifts, unpaid furlough, and legally protected leave may be excluded or coded differently depending on the research question. Remote and hybrid work further change the practical meaning of physical absence from a workplace. Johns (2011) shows that health, work conditions, and productivity need to be considered alongside attendance, while Johns (1994) highlights discrepancies and uncertainties in reported absence data. An adequate data dictionary distinguishes the work location expected, whether work could be performed remotely, and the classification applied by the employer. These choices directly determine the numerator and denominator of the reported rate and, therefore, the scientific object under study.
Absence may represent both a consequence of earlier conditions and an event that changes later conditions. A worker who takes protected medical leave might recover and remain employed; a worker who continues working through an illness could subsequently experience a longer absence; another might leave because the original job no longer accommodates a health limitation. Such pathways imply different temporal ordering and potential competing events. Skagen & Collins (2016) reviews longitudinal evidence on sickness presenteeism and health-related outcomes, while Mitra et al. (1992) synthesizes earlier associations between absence and turnover. A sufficiently rich panel can test whether absence predicts departure after accounting for measured health, tenure, workplace conditions, and available alternatives. It should not presume that absence caused leaving when the underlying illness or restructuring could produce both.
Presenteeism, Sickness Absence, and Working While Ill
Sickness presenteeism refers to attendance at work despite illness; it is operationally distinct from working efficiently and from ordinary absence. A person may attend because of financial necessity, staffing pressure, strong occupational identity, difficulty finding replacement coverage, or the belief that duties cannot be postponed. The study of attendance dynamics in Johns (2011) examines the correlates of presenteeism, absenteeism, and productivity loss, underscoring that these outcomes can coexist rather than moving in opposite directions. A system that records only missed days may therefore reward attendance in circumstances where rest, medical care, accommodation, or remote work would better preserve long-term function. The relationship between attendance and health requires context about symptoms, job demands, workplace protections, and subsequent recovery.
The systematic review by Skagen & Collins (2016) examined longitudinal evidence on the health and well-being consequences of sickness presenteeism over time. Its importance for retention research lies in the temporal possibility that continued attendance while ill can predict later negative health outcomes, which in turn may affect absence and employment continuity. However, observational longitudinal designs remain susceptible to reverse causation and confounding: people with more severe baseline symptoms may have both greater pressure to attend and greater later impairment. A careful synthesis should examine how the included studies defined illness, selected workers, measured changes, and adjusted for initial health. The review supports treating presenteeism as a serious occupational-health research variable, while causal or universal claims about all occupations require stronger evidence.
Work organization can influence sickness-attendance decisions through formal leave access, pay replacement, staffing levels, workload, supervisor responses, and employee discretion. The job demands–resources framework Bakker & Demerouti (2024) identifies work conditions likely to shape exhaustion and resources; Johns (2011) links the attendance decision to productivity considerations. An intervention designed to reduce absence should therefore measure health and operational effects, not simply the number of sick days used. For example, changing sick-leave eligibility can alter recording practices, willingness to report illness, and the duration of episodes. If the company observes fewer certified sickness days after tightening eligibility, the statistic alone cannot establish healthier employees. Health outcomes and worker experience must accompany attendance results for a defensible evaluation.
A further scientific concern is exposure externality: attending while infectious can affect coworkers, customers, and vulnerable populations, especially in care or shared indoor settings. This possibility makes sickness absence potentially protective at the organizational level. The relevant policy evidence depends on disease transmission, occupation, and jurisdiction; general organizational psychology reviews alone do not quantify those effects. The synthesis of work and health in Johns (2011) establishes why attendance cannot be assigned a single positive value independent of conditions. Researchers studying retention must also consider whether repeated sickness attendance decreases later work capacity, whether protected leave preserves the employment relationship, and whether job accommodations change both outcomes. These are linked but empirically distinct hypotheses.
Connections Between Absence and Employee Turnover
Absence and exit can be manifestations of a broader deterioration in job fit, health, or organizational attachment, yet they are not sequential stages for every employee. The early integrative review Porter & Steers (1973) considered both turnover and absenteeism as organizationally significant outcomes. The quantitative synthesis in Mitra et al. (1992) subsequently assessed the relationship between absence and turnover across studies, demonstrating the value of examining the association directly rather than assuming identical causation. Individual employees can be absent for legitimate health reasons while strongly intending to stay, and an employee may leave a favorable job without preceding absence after an unexpected opportunity. A scientifically defensible model therefore permits a common cause, an absence-to-exit pathway, an exit-to-absence anticipatory process, and independence under different circumstances.
Temporal structure can distinguish these alternatives. Under a shared-demand hypothesis, an increase in workload or a deterioration in health predicts both later absence and eventual leaving. Under an absence-consequence hypothesis, repeated episodes contribute to lost work connections or altered job assignments and thereby affect subsequent departure. Under an anticipatory-withdrawal hypothesis, a worker preparing to leave may reduce attendance before the resignation, although contract rules, health, and job opportunities may explain the same pattern. Harrison & Martocchio (1998) and Hom et al. (2017) provide reasons to model trajectories and intermediate events instead of relying solely on annual correlations. Tests require dated personnel information and, preferably, repeated measures of occupational conditions, health, and intentions before and after each event.
A simple statistical analysis that regresses later turnover on previous absenteeism may appear compelling, especially when both are measured by administrative records. Yet prior absence often proxies for health status, caregiving demands, shift instability, or an organization that poorly accommodates disability. Without those measurements, a turnover coefficient can be observationally predictive while remaining causally ambiguous. Johns (2011) demonstrates the larger attendance context, and Griffeth et al. (2000) emphasizes that exit has many determinants. A useful prospective design would stratify by absence reason and distinguish short episodes from sustained sickness absence, examining whether effects persist within occupational groups and after changes in medical circumstances. It would also test whether workers who take leave and return to modified duties have different retention trajectories from those without suitable accommodation.
Absence policy and retention policy can generate conflicting incentives. A punitive attendance score might reduce recorded unapproved absence in the short term while encouraging presenteeism and increasing dissatisfaction; generous leave might initially raise the number of recorded absence days as employees gain access to legitimate recovery. The possible consequences must be evaluated, not asserted in advance as universal results. Skagen & Collins (2016) gives reasons to monitor health over time, and Allen et al. (2010) grounds retention management in verifiable organizational outcomes. Interventions should report attendance subcategories, health or recovery indicators, employment duration, voluntary versus involuntary separations, and employee perceptions of fairness. A program that reduces an undifferentiated absence rate but raises preventable departures cannot be judged successful from the absence metric alone.
Measurement, Data Quality, and Construct Validity
Scientific measurement begins by specifying the event, time origin, population at risk, observation window, and expected recording process. For turnover, administrative systems may identify resignation dates, but coding errors can combine voluntary quits with contract expiration, employer termination, or intercompany transfers. For attendance, scheduling systems may contain missed shifts without identifying certified sickness, caregiving leave, or protected accommodation. Johns (1994) demonstrates why retrospective self-report of absence needs its own validity assessment, and Hom et al. (2017) reviews the wider evolution of turnover measures. A credible empirical study should audit a sample of records, document classification rules, report the prevalence of missing reason codes, and describe how transfers and simultaneous employment spells are handled. Measurement choices influence the observed result before statistical modeling begins.
The denominator is as important as the numerator. A turnover proportion computed from a workforce snapshot mixes employees with sharply different tenure and exposure periods; a short-time employee might have only two weeks at risk while another contributes the whole year. Person-time and event-history methods can handle staggered entry, although the choice depends on the analytic question. Absence counts likewise depend on scheduled workdays and shift length, making comparisons between full-time and part-time employees problematic unless exposure is standardized. Harrison & Martocchio (1998) emphasizes time-specific absence processes, while Steel & Lounsbury (2009) surveys turnover process models whose predictions implicitly depend on timing. Reporting both raw counts and appropriately defined rates makes it possible to reproduce analyses and identify artifacts created by staffing changes.
Self-reported satisfaction and retention reasons are vulnerable to recall, social desirability, common-method variance, and shifting interpretation of survey items. In Hausknecht et al. (2009), employees' stated reasons for staying provide informative perceptions but are conditioned on continued employment. Mitchell et al. (2001) demonstrates why embeddedness requires measures distinct from generic satisfaction and intention. A strong design combines appropriately validated scales with independently recorded exits, tests measurement equivalence across occupational or demographic groups, and checks whether missing survey responses are systematically related to impending resignation. If leavers disproportionately fail to answer a later questionnaire, an analysis of only responders will overrepresent long-tenure attitudes. Sampling and attrition should therefore be reported as part of the scientific findings.
The distinction among predictive validity, explanatory validity, and practical utility is critical. A machine-learning model may accurately classify departures from attendance, tenure, or schedule data, yet supply little evidence that changing one feature will prevent resignation. Similarly, a retention intervention may improve a satisfaction scale without altering the probability of exit in the relevant window. Rubenstein et al. (2018) shows that antecedent relations require careful analysis of moderators, and Bloom et al. (2024) demonstrates an intervention design with observed behavioral outcomes. Good reporting distinguishes between model discrimination, calibration, causal effects, and operational outcomes. It also states which workers were included, which people were excluded, and how the learned or inferred relation might fail under changes in organizational policy.
Event-History, Longitudinal, and Multilevel Research Designs
Employee turnover is naturally analyzed through time-to-event methods because the scientific question often concerns when a separation occurs and which exposures alter that timing. A proportional-hazards model can estimate an association between covariates and the instantaneous exit hazard among workers still employed, while a discrete-time hazard specification can accommodate monthly or quarterly personnel panels. These approaches handle staggered entry and right censoring more transparently than a simple binary model of whether an employee ever left. Yet they require definitions of time origin, risk set, time-varying predictors, and competing events. If involuntary termination prevents the observation of later voluntary resignation, a cause-specific hazard and a cumulative-incidence analysis address different questions. The process-oriented theories surveyed in Hom et al. (2017) and Steel & Lounsbury (2009) make these choices central rather than optional technical details.
Attendance data usually require a different family of models because a worker can experience multiple episodes. Recurrent-event survival models, multilevel count models with a scheduled-time offset, and hurdle or zero-inflated structures may be useful depending on the distribution and data-generating process. Short and long sickness spells may have different explanatory correlates and should be analyzed separately when the evidence supports doing so. Harrison & Martocchio (1998) stresses the role of time in absence research, and Johns (1994) underscores uncertainty in self-reported episodes. A model that includes many employees with zero recorded absences must distinguish genuine zero events from unobserved exposure and from incomplete records. Measurement audits and sensitivity analyses should precede substantive conclusions about personality or work attitudes.
A longitudinal model should represent changing circumstances explicitly. An employee can move between teams, receive a new supervisor, experience a health disruption, and change work arrangements during the same employment spell. Fixed-effects or within-worker models can reduce confounding by stable individual characteristics, but they do not automatically control time-varying confounders or measurement error. Event-study designs around policy changes can reveal temporal patterns if comparison units are suitable and pre-intervention trends are adequately assessed. Rubenstein et al. (2018) documents moderator variation in turnover correlates, while Bloom et al. (2024) illustrates the additional identification provided by a randomized organizational treatment. Researchers should specify whether they are estimating a predictive association, a total causal effect, or a conditional relationship within a selected surviving cohort.
Organizations contain teams, managers, occupations, and sites, and employees in the same unit share conditions. A multilevel model can estimate variation among those units and test whether job demands, turnover, or supervisor changes have collective effects. It should avoid treating individual employee responses as independent when policy assignment or managerial practice is common to an entire team. Hancock et al. (2013) synthesizes organizational-performance consequences of collective turnover, indicating that the team or firm may be the relevant causal unit for some questions. Aggregating individual satisfaction into a team average also raises compositional concerns: a high-turnover team may have changing membership and response rates, so its apparent attitude trend can partly reflect who remains available to be surveyed. Strong panel construction makes these dependencies visible.
Causal Inference, Selection, and Research Integrity
Turnover research frequently identifies associations among pay, leadership, satisfaction, health, external opportunities, and observed exits. Interpreting these associations causally requires a plausible account of which variables precede others and which common causes shape both exposure and outcome. For example, a supportive supervisor may reduce employee exits, but high-performing or stable teams may also attract more supportive supervisors. Surveys of remaining staff after a resignation can miss the attitudes of employees already gone. Griffeth et al. (2000) and Rubenstein et al. (2018) are important summaries of correlates; their pooled estimates remain dependent on the designs and measurements of included studies. A claim that changing a correlate will necessarily reduce quits is therefore stronger than the evidence supplied by a cross-sectional or conventional longitudinal association alone.
Selection can enter at many points. Employees choose jobs; firms choose applicants; workers with poor health leave some roles; survey response depends on engagement; and managers can selectively grant scheduling flexibility. Adjusting for a factor measured after treatment can itself create bias when that factor is affected by the intervention or acts as a collider. A causal diagram can make assumptions explicit before statistical adjustment, but the diagram must represent a considered scientific hypothesis rather than claim to have resolved hidden confounding. In Bloom et al. (2024), assignment to hybrid work was randomized in a defined workforce, addressing a key selection problem for the treatment comparison. The study still requires attention to compliance, the precise intervention offered, differential outcomes by subgroup, and external validity outside that firm.
Prediction studies should use temporal validation and leakage controls. If personnel data include codes created at resignation, a classifier may appear to predict exit while merely detecting an already initiated separation process. Models trained and evaluated on random records from the same workers can capture idiosyncratic histories without generalizing to future employees. The conceptual concerns in Steel & Lounsbury (2009) and the outcome distinctions in Hom et al. (2017) indicate why the prediction window must match the decision window. Evaluation should report calibration across cohorts, false-positive burdens, uncertainty around estimates, and changes under labor-market shocks. A high area under a receiver-operating-characteristic curve can coexist with poor precision when quitting is rare. Prediction does not create an ethical license to surveil or penalize employees based on opaque scores.
Cumulative scientific inference benefits from triangulation. Meta-analyses such as Griffeth et al. (2000) and Rubenstein et al. (2018) summarize broad predictor evidence and moderation; event-history panels reveal trajectories; qualitative exit interviews describe meanings and shocks; and randomized interventions permit stronger causal statements for specific policy contrasts. Agreement among designs increases confidence in a mechanism, while discrepancies can be informative about population, implementation, outcome coding, and time horizon. Reporting should include confidence intervals, absolute risks, subgroup analyses with appropriate caution, missing-data procedures, and the precise intervention under study. The absence of a statistically significant performance difference should not be described as proof of universal equivalence unless the study used a justified equivalence or noninferiority framework.
Randomized Evidence on Hybrid Work and Retention
The field experiment by Bloom et al. (2024) supplies a rare direct causal test of a work-organization policy with observed employee departures. In a large Chinese technology company, 1,612 graduate employees from engineering, marketing, and finance-related divisions participated in a six-month randomized comparison during 2021–2022. Assignment was based on parity of birth dates: the intervention offered work from home on Wednesday and Friday alongside three office days, while the comparison group was scheduled in the office five days a week. The design substantially reduces preexisting differences between the groups as an explanation for the intention-to-treat contrast. Its causal estimand nevertheless applies to the hybrid option as implemented in that company and period, not to all remote work arrangements or every occupational setting.
The authors report an observed quit rate of 7.2% in the control group and 4.8% in the hybrid group during the trial. This is a difference of 2.4 percentage points and a relative reduction of roughly one-third, reported with a two-sided test at P = .043 Bloom et al. (2024). The absolute difference is essential because its meaning changes with the baseline risk and observation window. It does not mean that every organization adopting hybrid work will experience a one-third reduction in employee separations. The paper also reports improved job-satisfaction survey scores, allowing a plausible psychological interpretation of the retention effect. A mediation claim, however, would require analysis establishing how much of the effect operates through satisfaction rather than commuting costs, work–family coordination, or other concurrent mechanisms.
An especially important result is the authors' assessment of performance. The study reports no detected deterioration in performance ratings over two years of evaluation following the initial experiment, with formal equivalence analyses for selected outcomes; it also reports no overall promotion difference and no evidence of a change in lines of code among computer engineers Bloom et al. (2024). These measured endpoints are more informative than a general claim that flexibility improves productivity everywhere. Performance grades and code output cover limited dimensions of work, and informal learning, team coordination, recruitment effects, or long-run innovation require their own designs and observation windows. The reported retention reduction was particularly pronounced in certain subgroups, including nonmanagers and employees with long commutes, underscoring potential variation in benefits. Subgroup results are best interpreted in light of the prespecified comparisons and statistical uncertainty.
The hybrid-work experiment also helps explain why policy evaluation should combine employee and organizational outcomes. Flexibility changes daily travel, control over work schedules, household logistics, and perceived support. Those changes may produce satisfaction effects and altered quit decisions without necessarily shifting objective performance in the same interval. Bloom et al. (2024) shows this pattern for the studied organization. The generalizable methodological lesson is the usefulness of identifying the exact work-design intervention, randomizing when feasible, preserving an appropriate control condition, and tracking objective departures beyond immediate attitude surveys. Future replications should include different countries, frontline occupations, employees with limited home working conditions, and organizations facing different staffing constraints before broad normative claims are made.
Retention Policies and Organizational Interventions
Organizational retention policy can modify recruitment realism, onboarding, supervisor quality, pay structures, career progression, work design, employee voice, flexibility, or health-related support. The evidence-oriented synthesis in Allen et al. (2010) argues for testing retention assumptions rather than adopting common managerial myths. A scientifically designed program states which employees it targets and which mechanism it intends to change. An onboarding intervention might increase information and coworker links; a supervisor-training intervention might alter daily support; flexible schedules might reduce work–family conflict and commuting burden. Those mechanisms should be measured in addition to exits. Programs bundled together may improve retention, but without experimental or credible quasi-experimental comparison, the organization cannot determine which component caused the difference.
Targeting high-risk groups presents both opportunity and methodological risk. Workers with short tenure, changing roles, intensive schedules, or recent managerial changes may have different exit hazards, but apparent risk also reflects recruitment practices and the composition of those groups. Hausknecht et al. (2009) supports examining variation in reasons for remaining, while Rubenstein et al. (2018) suggests the importance of moderators across turnover predictors. A retention strategy should therefore match problems to mechanisms rather than presume that a single benefit will affect every employee equally. It should compare treatment reach and outcomes across employment groups and monitor whether intervention eligibility reproduces existing organizational inequities. If only employees assessed as irreplaceable receive support, aggregate retention improvement may hide deteriorating conditions for other workers.
Evaluation should distinguish implementation from effect. Participation rates, supervisor compliance, actual schedule changes, and use of support services describe whether the intervention occurred. The primary outcome may be six- or twelve-month voluntary turnover, with secondary outcomes including sickness absence, performance, satisfaction, well-being, and internal mobility. Bloom et al. (2024) provides an example of measured objective quits following a specified policy contrast; Allen et al. (2010) provides broader evidence-based principles. Ideally, an experiment uses predeclared primary outcomes and sufficient follow-up for career decisions. When randomization is infeasible, difference-in-differences or interrupted time-series methods require defensible comparison groups and sensitivity checks. Program effectiveness should be expressed in absolute changes and uncertainty intervals rather than promotional descriptions.
A strong retention program also asks what is being retained. If a company experiences lower voluntary departure because employees cannot afford to lose benefits, the headcount metric may improve while trust and well-being remain poor. If a new staffing system eliminates harmful overtime and supports recovery, retention and health may improve together. The multi-component commitment analysis Meyer et al. (2002) and the embeddedness construct Mitchell et al. (2001) provide a conceptual basis for distinguishing valued attachment from perceived constraint. Employers and researchers should include employee agency, career development, health, and equitable treatment when deciding which retention outcomes are desirable. Reduced turnover is a useful organizational result only when interpreted together with the conditions that produce it and the consequences that follow.
Collective Turnover and Organizational Performance
Individual departure decisions accumulate into a collective turnover process that can affect an organization's skill resources, coordination, institutional memory, service quality, and staffing costs. The meta-analysis by Hancock et al. (2013) examined the association between collective turnover and organizational performance across empirical samples, with effects conditioned by outcome type and organizational circumstances. This synthesis makes clear why a headcount rate cannot be analyzed only as an individual psychological consequence. Departures may break established working relationships and impose replacement costs, while some departures create openings for renewal, improved person–job fit, or needed organizational restructuring. Earlier analysis by Dalton & Todor (1979) already argued that turnover can have positive as well as negative organizational implications under different conditions.
The performance consequence of turnover can depend on the knowledge held by departing employees, the density of working relationships, replacement availability, and the timing of exits. Losing a specialized technician during a critical operation differs from losing an employee in a role that can be rapidly refilled; an entire team resigning after a manager change differs from independent individual exits. Hancock et al. (2013) provides quantitative synthesis at the collective level but does not establish identical effect sizes in every industry. An organizational study should measure workforce stability alongside turnover composition, vacancies, skills, product quality, customer outcomes, and the cost of temporary coverage. It should also model feedback: deteriorating performance may produce employee exit, while employee exit may subsequently deteriorate performance.
Absenteeism and collective turnover can interact through staffing reserves. When experienced workers depart, fewer employees may be available to cover sickness leave, which can increase overtime and encourage presenteeism. Those pressures might in turn contribute to exhaustion and future resignation. These are plausible dynamic mechanisms supported conceptually by the attendance literature Johns (2011), job demands–resources theory Bakker & Demerouti (2024), and collective-turnover research Hancock et al. (2013), but the cited studies do not jointly test every link in a single identified pathway. Future panel designs could observe team staffing changes, workload, sickness episodes, perceived support, and subsequent exits with temporal lags. Such work would clarify when workforce shocks propagate through occupational health and organizational performance systems.
The economically relevant question is often the marginal effect of a particular preventable loss, not a universal mandate to minimize every separation. Retaining workers whose skills are central to safety or service continuity may be highly valuable, while turnover associated with normal career mobility is a legitimate dimension of labor markets. Dalton & Todor (1979) cautions against uniformly negative interpretations, and Allen et al. (2010) grounds strategic retention in the problems an organization can responsibly address. Evaluation should therefore differentiate regretted and non-regretted voluntary turnover using transparent criteria, avoid defining employee worth solely by managerial preference, and monitor fairness in how retention investments are allocated. Scientific precision improves organizational decision-making when outcomes and ethical considerations are specified together.
Occupational Differences, Equity, and Institutional Conditions
Turnover and attendance processes are shaped by material differences among occupations. A hybrid schedule is feasible for many information-processing roles and impractical for some forms of emergency care, manufacturing, transportation, or on-site service. The randomized evidence in Bloom et al. (2024) therefore identifies an important work-design effect within a bounded occupational population. Its implications for other workplaces should be evaluated through functionally comparable interventions: predictable shift scheduling, paid leave, staffing buffers, job autonomy, and reliable accommodations may supply similar resources through different operational means. The job demands–resources synthesis Bakker & Demerouti (2024) provides an organizing perspective for identifying each role's specific demands and resources. Research should compare intervention effects across work arrangements with explicit treatment definitions rather than assume that identical policies are feasible or useful in every job.
Differences in caregiving responsibilities, disability, immigration-related work constraints, travel distance, and access to healthcare may change both attendance opportunities and employees' capacity to leave. These differences require evidence grounded in a specific population and applicable legal regime, and they should never be inferred from demographic membership alone. The observed effects in Bloom et al. (2024) varied across study subgroups, inviting investigation of schedule benefits and unequal commuting burdens. The attendance analysis Johns (2011) similarly underscores the relevance of health and work conditions. A retention study that fails to measure protected leave, reasonable accommodations, or unequal access to flexibility may interpret institutional restrictions as a psychological preference for staying. Stronger research distinguishes the distribution of opportunities from individual evaluations of those opportunities.
Comparative analysis requires consistent constructs across countries and organizations. The same label, such as sickness absence, may include different statutory entitlements, record-keeping thresholds, probation rules, and pay replacement. Turnover classifications can also depend on the treatment of fixed-term contracts, retirements, and layoffs. The historical scholarship Porter & Steers (1973) and Hom et al. (2017) demonstrates that workplace relations evolve within a changing institutional context. Researchers should report country, sector, contract type, observation period, and definitions of both numerator and denominator. They should test whether measurement scales have similar meaning across groups before comparing average satisfaction or embeddedness. A theoretically credible international synthesis incorporates institutional variation as part of the explanation, rather than treating it as statistical noise to be averaged away.
Retention interventions also create distributive questions. Flexible work may advantage some occupations more than others; high-potential talent programs may omit workers whose contribution is less visible; attendance-based bonuses may penalize people managing chronic illness or care responsibilities. The evidence-based orientation of Allen et al. (2010) supports evaluating actual mechanisms, while Meyer et al. (2002) reminds researchers that staying may reflect attachment, obligation, or accumulated cost. An equitable evaluation examines both average treatment effects and whether certain groups are excluded, harmed, or compelled to remain under worse conditions. Privacy and consent matter when organizations use health and absence records. Administrative convenience cannot justify converting protected medical circumstances into individualized disciplinary predictions without appropriate safeguards and independent evaluation.
Critical Discussion of Mechanisms and Evidentiary Limits
The empirical literature converges on a limited set of defensible conclusions. Voluntary employee turnover is multiply determined; intentions, satisfaction, commitment, and labor-market opportunities are informative but incomplete predictors Griffeth et al. (2000), Rubenstein et al. (2018). Employee staying has positive attachments, relationship-specific investments, and constraint-based components Meyer et al. (2002), Mitchell et al. (2001). Absence follows its own recurrent temporal structure and can represent health protection, illness, or organizational withdrawal in different circumstances Harrison & Martocchio (1998), Johns (2011). These claims are mutually compatible, but they do not establish a single deterministic pathway from an unfavorable attitude to absence and then departure. The breadth of the evidence is a reason to formulate differentiated conditional explanations, not to assume that a universal withdrawal scale will summarize every process.
A first unresolved question concerns causal direction. Dissatisfaction may contribute to exit, yet impending exit can also change satisfaction ratings. Strained team conditions may produce absence and turnover, while previous departures intensify those very conditions. Statistical adjustments may control stable differences and observed contemporaneous covariates without eliminating feedback or unmeasured events. The process accounts in Lee & Mitchell (1994), Hom et al. (2017), and Steel & Lounsbury (2009) suggest multiple causal architectures with overlapping observable predictions. Discriminating among them requires repeated pre-event measures and interventions that modify one hypothesized mechanism while monitoring the others. Researchers should specify falsifiable contrasts: for example, whether a shock changes quit hazard after conditioning on prior satisfaction trajectory and objectively measured external opportunities.
A second problem is generalization from aggregate effect sizes. Meta-analysis identifies average patterns across a selected literature, and those patterns may depend on how studies defined voluntary exit, sampled occupations, measured attitudes, and reported findings. Rubenstein et al. (2018) explicitly examines moderators, and Hancock et al. (2013) shows why the consequences of collective turnover can vary with the organizational context. Publication bias, nonrepresentative samples, restricted tenure ranges, and unreported null results can influence a synthesis even when its statistical calculations are technically correct. A credible dossier therefore treats pooled estimates as structured evidence about a research population, not universal constants to be inserted into every firm's forecast. The most informative uncertainty analyses specify which policy decision could change if the underlying estimate were different.
A third limitation concerns the value assigned to retention. An intervention that reduces employee quits can be beneficial through continuity, trust, and professional development, but remaining can also reflect blocked opportunity, economic insecurity, or difficult-to-replace benefits. The distinction is visible in commitment theory Meyer et al. (2002) and embeddedness research Mitchell et al. (2001). The organizational perspective Dalton & Todor (1979) further demonstrates that turnover consequences must be evaluated rather than presumed. Good retention science treats voluntary worker mobility and employee well-being as legitimate endpoints alongside organizational cost. The question is whether working conditions support sustainable employment relationships with fair opportunities, not whether a numerical target has been met irrespective of the human process behind it.
A fourth limitation arises in policy extrapolation. The direct experimental result from Bloom et al. (2024) is unusually informative because it links a randomized hybrid-work offer to objective departures. It does not resolve the causal effect of remote work for workers who cannot work from home, the optimal number of home days in every sector, or effects over an unlimited period. The larger intervention evidence base includes many correlational and quasi-experimental analyses Allen et al. (2010). Replications should vary geography, role, tenure, management, and scheduling systems; they should measure absenteeism and health as well as quits. Combining detailed implementation records with trustworthy outcome data will be more valuable than repeating an intervention label that conceals major differences in policy design.
Future Research in the Artificial Era
The Artificial Era introduces new possibilities for measuring and reorganizing employment relationships, including automated scheduling, algorithmic matching, hybrid human–artificial work systems, and AI-supported workforce planning. These changes create research opportunities because job opportunities, task boundaries, performance evaluation, and the value of specific skills may shift together. The established foundations of turnover theory Hom et al. (2017) and work-design theory Bakker & Demerouti (2024) provide initial variables, but new claims about artificial systems require their own empirical evidence. Research can ask whether changes in task automation modify perceived job security, employee embeddedness, demand–resource balance, or actual exit hazards. It should observe organizational practices and worker outcomes rather than infer psychological consequences solely from the deployment of technology.
A promising design follows organizational units before and after a clearly defined automated-scheduling intervention, measuring schedule predictability, workload, sickness absence, turnover, and perceived procedural fairness. When phased introduction is plausibly exogenous, causal estimates may be feasible; when management selects teams for automation, selection must be addressed. The same principle applies to AI-mediated promotion, training, and performance management. The empirical lessons of Bloom et al. (2024) show the value of controlled policy contrasts, while Johns (2011) shows why attendance and health must be analyzed together. Research should also compare work outcomes for staff who use new tools with those assigned to different tasks after adoption, since exposure may vary with role and skill. Individual-level prediction without institutional and temporal context risks obscuring the actual mechanism.
Artificial systems can also support analysis of recurrent attendance and time-to-exit data, but their scientific value depends on valid outcome construction, temporal validation, and protection against unfair use. A predictive model trained on historical absence codes may reproduce prior inequities if protected leave or health-related accommodations were recorded as undesirable attendance. The measurement cautions in Johns (1994) apply regardless of algorithmic complexity, and the contextual variation documented by Rubenstein et al. (2018) limits untested transport of model coefficients. Researchers should evaluate calibration and error across occupations and protected circumstances, require substantive explanations for deployment decisions, and document permissible uses of sensitive records. A model that predicts resignation can inform aggregate staffing projections; it should not become a substitute for improving working conditions or respecting employee autonomy.
The longer-range question is how organizations maintain human competence, continuity, and fair career pathways as the division of tasks between Homo and Artificial changes. A change in demand for particular skills may increase internal transitions, alter the value of experience, and affect the feasibility of voluntary exit. These possibilities extend turnover research into questions of occupational restructuring and distributed work, but the relevant psychological findings remain to be established. Hom et al. (2017) illustrates the historical adaptability of turnover theory, and Allen et al. (2010) supplies the discipline of evidence-based interventions. A scientifically grounded program in the Artificial Era would preregister key outcomes when appropriate, preserve individual-level employment histories with robust privacy safeguards, compare alternative institutional designs, and report both worker welfare and organizational continuity.
Conclusion
Employee turnover, absenteeism, and retention form a coherent research subject when all three are treated as temporally situated observations of employment relationships. Turnover is a transition, absenteeism is a recurrent event or duration of missed scheduled work, and retention is continued membership in a defined cohort over a defined period. Their measurement requires explicit classification of voluntary and involuntary departures, distinct categories of absence, appropriate exposure denominators, and valid observation windows. Research from Porter & Steers (1973) to Hom et al. (2017) shows how the field developed from broad withdrawal explanations to dynamic, heterogeneous models of quitting and staying. The scientific understanding achieved through this history is that employment continuity emerges from several interacting processes with different time scales and institutional constraints.
The strongest theoretical integration combines the changing desirability and feasibility of exit, sequential and shock-driven decisions, affective and cost-based organizational commitment, job embeddedness, and the health-related attendance process. Mobley (1977) and Lee & Mitchell (1994) clarify alternative exit pathways; Meyer et al. (2002) and Mitchell et al. (2001) explain distinct forms of attachment and continued membership; Harrison & Martocchio (1998) and Johns (2011) require that absence be interpreted within temporal and health contexts. Each framework identifies a meaningful portion of the phenomenon. Their synthesis generates testable distinctions among reasons to stay, capacity to attend, and opportunities to leave without attributing every observed behavior to a single latent tendency to withdraw.
Empirical synthesis supports recurring associations between employee attitudes and actual turnover, variation in the organizational consequences of collective departures, and the importance of context and measurement Griffeth et al. (2000), Rubenstein et al. (2018), Hancock et al. (2013). The randomized trial by Bloom et al. (2024) additionally shows that a specified hybrid-work policy can causally improve retention in the studied workforce without the performance deterioration some feared. This result establishes the value of direct intervention evidence and careful generalization. Research quality depends on verified bibliographic attribution, appropriate designs, transparent measures, valid statistical contrasts, and honest treatment of missing data and causal uncertainty. Retention strategies should be evaluated by observable outcomes and human consequences, including health, fairness, and career agency.
Future inquiry should integrate reliable employment panels, event histories, repeated measures of health and attitudes, natural experiments, and randomized organizational policies. The Artificial Era extends this agenda as new technologies change scheduling, work organization, opportunities, and the continuity of skills. The decisive scientific task remains stable across these transformations: establish what happened, identify the mechanism plausibly producing it, distinguish association from intervention effect, and evaluate the conditions under which sustainable employment relationships are achieved. A fundamental monograph on employee turnover, absenteeism, and retention therefore provides a framework for understanding the organization of work as a dynamic relationship among persons, institutions, health, and changing opportunities.
References
- Allen, D. G., Bryant, P. C., & Vardaman, J. M. (2010). Retaining talent: Replacing misconceptions with evidence-based strategies. Academy of Management Perspectives, 24(2), 48–64. https://doi.org/10.5465/AMP.2010.51827775
- Bakker, A. B., & Demerouti, E. (2024). Job demands–resources theory: Frequently asked questions. Journal of Occupational Health Psychology, 29(3), 188–200. https://doi.org/10.1037/ocp0000376
- Bloom, N., Han, R., & Liang, J. (2024). Hybrid working from home improves retention without damaging performance. Nature, 630, 920–925. https://doi.org/10.1038/s41586-024-07500-2
- Crawford, E. R., LePine, J. A., & Rich, B. L. (2010). Linking job demands and resources to employee engagement and burnout: A theoretical extension and meta-analytic test. Journal of Applied Psychology, 95(5), 834–848. https://doi.org/10.1037/a0019364
- Dalton, D. R., & Todor, W. D. (1979). Turnover turned over: An expanded and positive perspective. Academy of Management Review, 4(2), 225–235. https://doi.org/10.5465/amr.1979.4289021
- Griffeth, R. W., Hom, P. W., & Gaertner, S. (2000). A meta-analysis of antecedents and correlates of employee turnover: Update, moderator tests, and research implications for the next millennium. Journal of Management, 26(3), 463–488. https://doi.org/10.1177/014920630002600305
- Hancock, J. I., Allen, D. G., Bosco, F. A., McDaniel, K. R., & Pierce, C. A. (2013). Meta-analytic review of employee turnover as a predictor of firm performance. Journal of Management, 39(3), 573–603. https://doi.org/10.1177/0149206311424943
- Harrison, D. A., & Martocchio, J. J. (1998). Time for absenteeism: A 20-year review of origins, offshoots, and outcomes. Journal of Management, 24(3), 305–350. https://doi.org/10.1177/014920639802400303
- Hausknecht, J. P., Rodda, J., & Howard, M. J. (2009). Targeted employee retention: Performance-based and job-related differences in reported reasons for staying. Human Resource Management, 48(2), 269–288. https://doi.org/10.1002/hrm.20279
- Hom, P. W., Lee, T. W., Shaw, J. D., & Hausknecht, J. P. (2017). One hundred years of employee turnover theory and research. Journal of Applied Psychology, 102(3), 530–545. https://doi.org/10.1037/apl0000103
- Johns, G. (1994). How often were you absent? A review of the use of self-reported absence data. Journal of Applied Psychology, 79(4), 574–591. https://doi.org/10.1037/0021-9010.79.4.574
- Johns, G. (2011). Attendance dynamics at work: The antecedents and correlates of presenteeism, absenteeism, and productivity loss. Journal of Occupational Health Psychology, 16(4), 483–500. https://doi.org/10.1037/a0025153
- Lee, T. W., & Mitchell, T. R. (1994). An alternative approach: The unfolding model of voluntary employee turnover. Academy of Management Review, 19(1), 51–89. https://doi.org/10.5465/amr.1994.9410122008
- Lee, T. W., Mitchell, T. R., Holtom, B. C., McDaneil, L. S., & Hill, J. W. (1999). The unfolding model of voluntary turnover: A replication and extension. Academy of Management Journal, 42(4), 450–462. https://doi.org/10.5465/257015
- Meyer, J. P., Stanley, D. J., Herscovitch, L., & Topolnytsky, L. (2002). Affective, continuance, and normative commitment to the organization: A meta-analysis of antecedents, correlates, and consequences. Journal of Vocational Behavior, 61(1), 20–52. https://doi.org/10.1006/jvbe.2001.1842
- Mitchell, T. R., Holtom, B. C., Lee, T. W., Sablynski, C. J., & Erez, M. (2001). Why people stay: Using job embeddedness to predict voluntary turnover. Academy of Management Journal, 44(6), 1102–1121. https://doi.org/10.5465/3069391
- Mitra, A., Jenkins, G. D., Jr., & Gupta, N. (1992). A meta-analytic review of the relationship between absence and turnover. Journal of Applied Psychology, 77(6), 879–889. https://doi.org/10.1037/0021-9010.77.6.879
- Mobley, W. H. (1977). Intermediate linkages in the relationship between job satisfaction and employee turnover. Journal of Applied Psychology, 62(2), 237–240. https://doi.org/10.1037/0021-9010.62.2.237
- Porter, L. W., & Steers, R. M. (1973). Organizational, work, and personal factors in employee turnover and absenteeism. Psychological Bulletin, 80(2), 151–176. https://doi.org/10.1037/h0034829
- Rubenstein, A. L., Eberly, M. B., Lee, T. W., & Mitchell, T. R. (2018). Surveying the forest: A meta-analysis, moderator investigation, and future-oriented discussion of the antecedents of voluntary employee turnover. Personnel Psychology, 71(1), 23–65. https://doi.org/10.1111/peps.12226
- Skagen, K., & Collins, A. M. (2016). The consequences of sickness presenteeism on health and wellbeing over time: A systematic review. Social Science & Medicine, 161, 169–177. https://doi.org/10.1016/j.socscimed.2016.06.005
- Steel, R. P., & Lounsbury, J. W. (2009). Turnover process models: Review and synthesis of a conceptual literature. Human Resource Management Review, 19(4), 271–282. https://doi.org/10.1016/j.hrmr.2009.04.002