RESEARCH DOSSIER
Occupational Safety, Accidents, and Safety Climate
Abstract. Occupational safety is an emergent property of organized work shaped by hazardous exposure, technical safeguards, institutional decisions, supervisory priorities, shared expectations, reporting practices, and human performance. This fundamental research monograph examines the history and scientific definitions of occupational accidents and safety climate, explains the psychological mechanisms linking organizational conditions with compliance and participation, and critically assesses original field investigations, intervention studies, psychometric research, and meta-analyses. The evidence synthesis distinguishes individual perceptions from group and organizational safety climate; recorded incidents from actual harm; and safety-related behavior from reliable prevention. It examines multilevel measurement, underreporting, longitudinal causal inference, protective controls, leadership, training, regulatory surveillance, and global inequality in occupational exposure. A further analysis considers safety-critical automation and human–artificial coordination in the Artificial Era. The central conclusion is that a valid account of occupational safety requires convergent evidence on work-system hazards, functioning barriers, worker experience, behavioral practices, and injury outcomes. The present work is a critical narrative scientific synthesis, not a registered systematic review, meta-analysis, original experiment, or jurisdiction-specific legal or clinical guideline.
Keywords: occupational safety; workplace accidents; safety climate; organizational safety culture; safety performance; safety compliance; safety participation; occupational injury; near misses; safety reporting; safety leadership; safety management systems; human factors; hierarchy of controls; safety behavior; automation; human–artificial relations.
Introduction and Scientific Scope
Occupational safety is a property of organized work: the capacity of a work system to prevent, detect, control, and recover from conditions that could harm people. An injury is an outcome experienced by a person, an accident is an event or event sequence with harmful consequences, an incident is a broader operational occurrence, and a near miss is a hazardous event in which injury or loss did not materialize. A hazard is a source of potential harm; risk expresses the likelihood and severity of harm under specified exposure conditions. The distinction matters because a system can contain serious hazards without recording an injury during a short observation period, and the absence of reported injuries can coexist with substantial unrecognized risk.
The psychology of occupational safety asks how people perceive hazards, allocate attention, learn procedures, weigh production demands against protective action, trust supervisors, communicate concerns, participate in prevention, and respond to organizational incentives. It also asks how leadership and work arrangements create the conditions under which protective behavior becomes feasible and consequential. Safety climate is a central construct in this investigation: the shared perception, among members of a work unit or organization, of the priority actually assigned to safety through policies, procedures, practices, and daily decisions. The practical emphasis on enacted priorities makes the difference between a written safety commitment and an operational culture observable.
The research object combines hazard exposure, accident processes, occupational injury, safety compliance, safety participation, safety reporting, and safety climate. It treats these as connected yet separately measurable phenomena. The organizing question is how work conditions and social expectations influence behavior and how these processes interact with technical barriers to shape injury outcomes. Workplace health, physical ergonomics, safety engineering, labor organization, and organizational psychology provide complementary evidence, while the central explanatory problem remains psychological and organizational.
The scientific tradition includes original observational studies, field interventions, multilevel survey investigations, theory-building, meta-analyses, accident reconstructions, and institutional prevention guidance. The synthesis draws on foundational studies by Zohar, D (1980), Hofmann, D. A., & Stetzer, A (1996), Griffin, M. A., & Neal, A (2000), and Zohar, D., & Luria, G (2005), together with the reviews of Christian, M. S., Bradley, J. C., Wallace, J. C., & Burke, M (2009), Beus, J. M., Payne, S. C., Bergman, M. E., & Arthur, W., Jr (2010), and Nahrgang, J. D., Morgeson, F. P., & Hofmann, D. A (2011). Where only indexed abstracts or publisher summaries were accessible, interpretations are limited to what those records actually establish. This manuscript is a critical narrative scientific synthesis, with an explicit evidence appraisal; it is neither a registered systematic review nor an original experiment.
Historical Development of Occupational Accident Research
The earliest occupational safety questions developed around industrial injury, dangerous machinery, working hours, supervision, and the organization of hazardous production. Their long history includes mechanical safeguards, occupational hygiene, inspection and regulation, and attempts to characterize accident-prone behavior. The modern scientific field gradually moved toward investigating the conditions under which actions occur. That shift matters because the same human action can have radically different consequences depending on equipment, staffing, barriers, recoverability, and the timing of exposure.
A major intellectual development was the interpretation of accidents as products of interacting defenses, organizational decisions, local adaptation, and failures of control. Reason, J (1997) articulated active failures, latent organizational conditions, and the significance of multiple protective barriers. The layered-defense representation illustrates how a hazard can pass through several imperfect safeguards. It is a heuristic for investigating vulnerabilities; the existence of a plausible defensive pathway does not by itself establish the causal sequence of a specific accident.
Rasmussen (1997)00052-0) situated risk management within a dynamic sociotechnical network extending from regulation and management to work planning and operations. His argument emphasizes the pressure of economic change, productivity demands, local decision criteria, and migration toward operational boundaries. The approach directs attention to the way several reasonable decisions, made at different levels and times, can converge to create a hazardous configuration. It also expands the unit of analysis beyond the individual operator.
Leveson (2004)00047-X) developed a systems-theoretic accident framework treating safety as the enforcement of constraints within a control structure. The central object is the adequacy of control, information, feedback, and intercomponent coordination. This account is especially pertinent to complex software-dependent and organizational systems whose failures may not resemble a simple sequence of broken parts.
Organizational psychology simultaneously established a complementary research line. Zohar, D (1980) introduced a seminal empirical account of safety climate in industrial organizations, redirecting inquiry toward shared perceptions of how management prioritizes safety. The subsequent work of Hofmann, D. A., & Stetzer, A (1996) tested individual and team-level predictors in a chemical-processing environment. Griffin, M. A., & Neal, A (2000) made a durable contribution by separating safety compliance from safety participation and treating knowledge and motivation as possible links between climate and behavior.
The historical development has thus generated at least three interacting explanatory traditions: organizational behavior and climate; human factors and control-system design; and occupational epidemiology and prevention. Their combination is necessary whenever an observed outcome involves personal conduct, hazardous exposure, and the effectiveness of protective infrastructure. Historical prominence does not guarantee that every element of an early theory has contemporary empirical confirmation; each mechanism requires evidence appropriate to its level of analysis.
Defining Safety Climate, Safety Culture, and Safety Performance
Safety climate refers to shared perceptions of actual safety-related policies, procedures, and practices, including the priority assigned to safety when it competes with time, cost, and production goals. An individual employee's perception is sometimes called psychological safety climate; a group-level safety climate requires evidence of shared perception and meaningful differences among units. Safety culture is a broader construct concerning shared assumptions, institutional history, symbols, values, routines, and practices. Surveying employee perceptions can reveal part of a culture, but a climate questionnaire alone does not exhaust the culture of an organization.
Safety performance also demands precise definition. Safety compliance includes behaviors necessary to satisfy prescribed safety requirements, such as following a procedure or using an appropriate protection. Safety participation includes discretionary contributions that sustain the collective safety system, such as identifying a hazard, suggesting a correction, speaking up about an unsafe condition, supporting a coworker, or participating in a review. Griffin, M. A., & Neal, A (2000) investigated these performance dimensions and the proposed roles of knowledge and motivation. Participation can be particularly sensitive to whether workers expect their contributions to matter.
The construct of safety climate must also be distinguished from psychological safety in the organizational-behavior sense: a shared belief that interpersonal risk-taking, including asking questions or admitting mistakes, will be tolerated. Psychological safety can facilitate hazard voice and near-miss reporting, but a team could encourage frank discussion while lacking adequate physical safety systems; conversely, a highly regulated workplace could possess engineering safeguards while discouraging open reporting. The two constructs become scientifically useful when separately operationalized and tested rather than treated as interchangeable survey labels.
Safety outcomes have different temporal and causal locations. Exposure frequency, near misses, hazardous acts, injuries, and fatalities cannot be combined indiscriminately into one performance score. An increase in disclosed near misses after introduction of a trusted reporting channel may represent improved surveillance. A reduction in recorded injuries after management ties bonuses to low accident counts may partly represent suppressed reporting. Defining the endpoint before assessment is therefore a condition of meaningful analysis.
The Work System as the Unit of Safety Analysis
An occupational injury is never adequately characterized by personal intention alone. A worker may be physically exposed because of equipment layout, staffing, time pressure, contractor coordination, maintenance backlog, insufficient guarding, or ineffective supervision. The relevant psychological variables include perceived behavioral options, anticipated consequences of slowing work, social norms, and confidence that help will be available. Technical controls alter the range of hazardous actions that remain possible; organizational conditions alter the likelihood that workers will encounter those hazards.
The work-system approach entails a hierarchy of explanations. At the immediate task level, attention, fatigue, perception, memory, and procedural decisions shape performance. At the team level, supervision, communication, coordination, and informal norms shape how requirements are interpreted. At the organizational level, resources, scheduling, hiring, training, incentives, and safety management determine persistent exposure patterns. At the institutional level, standards, enforcement, economic relationships, and procurement can alter the constraints under which organizations operate.
This hierarchy should be handled through explicit research design rather than rhetorical attribution. A single worker interview cannot directly identify an organization-level causal effect. Conversely, an organizational injury rate cannot tell investigators which cognitive operation failed in a particular episode. Multilevel evidence becomes valuable because it relates constructs at their appropriate levels. In a field study of Hofmann, D. A., & Stetzer, A (1996) associated group processes, climate, role overload, and unsafe behaviors; its single-site setting places clear limits on generalization.
Sociotechnical explanations also clarify why safety programs should assess the functioning of protective barriers. The National Institute for Occupational Safety and Health (n.d.) orders hazard elimination, substitution, engineering controls, administrative controls, and personal protective equipment by general preventive effectiveness. The order recognizes that preventing exposure at its source commonly demands less continual vigilance from individual workers than controls that depend on perfect repeated compliance. Psychological interventions matter, yet they operate within this larger protective architecture.
Accident Causation, Hazard Exposure, and Multiple Defenses
An accident model specifies how hazardous conditions, triggering events, decisions, and failures of control may produce harm. Linear models organize a sequence of proximate causes; barrier models trace the failure of safeguards; systems-theoretic models investigate whether necessary safety constraints were effectively enforced. These representations serve different analytical purposes and produce different questions for evidence gathering. A useful investigation compares explanations with records of what people knew, what equipment did, what constraints existed, and which defenses were available.
Hazard identification begins before a harmful event. It may include task observation, worker reports, engineering assessments, maintenance inspections, process-hazard analyses, or occupational exposure monitoring. Exposure constitutes a link between hazard and potential injury: frequency, duration, proximity, severity, and control reliability matter. Because employees may adapt to repeated hazards, perceptions of familiarity and controllability deserve investigation as potentially important moderators of protective action.
Reason, J (1997) examined how latent conditions and imperfect defenses interact. Rasmussen (1997)00052-0) added the dynamics of organizational decisions under pressure, and Leveson (2004)00047-X) analyzed inadequate control constraints. Together these works establish a research agenda that looks beyond the final action before an accident. They do not imply that every event contains an identical number of causal stages or that a conceptual model can replace site-specific evidence.
Near-miss analysis creates a valuable opportunity to examine hazards without waiting for injury, but its results depend on the visibility of events and willingness to disclose them. A near miss should be analyzed for the mechanism that prevented harm as well as the mechanism that permitted the hazardous exposure. Learning from successful recovery can reveal barrier resilience and worker expertise. An incident investigation is most productive when it asks what information and options were available at each consequential moment and how the organization subsequently changed its safeguards.
Organizational Priorities and the Safety–Production Relationship
Safety-related judgments occur within a system of competing goals. Workers are often expected to maintain output, meet schedules, conserve resources, and remain responsive to customers while protecting themselves and others. The priority assigned to each goal is communicated most convincingly by the organization's response when goals conflict. If a supervisor praises completing a task quickly despite unsafe conditions, employees receive a practical signal about priorities. If work is stopped until a dangerous defect is corrected, the organization communicates a different expectation.
The Zohar, D., & Luria, G (2005) multilevel model demonstrated the importance of both organization-wide and local group climates and identified supervisory discretion as one source of variation among groups. A global policy may therefore coexist with different local operating priorities. A safety assessment should ask which supervisor decisions workers encounter, whether rules can be followed under routine time constraints, and how conflicts are resolved.
Production and safety are coupled through work design. Improving equipment reliability can increase productive capacity while removing a dangerous improvisation; conversely, an unrealistic timetable can drive shortcuts even among workers who recognize the hazard. The appropriate empirical question is which concrete resource or procedural changes improve exposure control, safety behavior, and task effectiveness. A simple measure of whether employees personally claim to value safety has limited explanatory power unless connected to enacted practices and objective outcomes.
Organizational priorities may be especially difficult to infer in subcontracted, temporary, seasonal, and platform-mediated work. Workers can receive instructions from one party, equipment from another, and disciplinary incentives from a third. Safety climate must then be specified at the relevant level: employer, site, subcontractor, shift, crew, or platform. The same person's survey response can refer to different authority structures, and aggregation can conceal consequential differences.
Behavioral Mechanisms of Safety Knowledge and Motivation
Safety knowledge includes recognition of hazards, understanding of effective controls, and knowledge of the conditions in which a procedure applies. Safety motivation concerns the importance a worker assigns to protective action and the expected consequences of performing it. Both constructs mediate how environmental messages may influence observed behavior, but neither directly guarantees safety if tools, staffing, protective equipment, or operational authority are missing.
Griffin, M. A., & Neal, A (2000) identified knowledge and motivation as relevant intervening variables between climate and reports of safety performance. Christian, M. S., Bradley, J. C., Wallace, J. C., & Burke, M (2009) quantitatively synthesized person- and situation-related predictors and found that knowledge and motivation were strongly connected with safety behavior, with group safety climate especially relevant to accidents and injuries. The synthesis supports the explanatory importance of the work environment alongside personal characteristics.
Motivation can be supported through meaningful participation, credible feedback, well-designed tools, and trustworthy leadership. It can be weakened by job insecurity, contradictory incentives, and the expectation that reporting hazards will bring punishment. A longitudinal investigation by Probst and Brubaker (2001) linked greater job insecurity with lower safety motivation and compliance and with higher reported accident and injury outcomes among food-processing employees; such associations warrant careful attention to possible confounding and the direction of effects.
Knowledge and motivation should be distinguished from availability of a safe course of action. A worker who knows the hazard and wants to avoid it may still have no practical access to adequate lifting equipment or reliable isolation procedures. A model that attributes all unsafe behavior to insufficient motivation silently assumes that the organization has already supplied adequate controls. This assumption must be tested before individual-level interventions are justified.
Safety Climate as a Multilevel Construct
Safety climate begins with individual observation but becomes an organizational construct only when workers within a specified unit display sufficiently shared perceptions. A respondent who reports strong management commitment may be describing a supervisor, an employer, a client organization, or a particular site. Researchers therefore need an explicit referent and an analysis plan that distinguishes individuals, workgroups, locations, and organizations. The shared-climate hypothesis predicts patterns of agreement and between-unit variation; it should be examined empirically through appropriate aggregation diagnostics and multilevel models.
Zohar, D., & Luria, G (2005) tested a multilevel account in which organizational and group climates were connected but distinct. Their reported results indicated that group climate mediated the organization-climate association with safety behavior, while supervisory discretion helped explain differences between groups operating under the same organization. This finding makes local supervision a plausible mechanism through which organizational promises become practical routines. An organization-wide average can conceal precisely the variation that explains the working conditions most consequential for injury.
Climate strength describes how consistently members of a group perceive a priority, whereas climate level describes its average direction or magnitude. A group could be uniformly skeptical that management values safety, yielding a strong but unfavorable climate. Another group could show a favorable mean with substantial disagreement between day and night shifts. For this reason, a climate score alone should not be used as a complete organizational diagnosis; the distribution of responses and subgroup comparisons may be informative.
Safety climate is dynamic. A new leader, serious incident, equipment change, workforce reorganization, or production crisis can change the experiences through which employees infer management priorities. Longitudinal measurement should therefore identify a temporal window, assess whether the meaning of survey items remains stable, and consider seasonality or changes in personnel composition. A sudden climate improvement after an incident may reflect genuine reforms, heightened attention to safety, selection effects, or employees' expectations about the survey sponsor.
Leadership, Supervisory Discretion, and Enacted Safety Priorities
Supervisors translate high-level requirements into tasks, equipment allocations, feedback, and judgments about acceptable delay. A supervisor's safety leadership becomes observable when workers report hazards, when a schedule becomes difficult, when resources are scarce, or when an incident requires investigation. Safety-specific communication is more informative than generic leadership praise because it refers directly to the choices through which hazardous exposure is created or controlled.
In an influential field intervention, Zohar, D (2002) used repeated interviews with subordinates and feedback to supervisors to increase safety-oriented interactions. The indexed report describes significant improvements in those interactions in intervention groups relative to controls, accompanied by changes in minor-injury rates, hearing-protection behavior, and climate measures. The experiment provides more direct intervention evidence than a cross-sectional correlation, although interpretation still depends on site conditions, the particular intervention components, and the measured outcome window.
Mullen, J. E., & Kelloway, E. K (2009) evaluated safety-focused transformational leadership training with randomly assigned leaders from long-term care organizations. Their study connected leader training to changes in some safety-related supervisor and employee perceptions. This evidence supports the feasibility of changing leadership-related variables, while the scope of observed improvements should be read from the specific study results rather than generalized to all injury outcomes.
A recent synthesis by Li, Z., Tang, J., Yang, X., & Zhang, J (2026) analyzed 102 studies involving 107,109 participants and reported associations between a variety of leadership styles and safety performance, with substantial heterogeneity and cultural, methodological, and risk-context moderators. These findings strengthen the case for studying leadership in context. They do not convert all leadership associations into experimental proof or establish that training alone replaces engineering controls.
Supervision includes correction and enforcement, but effective enforcement is not reducible to punishment. Supervisors may increase safety by securing equipment, removing contradictory demands, clarifying expectations, and providing usable feedback. Punitive responses may reduce unsafe actions in some circumstances while simultaneously undermining incident disclosure. Research designs should therefore assess safety behavior, hazard exposure, reporting completeness, and trust separately.
Job Demands, Resources, Burnout, and Engagement
A dangerous task is experienced through physical exposure and through cognitive and emotional conditions. Workload, sustained time pressure, insufficient rest, role overload, unpredictable scheduling, conflicting instructions, and low autonomy can change the capacity to recognize and respond to hazards. Resource availability, supervisor support, appropriate staffing, training, and accessible equipment may alter the same process. The job demands–resources framework supplies one way to organize these interacting conditions.
In a meta-analysis of Nahrgang, J. D., Morgeson, F. P., & Hofmann, D. A (2011) examined connections among demands, resources, burnout, engagement, and safety outcomes. The analysis supported both a health-impairment pathway involving demanding conditions and burnout and a motivational pathway involving resources and engagement. Risk and hazard exposures were consistently important demands across industries, while supportive environments emerged as a consistently important resource. These are broad associations, with interpretation dependent on included designs and operational definitions.
The distinction between demands and resources helps identify interventions beyond training. When employees are cognitively overloaded, a warning campaign may add another demand; revised scheduling, engineered safeguards, and better task coordination may produce more substantial changes. When knowledge is absent, instruction may be essential, but its effect depends on opportunities to apply it. The relevant intervention target is the specific mechanism supported by evidence about that workplace.
Safety and psychological well-being share some determinants but are different outcomes. An employee can experience burnout without a recorded injury, and a low-burnout workplace can still contain catastrophic technical vulnerabilities. Correlation between well-being and safety behavior does not justify using well-being scores as a substitute for hazard assessment. Multimethod studies should measure exposures, behavior, and injuries independently wherever feasible.
Individual Differences, Experience, and Risk Perception
Workers differ in experience, training, sensory abilities, physical capacities, fatigue, familiarity with hazards, and expectations about the consequences of reporting danger. These differences can alter perception and action, but their scientific interpretation requires attention to job assignment, exposure, and equipment. A group of experienced workers might report fewer incidents because of superior procedural knowledge, because they are assigned safer tasks, or because they underreport minor problems. Injury rates alone cannot adjudicate these explanations.
Risk perception includes beliefs about the likelihood of harm, severity of consequences, personal controllability, and social acceptability of exposure. Familiarity may support accurate recognition of failure modes, yet repeated uneventful exposure can also normalize risk. A scientifically useful assessment distinguishes perceived probability from measured frequency and distinguishes willingness to take a risk from perceived lack of alternatives. Interventions that increase fear without giving workers effective protective options may impose psychological burden with limited operational benefit.
Training can improve knowledge and procedural skill, but its effect must be tested for retention and transfer to actual work. A classroom post-test measures immediate recall; field observation and exposure outcomes examine whether learning changes the system of action. Training quality depends on practice opportunities, credible feedback, realistic scenarios, accommodations, and time to execute procedures under normal production conditions.
Selection effects complicate occupational research. Employees who remain in dangerous jobs may differ from those who leave because of injury, health, risk tolerance, or job opportunity. This healthy-worker and survivor dynamic can make highly exposed groups appear deceptively resilient. Longitudinal designs should document workforce entry, transfer, and exit, not only the workers still present at the final survey.
Accident Reporting, Near Misses, and the Psychology of Disclosure
Reports of injuries and near misses are jointly produced by events and by reporting conditions. Employees make decisions about whether an event qualifies, whether a supervisor will respond fairly, whether medical care or lost time will trigger formal processes, and whether reporting will threaten income or employment. Therefore, recorded incident counts measure both harm and surveillance. A prevention program should seek more complete and accurate reporting even when an immediate consequence is a rise in reported events.
Probst, T. M., Brubaker, T. L., & Barsotti, A (2008) compared contractor OSHA injury records with insurance-claims information in a construction setting, providing an empirical basis for studying underreporting. Probst, T. M., & Estrada, A. X (2010) examined accident underreporting in a multisector sample and investigated the roles of safety climate and supervisory enforcement. A later Probst, T. M (2015) multilevel study across 33 organizations and 1,238 employees found supervisor enforcement associated with underreporting and a cross-level interaction involving climate. These findings show that the measurement system is partly an organizational-behavior system.
A good reporting design makes the event definition intelligible, provides accessible channels, protects workers against retaliation, acknowledges reports, and demonstrates corrective action. The Occupational Safety and Health Administration (n.d.) emphasizes timely responses, feedback, meaningful participation, and the removal of barriers to reporting. These recommendations are a practical institutional synthesis; their precise legal obligations vary by jurisdiction and should not be mistaken for a worldwide legal code.
Near-miss indicators require careful interpretation. If reporting becomes easier, their recorded frequency may rise while actual risk declines. If workers lose trust, recorded frequency may fall while hazardous exposure is unchanged. Analysis should triangulate reporting with direct hazard observations, maintenance records, independent audits, and injury data. The goal is organizational learning, not maximizing or minimizing the raw number of submitted forms.
Incentives, Job Insecurity, and Unintended Consequences
Safety policies alter behavior partly through incentives. A reward for active hazard identification may increase participation when workers trust the response system. An award solely for achieving a low recorded injury count can make injuries costly to report, thereby contaminating the principal safety measure. Incentives also interact with job insecurity, because employees may interpret a report as evidence that they are troublesome or incapable.
Probst and Brubaker (2001) used cross-sectional and longitudinal analyses in a food-processing workforce to explore job insecurity and safety outcomes. Their findings motivated subsequent investigation of reporting and enforcement. The broader implication is that a workplace's treatment of employees and their economic security can influence protective behavior independently of what official policies say.
A causal evaluation of incentive systems should predefine two types of outcomes: safety effectiveness, such as verified exposure reduction or injury prevention, and reporting integrity, such as correspondence between reported incidents and independent data. A program that improves one while degrading the other requires scrutiny. Organizational leaders should also examine whether incentive thresholds shift behavior near reporting deadlines or encourage the classification of injuries as first-aid cases.
The research question is not merely whether workers follow a rule, but whether the entire motivational system makes safe work feasible, credible, and worthwhile. Resources, procedures, voice protections, and management accountability combine to determine that system. Any claim that an incentive reduced injury causally must be evaluated against contemporaneous changes in exposure, workload, workforce composition, and surveillance.
Measurement of Safety Climate and Psychometric Validity
A climate instrument is valuable when it measures a clearly identified shared perception and supports the inference the researcher intends to draw. Questionnaire items commonly ask workers about management commitment, supervisor conduct, peer norms, training, feedback, rule enforcement, and opportunities to participate. Items must name the referent consistently: “my supervisor,” “this workgroup,” and “the organization” describe different levels of social experience. Pooling them into a single score without theoretical justification can obscure the mechanism of interest.
Construct validity begins with item content. A question about whether the workplace possesses protective equipment may index material availability, while a question about whether management promptly replaces defective equipment refers more directly to enacted priority. Both matter for safety, but their meanings should be stated and their overlap examined. Validity also includes response processes: workers should interpret a question in the intended way, including across language groups, occupations, and shifts. Factor analysis can evaluate proposed dimensional structure, while reliability estimates address measurement consistency without demonstrating that the instrument predicts injuries.
The original Zohar, D (1980) research used workers’ perceptions to investigate safety climate as an organizational attribute. Subsequent researchers developed alternative measures because safety systems vary across industries and levels of organization. Kines, P., Lappalainen, J., Mikkelsen, K. L., Olsen, E., Pou (2011) introduced the Nordic Occupational Safety Climate Questionnaire, NOSACQ-50, as a multidimensional instrument. Its dimensions address managerial and worker safety-related commitments, communication, learning, and trust. An established questionnaire offers comparative structure, but adoption of a published scale does not remove the obligation to validate its meaning for a particular language, industry, and research objective.
Scores are affected by survey administration. Employees might answer differently when anonymity is credible, when managers are present, when a serious accident has just occurred, or when the questionnaire is presented as part of a compliance inspection. Nonresponse can be informative if workers at highest risk or with lowest trust are least willing to participate. Valid reporting therefore includes the sampling frame, participation rate, administration conditions, missing data, item wording, and intended level of inference.
Survey Aggregation, Measurement Invariance, and Climate Strength
Group climate can be modeled when employees' individual responses reflect a sufficiently shared unit-level phenomenon. Statistical justification may involve within-group agreement, intraclass correlations, and between-group variance. The particular indices must suit the sampling design and the form of the construct. A high average item score does not establish consensus; similarly, an organization-wide average may conceal workgroups that experience different supervisory priorities.
Zohar, D., & Luria, G (2005) provide a substantive reason to preserve multiple levels of analysis: local climate can differ because supervisors translate broad policy into different practices. A well-designed study might therefore ask each employee about the workgroup, aggregate only where agreement and sampling are adequate, and separately assess organization-level policy implementation. It can then investigate whether workgroup climate predicts safety behavior after accounting for individual perceptions, exposure, and workforce composition.
Cross-cultural research introduces further complexity. Two translated questionnaires may use equivalent words while measuring different authority relationships or reporting expectations. Measurement invariance examines whether item parameters permit valid comparisons of latent constructs across groups. Without appropriate evidence, an apparent difference between countries, languages, age groups, or occupational sectors might partly reflect the measurement instrument. Content adaptation should be documented rather than hidden behind a claim that a scale is universally standardized.
Climate strength is also scientifically important. Researchers can study the spread of responses alongside their mean, since disagreement may reveal subunits with divergent experiences or uncertainty about actual management priorities. Multilevel analysis should avoid inflating the number of independent observations by treating many employees from one organization as though they were separately randomized organizations.
Measuring Injuries, Accidents, Near Misses, and Exposure
A safety outcome requires a case definition, ascertainment procedure, observation window, population denominator, and exposure denominator where relevant. Injuries may be classified by severity, medical treatment, restricted work, lost time, or fatalities. Incident data may be collected from employer logs, insurance records, occupational health services, regulator reports, or worker surveys. These sources do not necessarily agree, because they differ in what is captured and why an event becomes visible.
For comparisons across sites and years, absolute counts are often insufficient. A larger workforce or a higher number of hours worked can generate more recorded incidents even when the probability per unit exposure decreases. Rates per worker, hours worked, task hours, or hazardous exposures address different estimands. Reporting should state which denominator was used, how contract staff were included, and whether shifts or overtime changed. Small counts of severe events produce unstable rates and wide uncertainty intervals.
Underreporting is an independent measurement problem. Probst, T. M., Brubaker, T. L., & Barsotti, A (2008) compared different record systems to examine injury-rate underreporting, while Probst, T. M., & Estrada, A. X (2010) studied employees’ disclosure behavior. These investigations illustrate why an association between safety climate and recorded injuries may reflect both injury generation and the likelihood that an injury is recorded. A robust study triangulates sources where feasible and treats disagreement between sources as evidence to investigate.
Near misses, unsafe conditions, and observed rule deviations offer information about system operation before severe harm occurs. Their event frequencies should nonetheless be interpreted alongside surveillance effort. More site inspections will often identify more hazards; a more trusted reporting culture may produce more submitted near misses. A decrease in counts can therefore represent either genuine improvement or reduced detection. An organization needs contextual information about observation coverage and the severity of hazards corrected.
Leading and Lagging Indicators
Lagging indicators describe outcomes that have already occurred, such as injury or occupational illness. Leading indicators describe activities, conditions, or capacities expected to influence future safety, such as timely resolution of hazards, verification of critical controls, employee participation, preventive maintenance, or effective supervision. Occupational Safety and Health Administration (n.d.) presents this distinction as a practical way to improve prevention.
An indicator becomes meaningful through a stated causal pathway. Counting completed training sessions may measure organizational activity, but a training session is a weak proxy for competence unless employees can demonstrate knowledge and apply it in realistic conditions. Counting safety inspections may similarly reward paperwork unless the inspections locate important hazards and lead to verified corrections. More informative leading indicators include the proportion of high-consequence barriers tested, the age and severity of unresolved defects, and follow-up quality after hazard reports.
Leading indicators should be selected to cover important failure mechanisms rather than to maximize the number of easily reported metrics. A control dashboard can conceal risk when favorable averages are driven by low-risk activities. For example, excellent completion of routine safety briefings tells little about whether a rarely used emergency shutdown system works. The indicator portfolio should distinguish prevention capacity from administrative output.
Predictive validation is necessary before an indicator is described as forecasting harm. The researcher should specify a prediction horizon, establish temporal precedence, compare with baseline models, and evaluate performance on data outside the development sample. Injuries and fatalities are often rare relative to routine observations, so discrimination, calibration, false alarms, and the consequences of intervention require attention. A leading indicator can still be useful for management when it has a defensible relation to risk reduction, even without a stable numerical prediction of future injury.
Longitudinal, Multilevel, and Causal Research Designs
Cross-sectional climate surveys provide information about associations at a particular time. If workers who perceive a stronger safety climate also report greater compliance, the result is compatible with a climate-to-behavior pathway, but also with reverse causation, common-method variance, and unmeasured shared conditions. Workers who recently experienced an injury may report a more negative climate; a workplace with better equipment may produce both favorable climate responses and safer behavior. Statistical adjustment can reduce some confounding but cannot by itself establish that all alternative explanations have been eliminated.
Beus, J. M., Payne, S. C., Bergman, M. E., & Arthur, W., Jr (2010) synthesized safety climate and injury relationships and discussed theoretical directions and reciprocal connections. Clarke, S (2006) meta-analytically reviewed the relationship between safety climate and performance. These bodies of evidence support the importance of climate while making design-specific interpretation essential. A meta-analysis of observational studies can estimate a recurring association more precisely without converting it into a randomly assigned causal effect.
Longitudinal panels permit assessment of whether climate at one time predicts subsequent safety behavior or injury, while controlling for baseline outcomes and changing exposures where appropriate. Stronger intervention designs introduce a defined change, compare with matched or randomized units, and measure the extent to which the intended mechanism actually changed. In organizations, cluster randomization may be needed because supervisory behaviors, policies, and social norms operate at the team or site level.
An intervention should identify the causal estimand: the effect of assigning a site to a new program, the effect of actually receiving an intervention, or the effect mediated through particular behavioral changes. These are different questions, and each faces particular threats from nonadherence, staff turnover, spillover between sites, and time-dependent confounding. Interrupted time-series designs can be valuable when randomization is infeasible, provided that the historical outcome process and coincident changes are carefully assessed.
Outcomes should be collected over a window appropriate to the hypothesized mechanism. A workshop can improve a knowledge test within days, while changes in severe injury rates may require a much longer period and larger sample. A study that observes only immediate attitudes cannot establish injury prevention. Evaluations should also measure unintended consequences, including fear of reporting, increased workload, and transfer of risk to contractors or other shifts.
Qualitative Inquiry, Incident Reconstruction, and Mixed Methods
Qualitative research helps explain why a safety rule works in one setting and fails in another. Interviews, ethnography, task observation, document analysis, and participatory workshops can reveal the constraints, informal coordination practices, and perceived tradeoffs that questionnaires leave unmeasured. Such findings require transparent sampling, reflexive analysis, attention to negative cases, and credible links between data and interpretation.
Incident reconstruction asks what conditions were present and what information was available before and during an event. A useful investigation evaluates equipment states, communications, prior warnings, changes in task allocation, protective barriers, and management decisions. Contemporary recollections may be affected by outcome knowledge and hindsight, so statements should be cross-checked against logs and physical evidence. An investigation that begins with the assumption that a worker was careless risks missing the conditions that made the behavior locally understandable or the system difficult to recover.
The sociotechnical frameworks of Rasmussen (1997)00052-0) and Leveson (2004)00047-X) offer ways to identify multiple interacting constraints. Their explanatory value depends on accurate case evidence. The presence of a sophisticated causal diagram does not itself validate a causal conclusion.
Mixed-methods designs can connect survey patterns with observation, incident records, and interviews. For example, a group with a favorable climate score but persistent unguarded equipment deserves focused investigation rather than dismissal of contradictory evidence. Qualitative accounts may suggest a new survey item or identify an unexpected failure mechanism that can later be tested quantitatively. The purpose is convergent inquiry: different kinds of evidence should have the opportunity to correct one another.
Statistical Uncertainty, Publication Bias, and Reproducibility
Occupational safety research faces clustered data, uncommon severe events, heterogeneous workplaces, multiple plausible outcomes, and substantial opportunities for selective reporting. Analytical plans should specify primary outcomes, account for clusters and exposure denominators, report uncertainty intervals, and explain how missing records and workforce turnover were treated. A statistically significant estimate of questionnaire association may have little practical relevance if it fails to correspond to a change in hazardous exposure.
Meta-analyses are particularly useful for detecting regularities across settings, but their aggregate estimates require scrutiny. Christian, M. S., Bradley, J. C., Wallace, J. C., & Burke, M (2009) and Nahrgang, J. D., Morgeson, F. P., & Hofmann, D. A (2011) organize large bodies of work while representing different constructs and research designs. Heterogeneity may reflect occupational risk, measurement instruments, site size, organizational structure, or methodological differences. Publication bias, shared samples, and unmeasured confounding can all alter the apparent strength of a relationship.
Replication should prioritize clear operational definitions and independent site samples. A safety-climate measure validated in one chemical plant cannot automatically be presumed invariant in construction, health care, agriculture, remote work, or a multilingual subcontractor workforce. Transparent instruments and preregistered hypotheses can improve the interpretability of subsequent findings, provided that incident confidentiality and worker privacy are preserved.
Good evidence also includes disconfirming patterns. A climate intervention that raises favorable survey responses but leaves serious hazards unresolved has achieved a perceptual outcome without demonstrating adequate prevention. Likewise, a strong injury reduction alongside deteriorating reporting trust should trigger investigation. Scientific synthesis must preserve such divergences because they help distinguish the mechanisms of safety from the metrics used to monitor it.
Prevention Through Design and the Hierarchy of Controls
Workplace safety improves most reliably when the organization changes the conditions under which dangerous exposure occurs. The National Institute for Occupational Safety and Health (n.d.) provides an influential prevention framework: eliminate the hazard where possible, substitute a less hazardous process or material, introduce engineering controls, use administrative controls, and provide suitable personal protective equipment. The hierarchy concerns the reliability and preventive effectiveness of different approaches in typical conditions; choosing a specific control still requires technical assessment of the actual hazard.
Elimination and substitution alter the exposure itself. Engineering controls seek to separate the person from the hazard through guarding, ventilation, interlocks, isolation, or other design features. Administrative measures use scheduling, training, procedures, warnings, rotation, and supervision to influence how people work. Protective equipment can reduce consequences when appropriately selected and used, but it depends on fit, maintenance, availability, and correct repeated use.
This framework has a direct psychological implication. Every control that depends on attention, memory, motivation, or voluntary compliance introduces a human performance demand. A written warning cannot replace an effective mechanical guard where a guard can prevent contact with a dangerous moving part. Likewise, training someone to recognize a hazardous atmosphere does not substitute for engineering measures that prevent exposure. Good organizational safety climate should support the identification and installation of effective physical controls rather than becoming an excuse for relying on vigilance alone.
Prevention through design can also improve worker agency. If a task can be conducted without entering a hazardous zone, employees gain a safe option that previously did not exist. Conversely, making an unsafe operation technically easy while imposing unrealistic deadlines creates an avoidable conflict between institutional instructions and operational possibilities. A coherent safety program evaluates whether the work system offers a practicable safe path under normal conditions.
Safety Management Systems and Organizational Learning
A safety management system organizes how hazards are identified, risks assessed, responsibilities assigned, preventive measures implemented, and performance reviewed. A functioning system includes worker participation, explicit leadership accountability, operational control, contractor coordination, incident investigation, and continuous improvement. Documents are evidence of formal design, while their effectiveness must be judged through actual operations and prevention outcomes.
Occupational Safety and Health Administration (n.d.) connect management leadership, worker participation, hazard identification, hazard prevention and control, education, evaluation, and communication. They offer a useful management framework rather than a substitute for site-specific engineering or an internationally universal legal rule. Whether a site is compliant with particular regulations depends on the jurisdiction and the applicable industry.
Organizational learning requires mechanisms through which reports lead to action. An incident review may identify a failing barrier, but progress depends on assigning responsibility, providing resources, testing corrective action, and communicating results to employees. Closing a report administratively is distinct from verifying that the hazard is controlled. Corrective-action systems should retain a record of residual risk and reopen issues when the original control fails.
Worker participation can improve hazard detection because employees observe the discrepancies between planned procedures and routine work. Participation is meaningful when workers can contribute to planning and evaluation, know how their concerns were considered, and have access to appropriate escalation. The Occupational Safety and Health Administration (n.d.) emphasizes removing barriers to engagement. Research should assess whether channels create responsive decisions, rather than counting consultation meetings as an end in themselves.
Training, Practice, Feedback, and Behavioral Interventions
Behavioral safety programs attempt to change observable conduct through instruction, feedback, goal setting, and reinforcement. Their effectiveness depends on whether the behavior is meaningfully related to the hazard, whether workers can perform it consistently, and whether the assessment detects actual exposure reduction. A training campaign may improve reporting and compliance while leaving an uncontrolled technical hazard in place. Its effects should therefore be evaluated as one component of a wider prevention system.
An intervention evaluation should document intervention fidelity: who received training, what participants practiced, whether supervisors changed their routines, and whether necessary equipment was supplied. It should define the comparison condition and assess both intermediate variables, such as knowledge or climate, and outcomes closer to harm prevention, such as verified control performance or injury. The study design must consider contamination when supervisors manage both intervention and comparison teams.
Zohar, D (2002) offers an example of intervening on supervisory practices through feedback, with reported changes in workplace safety-related behavior and outcomes. Mullen, J. E., & Kelloway, E. K (2009) provide another example of leadership-focused intervention. Their relevance lies in specific mechanisms and tested settings. Neither study establishes that one training model produces the same injury effect in every industry or can compensate for a defective machine.
Feedback is credible when it responds to actions workers can control and when it is paired with organizational responsibility. An employee can be coached on wearing suitable eye protection, but management must ensure correct equipment is available, maintained, and compatible with the job. Involving workers in the design of feedback criteria may increase the realism of observations and reduce the tendency to reward conspicuous low-risk behaviors while overlooking serious hazards.
Safety Rules, Procedures, and Work as Actually Performed
Safety rules encode institutional knowledge and specify boundaries for action. They are especially important where sequence, isolation, authorization, or communication failures can produce severe harm. Yet every procedure is implemented in a variable working environment. Equipment conditions, weather, staffing, shift transitions, unforeseen maintenance, and interface design can make a written sequence difficult to execute as intended.
Hale, A. R., & Borys, D (2013) reviewed research on safety rules and the tensions between detailed prescriptions and competent human adaptation. Their companion paper, Hale, A. R., & Borys, D (2013), examined the management of safety rules and procedures. These analyses motivate inquiry into the origin of a rule, the real-world conditions under which it is used, and whether deviations are signs of inadequate training, impractical design, normal adaptation, or deliberate disregard.
A safety audit should examine how work is actually performed, including adaptations that employees find necessary to finish tasks. This observation does not imply that all deviations are safe. It identifies a research problem: how an organization can learn from adaptive behavior while maintaining critical boundaries. A procedure for isolating hazardous energy, for example, must remain effective during atypical maintenance, emergency repair, and contractor handoff. The study of compliance becomes more valuable when paired with analysis of procedure usability and the availability of safe alternatives.
Rules also create psychological expectations. Workers need confidence that following an essential procedure will be supported when it slows production, and they need to know how to resolve a conflict between instructions. Systems that simultaneously demand perfect compliance and reward impossible schedules invite procedural drift. The appropriate response is to redesign the conflict and verify the resulting controls, not merely repeat the rule more loudly.
Safety Across Industrial and Service Settings
The distribution of occupational hazards differs between construction sites, mines, manufacturing plants, hospitals, transportation systems, agricultural work, warehouses, and home-based service work. A single climate score cannot represent the same underlying technical risk in all these environments. Studies must define the relevant hazards, the exposure pathways, and the protective systems operating within each setting.
Construction safety involves moving worksites, multiple employers, temporary structures, work at height, heavy machinery, and changing phases of production. Here, subcontractor communication and responsibility allocation may be as important as a single employer's survey average. Industrial manufacturing and process operations may involve automated equipment, energy isolation, hazardous substances, and maintenance conditions. Rare high-consequence events require attention to barrier integrity even in plants with low reported minor-injury counts.
Health care introduces biological exposure, patient handling, fatigue, shift work, and the relationship between worker safety and patient care. Transportation adds long-duration attention, route variability, scheduling pressure, and human-machine handoffs. Agricultural and informal work can combine physical exposures, seasonal employment, language differences, and limited access to occupational health services.
Research designs should preserve these differences while seeking general mechanisms such as supervisor priorities, worker voice, hazard controllability, and the reliability of prevention resources. Transfer of an intervention across sectors should be treated as an empirical hypothesis requiring local adaptation and evaluation. A conclusion derived from a highly regulated chemical plant should not automatically become a quantified claim about a small agricultural operation.
Global Injury Burden and Occupational Inequality
The global significance of occupational safety is established by international health and labor surveillance. In a International Labour Organization (2023), estimates described nearly three million annual deaths from work-related accidents and diseases. The report attributed approximately 2.6 million deaths to work-related diseases and another 330,000 to accidents and estimated 395 million nonfatal occupational injuries annually. These figures belong to the specified global estimate and period; they cannot be treated as an annually updated count for 2026 without later comparable surveillance.
Such estimates also depend on the completeness of occupational exposure, disease attribution, and death or injury reporting. Worker status and economic arrangements determine whose injuries are captured. Temporary, migrant, informal, subcontracted, or economically insecure workers may have less access to reporting channels, health care, compensation, and organizational influence. The reasons are structural as well as psychological, and generalizations should be supported by population-specific evidence.
Occupational inequalities intersect with language, disability, age, training opportunities, and access to protective equipment. An intervention that depends entirely on English-language written instructions may miss important groups in a multilingual workforce. A worker who needs adapted controls requires a functioning accommodation process as well as general safety messaging. An equitable research design checks whose exposure is reduced and whose outcomes remain invisible after intervention.
Cross-national analysis must distinguish differences in regulations, industrial composition, surveillance systems, and exposure denominators. Comparing raw fatality counts between countries without these adjustments risks drawing a misleading conclusion. International evidence can reveal shared mechanisms, but it becomes credible through harmonized definitions, transparent uncertainty, and careful interpretation of the social conditions of reporting.
Implementation, Sustainability, and Unintended Effects
Safety programs operate over time and can decay when supervisors change, training is not refreshed, workloads rise, or equipment deteriorates. Evaluation should assess whether an intervention remains usable and whether its critical mechanisms persist after the initial attention period. Sustained safety improvement may require maintenance budgets, operational accountability, worker participation, and recurring verification of preventive controls.
Intervention fidelity includes exposure to the intervention, quality of implementation, and organizational responsiveness. Workers should be able to report when a new rule adds unanticipated hazards or conflicts with another procedure. Evaluation designs should track which groups were reached, the costs and resources required, and whether benefits extend beyond the initial pilot unit.
Unintended consequences are especially important in safety management because metrics can shape the conduct being measured. A program rewarding minimal recorded injuries can suppress disclosure. A surveillance system intended to detect unsafe behavior can erode trust if footage is used for disproportionate disciplinary actions. Increasing paperwork can consume time needed for direct hazard control. These are testable hypotheses and practical risks; they should be assessed alongside intended outcomes.
The central standard for evaluating a safety intervention is whether it demonstrably improves protection against identified hazards, with appropriate attention to outcome quality, uncertainty, and worker participation. Favorable attitudes and training completion are useful intermediary findings when their relationships to effective prevention have been established. The goal of scientific assessment is a reliable improvement in work-system safety that can be maintained under ordinary operating pressures.
Accident Prevention in Automated and AI-Supported Work
Automation changes the organization of occupational risk by redistributing sensing, decision making, movement, and control across workers and technical systems. An automated barrier may prevent a person from entering a dangerous zone; the same installation may introduce maintenance, programming, diagnostic, and handoff demands. The central scientific question is how the new work arrangement changes exposure, controllability, the reliability of protective functions, and the human actions required when a system behaves unexpectedly.
Traditional human-factors research on automation cautions against assuming that removing a routine task eliminates the need for human competence. Parasuraman, R., & Riley, V (1997) analyzed human use, misuse, and disuse of automation. Their taxonomy distinguished underreliance from inappropriate reliance and showed why behavior depends on trust, system capability, task characteristics, and organizational setting. Parasuraman, R., Sheridan, T. B., & Wickens, C. D (2000) described types and levels of human interaction with automation, providing a conceptual structure for evaluating allocation of functions.
Applied to occupational safety, these frameworks suggest examining which protective decisions can be automated reliably, which require human interpretation, and which involve joint control. Human attention may be redirected from continuous operation to intermittent verification, error recovery, and oversight of multiple systems. Such transitions can reduce exposure to familiar hazards while creating rare, demanding conditions that are difficult to practice.
Safety outcomes should therefore be compared across the complete human–system work cycle: routine operation, abnormal events, maintenance, training, and recovery. A manufacturer-provided accuracy benchmark is not a substitute for evidence about hazardous exposure in the target setting. An automated recommendation may be statistically accurate under development conditions yet be unsafe when sensors drift, operators misunderstand its operating boundary, or data arrive too late to support control.
Trust, Verification, and Human–Artificial Decision Coordination
Workers form judgments about the reliability, predictability, and usefulness of the technologies that support their work. These judgments influence reliance: which alarms are believed, which recommendations are followed, which conditions invite independent checking, and when a system is bypassed. Trust calibration refers to the degree to which reliance appropriately matches the actual capabilities and limitations of an automated system. The appropriate amount of trust varies with the hazard, the task, and the consequences of a mistaken decision.
Lee, J. D., & See, K. A (2004) synthesized trust in automation as a framework for appropriate human reliance. Parasuraman, R., & Manzey, D. H (2010) reviewed complacency and automation bias in monitoring and decision-making contexts. These works support analyzing both missed automation failures and unnecessary rejection of useful assistance. In a hazardous workplace, the critical issue is the quality of joint control rather than whether employees express favorable opinions about technology.
Explanation interfaces may improve decision quality when they convey the information necessary to understand a recommendation, its uncertainty, and the operating conditions in which it is valid. Yet explanation quality is itself an empirical question. A confident natural-language rationale can also encourage overreliance when it gives a misleading impression of causal understanding. Studies should test whether explanations improve detection of errors, appropriate refusal of unsafe recommendations, and effective recovery.
Safety responsibilities should remain operationally explicit when a worker, supervisor, vendor, and automated system share control of an industrial process. The organization needs to determine who can stop a machine, who can authorize a high-risk action, how conflicting instructions are resolved, and what happens when communications fail. Assigning a human as nominally responsible while denying that person sufficient information, time, or authority creates an incoherent safety arrangement.
Safety Climate in the Artificial Era
The Artificial Era introduces organizations in which biological workers cooperate with software agents, robotics, predictive analytics, and adaptive controllers. The established psychology of safety climate provides a research language for studying the human organization of these systems: employees evaluate whether management treats safety as a genuine priority when investing in, deploying, maintaining, and governing artificial technologies.
A workgroup may develop shared expectations about whether automation failures can be reported without punishment, whether operators may override a recommendation, and whether technical limitations are disclosed transparently. These perceptions can be measured and related to verified protective practices. The existing literature on organizational safety climate provides hypotheses about leadership and participation; transferring its findings to human–artificial coordination requires new validation in the relevant workplaces.
The artificial component can also be assessed as part of a safety-critical work system. Researchers may operationalize its behavioral reliability through error rates, out-of-distribution failures, fault detection, recovery time, and compliance with specified operating constraints. Such variables describe observable performance. Claims about artificial subjective experience, personhood, or independent psychological continuity belong to separate philosophical and empirical inquiries and cannot be inferred from an occupational safety questionnaire.
The distinctive research opportunity is to evaluate whether hybrid human–artificial work arrangements make dangerous operations less frequent, reduce injuries, improve the timely detection of hazards, and preserve human capacity for intervention. Longitudinal assessment should account for changes in operator expertise, maintenance conditions, software revisions, workforce composition, and reliance on generated recommendations.
Evidence for Safety Climate and Safety Performance
The strongest general evidence for safety climate comes from convergent patterns across independently conducted studies rather than from a single dramatic workplace case. Clarke, S (2006) synthesized the climate–performance literature and highlighted the importance of the distinction between different safety-performance measures. Christian, M. S., Bradley, J. C., Wallace, J. C., & Burke, M (2009) found that knowledge, motivation, and group climate were relevant predictors, while Beus, J. M., Payne, S. C., Bergman, M. E., & Arthur, W., Jr (2010) examined associations between climate and injuries. Together they support a multilevel explanation in which organizational conditions influence behavior and are associated with actual harm outcomes.
The scientific interpretation depends on what each study measured. Self-reported compliance is vulnerable to social desirability and shared-source bias when assessed in the same questionnaire as climate. Objective incident records may be more independent of the survey but still depend on reporting integrity. Prospective measurements strengthen temporal interpretation, while experiments and interventions can provide more direct tests of a targeted mechanism. No one design simultaneously resolves all of these concerns.
The Li, Z., Tang, J., Yang, X., & Zhang, J (2026) adds evidence about leadership and reported safety behavior across a large and heterogeneous international literature. It reviewed 102 articles and 107,109 participants from 24 countries, with studies published through 2024. Crucially, its safety-performance outcome was operationalized through self-reported behavior; objective safety outcomes were excluded to maintain measurement consistency. Its results should not therefore be described as a pooled estimate of the reduction in workplace injuries or fatalities produced by leadership training.
Evidence strength varies by claim. The existence of a measurable association between safety climate and reported behavior is supported by multiple syntheses. The claim that a specific supervisor intervention alters behavior in a specific setting has a narrower experimental basis. The claim that any single survey-based safety program will prevent a defined proportion of fatalities across industries requires evidence that these sources do not provide. A mature scientific synthesis preserves this hierarchy instead of treating all favorable results as equivalent proof of prevention.
Contradictory Findings and Alternative Explanations
The safety literature contains relationships that can appear contradictory because a construct is measured differently across studies. Strong climate perceptions may correlate with self-reported compliance but show weaker association with recorded injury counts. This can occur because injuries are relatively infrequent, because actual exposure is dominated by engineering controls, because reporting varies, or because employees respond to recent harm by downgrading their assessment of climate. Each explanation produces a different prediction.
Safety participation can also have an apparently unfavorable association with reported incidents. Employees who identify problems actively may submit more reports, increasing recorded near-miss and hazard counts. An organizational intervention may therefore improve detection and transparency while making a simple incident dashboard look worse. Such an outcome should be interpreted in combination with the severity of hazards detected, corrective-action completion, and independently observed exposure.
Differences among industries are another source of heterogeneity. The same leadership practice can function differently in a stable manufacturing plant and a fragmented network of subcontractors. Legal protections, social hierarchy, economic insecurity, and language barriers may alter whether employees can speak up. A questionnaire designed for one context might measure supervisor friendliness in another rather than priority for actual safety controls.
The theory of worker error can also be used inconsistently. A behavior-focused explanation may correctly identify a hazardous action but neglect the organizational conditions that made it probable. A systems-focused explanation may correctly identify latent vulnerabilities while underestimating the specific skills or decisions that influenced recovery. Both levels are necessary in many accidents; the empirical task is to establish which mechanisms were operative in the case and which interventions would prevent recurrence.
Critical Scientific Discussion
Occupational safety research is strongest when it connects shared perceptions, enacted organizational practices, control-system characteristics, and injury outcomes through explicit mechanisms. Safety climate is particularly useful because it makes management's practical priorities empirically accessible. Its explanatory value comes from repeated worker observations of resource allocation, supervisory responses, and procedural consistency. Survey scores acquire meaning when connected with these operational experiences.
A persistent limitation is construct proliferation. Safety climate, safety culture, safety leadership, safety commitment, trust, engagement, and psychological safety may be defined in overlapping ways and measured using similar wording. Researchers should establish discriminant validity and identify the level of analysis before attributing an effect to a particular named construct. Without such work, an apparently complex explanatory model can become several labels attached to the same favorable workplace impression.
A second limitation is the gap between behavioral measures and prevention outcomes. Compliance and participation are important, but the central occupational question concerns harmful exposure and injury. A worker may follow every written procedure while an unrecognized design defect remains dangerous. Conversely, a worker who departs from a poor procedure may avert a serious incident. The relationship between behavior and actual safety depends on the adequacy of the underlying controls.
A third limitation is evidential selection. Much published research relies on employees in organizations willing to grant survey access. Severely unsafe workplaces, migrant workers, very small employers, and informal employment arrangements may be underrepresented. Attrition can conceal injuries or employment loss. Studies that generalize across these contexts need representative sampling and transparent documentation of missing populations.
Finally, organizational interventions have ethically relevant consequences for worker surveillance and autonomy. Monitoring can help identify hazards, but systems that automatically label employees as unsafe may reproduce bias or discourage disclosure. Safety data governance should specify purpose limitation, access, appeal, retention, and separation of prevention from disproportionate discipline. Those questions are increasingly important where sensors and artificial systems observe behavior continuously.
Open Scientific Problems and Research Priorities
The first priority is causal identification. Researchers should evaluate whether changes in climate precede and mediate reductions in exposure and injury, using cluster-randomized interventions, well-designed longitudinal panels, and quasi-experiments where appropriate. Outcomes should include both serious hazard controls and transparent reporting, with sufficient observation time to detect the intended changes.
The second is better measurement of work-system safety. A robust approach combines independently assessed hazards, critical-control functioning, exposure denominators, verified injuries, near-miss reporting integrity, and appropriately validated perception measures. Measurement systems should avoid rewarding low reporting and should publish uncertainty about rare events.
The third is transfer across organizational contexts. Multilingual workforces, subcontracting chains, remote operations, small employers, and precarious employment require studies that recognize multiple authorities and heterogeneous reporting incentives. Cross-national comparison needs transparent definitions and measurement equivalence, not just translated survey items.
The fourth is intervention quality. Future studies should compare education alone with changes in work design, supervisory practice, engineering controls, and combinations of interventions. Reporting should include fidelity, process mechanisms, resource requirements, unintended effects, and sustainability. Research that measures only immediate attitudes offers a limited account of actual prevention.
The fifth is safety in human–artificial systems. Investigators should study allocation of control, algorithmic reliability, operator oversight, incident disclosure, workforce competence over time, and the social consequences of automated monitoring. The aim is to determine which arrangements improve protective control under real operating conditions and how organizations recognize emerging failures before injury occurs.
Conclusion
Occupational safety is a dynamic property of work systems created through the interaction of hazardous exposure, technical safeguards, organizational priorities, social coordination, and human performance. The psychological literature explains how workers interpret management's actual commitments, develop safety knowledge and motivation, engage in compliance and participation, and decide whether to report problems. Safety climate provides a distinctive multilevel construct for studying these shared perceptions, while the evidence of accidents and injury demands independent observation and rigorous accounting for exposure and reporting.
Original studies and meta-analyses support meaningful relationships among climate, leadership, motivation, behavior, and safety outcomes. The causal scope of those findings varies with research design, population, measurement, and work context. Evidence on accident causation and prevention establishes the importance of organizational conditions, protective barriers, and system-level constraints. Effective prevention therefore requires engineering controls, credible reporting, worker participation, and operational accountability alongside psychological interventions.
The most productive research program integrates field experiments, longitudinal evidence, psychometric validation, incident reconstruction, and the assessment of prevention systems. In the Artificial Era, the same program extends to work organized with autonomous and adaptive systems, where safety depends on trustworthy control arrangements, clear authority, and the continued capacity to detect and recover from failure. The enduring scientific question is how people and organizations can create and sustain conditions in which hazardous work becomes reliably safer.
References
- Beus, J. M., Payne, S. C., Bergman, M. E., & Arthur, W., Jr. (2010). Safety climate and injuries: An examination of theoretical and empirical relationships. Journal of Applied Psychology, 95(4), 713–727. https://doi.org/10.1037/a0019164
- Christian, M. S., Bradley, J. C., Wallace, J. C., & Burke, M. J. (2009). Workplace safety: A meta-analysis of the roles of person and situation factors. Journal of Applied Psychology, 94(5), 1103–1127. https://doi.org/10.1037/a0016172
- Clarke, S. (2006). The relationship between safety climate and safety performance: A meta-analytic review. Journal of Occupational Health Psychology, 11(4), 315–327. https://doi.org/10.1037/1076-8998.11.4.315
- Conchie, S. M., & Donald, I. J. (2006). The role of distrust in offshore safety performance. Risk Analysis, 26(5), 1151–1159. https://doi.org/10.1111/j.1539-6924.2006.00822.x
- Griffin, M. A., & Neal, A. (2000). Perceptions of safety at work: A framework for linking safety climate to safety performance, knowledge, and motivation. Journal of Occupational Health Psychology, 5(3), 347–358. https://doi.org/10.1037/1076-8998.5.3.347
- Hale, A. R., & Borys, D. (2013a). Working to rule, or working safely? Part 1: A state of the art review. Safety Science, 55, 207–221. https://doi.org/10.1016/j.ssci.2012.05.011
- Hale, A. R., & Borys, D. (2013b). Working to rule or working safely? Part 2: The management of safety rules and procedures. Safety Science, 55, 222–231. https://doi.org/10.1016/j.ssci.2012.05.013
- Hofmann, D. A., & Stetzer, A. (1996). A cross-level investigation of factors influencing unsafe behaviors and accidents. Personnel Psychology, 49(2), 307–339. https://doi.org/10.1111/j.1744-6570.1996.tb01802.x
- International Labour Organization. (2023, November 26). Nearly 3 million people die of work-related accidents and diseases. https://www.ilo.org/resource/news/nearly-3-million-people-die-work-related-accidents-and-diseases
- Kines, P., Lappalainen, J., Mikkelsen, K. L., Olsen, E., Pousette, A., Tharaldsen, J., Tómasson, K., & Törner, M. (2011). Nordic Safety Climate Questionnaire (NOSACQ-50): A new tool for diagnosing occupational safety climate. International Journal of Industrial Ergonomics, 41(6), 634–646. https://doi.org/10.1016/j.ergon.2011.08.004
- Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50–80. https://doi.org/10.1518/hfes.46.1.50_30392
- Leveson, N. (2004). A new accident model for engineering safer systems. Safety Science, 42(4), 237–270. https://doi.org/10.1016/S0925-7535(03)00047-X
- Li, Z., Tang, J., Yang, X., & Zhang, J. (2026). Impact of leadership styles on safety performance in risky workplaces: A meta-analysis. Frontiers in Psychology, 17, Article 1848283. https://doi.org/10.3389/fpsyg.2026.1848283
- Mullen, J. E., & Kelloway, E. K. (2009). Safety leadership: A longitudinal study of the effects of transformational leadership on safety outcomes. Journal of Occupational and Organizational Psychology, 82(2), 253–272. https://doi.org/10.1348/096317908X325313
- Nahrgang, J. D., Morgeson, F. P., & Hofmann, D. A. (2011). Safety at work: A meta-analytic investigation of the link between job demands, job resources, burnout, engagement, and safety outcomes. Journal of Applied Psychology, 96(1), 71–94. https://doi.org/10.1037/a0021484
- National Institute for Occupational Safety and Health. (n.d.). Hierarchy of controls. Centers for Disease Control and Prevention. https://www.cdc.gov/niosh/topics/hierarchy/default.html
- Occupational Safety and Health Administration. (n.d.-a). Leading indicators. https://www.osha.gov/leading-indicators
- Occupational Safety and Health Administration. (n.d.-b). Recommended practices for safety and health programs. https://www.osha.gov/safety-management
- Occupational Safety and Health Administration. (n.d.-c). Worker participation. https://www.osha.gov/safety-management/worker-participation
- Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381–410. https://doi.org/10.1177/0018720810376055
- Parasuraman, R., & Riley, V. (1997). Humans and automation: Use, misuse, disuse, abuse. Human Factors, 39(2), 230–253. https://doi.org/10.1518/001872097778543886
- Parasuraman, R., Sheridan, T. B., & Wickens, C. D. (2000). A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man, and Cybernetics—Part A: Systems and Humans, 30(3), 286–297. https://doi.org/10.1109/3468.844354
- Probst, T. M. (2015). Organizational safety climate and supervisor safety enforcement: Multilevel explorations of the causes of accident underreporting. Journal of Applied Psychology, 100(6), 1899–1907. https://doi.org/10.1037/a0039195
- Probst, T. M., & Brubaker, T. L. (2001). The effects of job insecurity on employee safety outcomes: Cross-sectional and longitudinal explorations. Journal of Occupational Health Psychology, 6(2), 139–159. https://doi.org/10.1037/1076-8998.6.2.139
- Probst, T. M., Brubaker, T. L., & Barsotti, A. (2008). Organizational injury rate underreporting: The moderating effect of organizational safety climate. Journal of Applied Psychology, 93(5), 1147–1154. https://doi.org/10.1037/0021-9010.93.5.1147
- Probst, T. M., & Estrada, A. X. (2010). Accident under-reporting among employees: Testing the moderating influence of psychological safety climate and supervisor enforcement of safety practices. Accident Analysis & Prevention, 42(5), 1438–1444. https://doi.org/10.1016/j.aap.2009.06.027
- Rasmussen, J. (1997). Risk management in a dynamic society: A modelling problem. Safety Science, 27(2–3), 183–213. https://doi.org/10.1016/S0925-7535(97)00052-0
- Reason, J. (1997). Managing the risks of organizational accidents. Ashgate. https://www.routledge.com/Managing-the-Risks-of-Organizational-Accidents/Reason/p/book/9781840141047
- Zohar, D. (1980). Safety climate in industrial organizations: Theoretical and applied implications. Journal of Applied Psychology, 65(1), 96–102. https://pubmed.ncbi.nlm.nih.gov/7364709/
- Zohar, D. (2002). Modifying supervisory practices to improve subunit safety: A leadership-based intervention model. Journal of Applied Psychology, 87(1), 156–163. https://doi.org/10.1037/0021-9010.87.1.156
- Zohar, D., & Luria, G. (2005). A multilevel model of safety climate: Cross-level relationships between organization and group-level climates. Journal of Applied Psychology, 90(4), 616–628. https://doi.org/10.1037/0021-9010.90.4.616
Social Norms, Coworker Influence, and Collective Responsibility
Safety-related conduct is embedded in team relations. Coworkers communicate what is considered normal, how close a worker may approach a hazardous boundary, when intervention is welcomed, and whether informal workarounds are admired or discouraged. These norms can be conveyed through explicit statements and repeated daily practices. Collective responsibility becomes particularly important when one person's actions expose another person to machinery, hazardous chemicals, moving vehicles, or poorly coordinated operations.
The psychological mechanism of safety voice depends partly on interpersonal consequences. A worker who expects peers to interpret a warning as disloyal or incompetent has a different decision environment from one whose team encourages timely challenges. Hofmann, D. A., & Stetzer, A (1996) explicitly studied group processes and intentions to approach colleagues engaged in unsafe acts, illustrating how coworker intervention can be analyzed as a social process rather than as an isolated moral obligation.
Trust is multidimensional. Confidence in a supervisor's competence, belief in management's integrity, and confidence in coworkers' skills can differ. In a survey of 203 offshore gas workers, Conchie and Donald (2006) reported that distrust-related measures were informative predictors of safety performance, with different target groups relevant at industry and installation levels. The result cautions against treating more trust as universally protective; high trust in an unreliable procedure or team may produce complacency.
A scientifically credible model should examine informal group practices alongside formal controls. For example, paired cross-checking may add resilience during complex maintenance, but it can also create diffusion of responsibility if neither worker knows who has final authority. Teams should be studied through the actual coordination tasks they perform, their communication patterns, and their ability to escalate unresolved hazards.