Who should manage worker safety to reduce occupational accidents?

Workplace safety remains a global concern despite increasing efforts of advocates, employers, occupational health and safety professionals, regulators, unions and management researchers to make operational work safe (Pagell et al., 2020; Wiengarten et al., 2019; Awasthy and Hazra, 2019). According to the United Nations Global Compact (2022), 2.78 million workers die annually from occupational accidents and work-related diseases worldwide. This issue is not confined to developing nations and is equally critical in developed economies, such as those in Europe and the USA. In 2021, the European Union recorded 2.9 million non-fatal workplace accidents and 3,347 fatal accidents (Eurostat, 2024). Similarly, in 2022, the USA recorded 2.8 million non-fatal workplace accidents and 5,486 fatalities (U.S. Bureau of Labor Statistics, 2024). While the moral responsibility to ensure safety is evident, the economic implications are also significant. The U.S. National Safety Council estimated that in 2022, workplace accidents resulted in wage and productivity losses of $50.7 billion, medical expenses of $37.6 billion and administrative expenses of $54.4 billion. Therefore, ensuring workplace safety should be a critical societal, managerial, and political goal.

As a result, working conditions and occupational health and safety (OHS) have become fundamental components of sustainable operations management. Social sustainability focuses on both an organization’s or supply chain’s workers, as well as encompassing wider communities (e.g. Pullman et al., 2009). As such, OHS has become an integral part of the social pillar of sustainable operations management (Gimenez et al., 2012). Prior research has explored the performance implications of both safe or unsafe workplaces and investigated the operational factors that enhance or harm workplace safety (Pagell and Gobeli, 2009; Pagell et al., 2020; Wiengarten et al., 2019, 2021). Operations managers are responsible for ensuring workers are safe, and safety is often regarded as a core operational outcome, alongside performance outcomes like cost, quality, delivery, and flexibility (e.g. Pagell et al., 2015; Wiengarten et al., 2021). However, in many firms, the management of safety is delegated to dedicated safety managers, who work separately from the operations managers even though the empirical evidence suggests safety and operations should be managed jointly (e.g. Pagell et al., 2015). This separation of managerial responsibility and the inherent connections between safety and other operational outcomes, underscores the importance of better understanding who manages safety within the operations management domain (Neri et al., 2022).

Research linking safety to operations management has mainly explored operational polices, practices and decisions. For instance, safety certifications such as OHSAS 18001 (replaced by ISO 45001) have been linked to improved operational outcomes (e.g. Lo et al., 2014) and creating process rigidity that stymies responsiveness (Ye et al., 2023). Other research has linked operational decisions such as how much slack is in the system to safety outcomes (e.g. Wiengarten et al., 2017). Finally, research has suggested a systemic approach to manage both safety and operations simultaneously (e.g. Pagell et al., 2015).

These studies provide valuable insights into the operational practices and decisions that affect occupational safety. However, they are mostly silent on who should manage safety in an establishment. Safety is mainly the result of how the operations are managed and impacts operational workers, yet it is often managed by a separate safety function (e.g. Pagell et al., 2015). Therefore, the unanswered question of who should manage safety is not trivial. Equally important, it is very relevant to our discipline as accidents occur on the shopfloors, building sites, kitchens, and so on that we as operations managers are supposed to manage. Although the EU Framework Directive 89/391/EEC and ISO 45001 mandate that employers/leadership ensure that appropriate measures are taken to protect workers’ health and safety, they do not specify the exact job roles or individuals who should be responsible for the management of safety. This (regulatory) flexibility means that employers can choose who has the managerial responsibility for safety within their establishment.

We focus on who manages safety rather than formal titles for safety. Considering the job role of the person who manages safety should directly capture the influence of individuals who actively engage with and manage safety issues on a daily basis, rather than relying solely on formal titles. Recognizing that those who are tasked with OHS are typically the driving force behind OSH practices, we posit that they have a direct impact on the effectiveness of safety measures and the resulting accident rates. By emphasizing the role of active management over formal titles, we aim to highlight the practical influence of operational expertise on safety outcomes.

Drawing on stakeholder-agency theory (Hill and Jones, 1992) and related literature, we argue that owners tend to prioritize profitability and growth, potentially at the expense of worker safety. In contrast, employees who manage safety are more likely to prioritize safety due to their direct stake in working conditions. However, the complexities faced by the person managing safety likely vary with firm size. Larger establishments have more resources, experience greater external pressures and require more rigid standard operating procedures, making the role of the person managing safety more challenging. Therefore, effective role assignment might differ between smaller and larger establishments. To explore these conceptual arguments further, we propose the following research questions.

1. Do establishments where owners manage OHS experience an increase in accidents compared to companies where employees manage OHS?

2. Does an establishment’s size moderate the relationship between the job role of the person managing safety and accidents?

Answering these research questions helps to advance our understanding of how to improve safety in the workplace. We thereby complement existing literature in our domain that has predominantly focused on operational-level traits such as certifications. In doing so, we also follow the suggestion by Lornudd et al. (2021), “to better understand how occupational safety is managed in organizations, there is a need to extend the focus beyond the operational levels …” (p. 1). The findings provide valuable insights for scholars and managers and also offer clear guidance for policymakers in developing future strategies to create safer operations.

To answer our research questions, we utilize survey data collected through the European Survey of Enterprises on New and Emerging Risks (European Agency for Safety and Health at Work, ESENER, 2024), administered by the European Agency for Safety and Health at Work (EU-OSHA). Using survey data from ESENER I (2014) and ESENER II (2019), we developed a pseudo-panel of 88,281 firm-year observations to create representative samples for the economies of the 32 countries repeatedly surveyed. In addition, since firm-level data from 2014 and 2019 could not be matched due to the anonymity of the firms, our analysis focused on comparing differences at the level of the 32 economies across these two years.

Our findings indicate that when owners manage safety, accident rates increase, irrespective of establishment size. Additionally, we find that in smaller establishments, having employee representatives manage safety reduces the accident rate. In contrast, in larger establishments, it is more effective to assign a dedicated OHS officer to manage safety.

Brown (1996) and Pagell and Shevchenko (2014) argue that safety should be treated as a key operational performance outcome, which is now being reflected in the conceptualization of sustainability research in our domain (Lo et al., 2014; Wiengarten et al., 2017). Workplace safety is a core component of the sustainable operations management literature (Lo et al., 2014; Pagell et al., 2015, 2020; Wiengarten et al., 2017, 2019). Previous research in operations management has explored various antecedents of occupational safety (e.g. process design, human capital, availability of resources) and the operational and financial performance outcomes of occupational safety. A literature review by Fan et al. (2014) identified four major research dimensions in the operations management literature on OHS: safety climate, management systems integration, voluntary OHS systems and sustainable operations. More recently, Neri et al. (2022) conducted a systematic literature review on safety and operations. They particularly highlighted that the literature still needs to explore “the mutual relationships between safety and operations from a sustainability perspective”. Furthermore, they highlighted shortcomings concerning the empirical investigation of contexts, climates, specific interventions, and the decision-making process (Neri et al., 2022).

Much of the previous research in our field has focused on whether lean manufacturing is safe or unsafe (e.g. Wiengarten et al., 2019), or whether firms fare better when they reduce worker safety (Pagell et al., 2020). In addition, recent policies such as the Corporate Sustainability Due Diligence Directive in the EU and national policies in states and countries such as California, Germany, France, and Norway all emphasize safety. Researchers and managers are interested in exploring how workplace safety – as well as other social sustainability objectives, such as diversity, and human rights – can be achieved (Longoni et al., 2019; Durach et al., 2024). However, how worker safety is affected by who manages safety in an establishment on a day-to-day basis has not been sufficiently explored.

Agencies such as EU-OSHA or the Occupational Safety and Health Administration of the U.S. Department of Labor highlight and recommend multiple best practices for managing occupational safety. These agencies not only wield enforcement tools like site inspections and fines but also advocate for best practices that firms can proactively adopt to reduce the risk of accidents. The U.S. regulator, for example, proposes ten best practices to manage safety: establishing safety and health as a core value, leading by example, implementing a reporting system, providing training, conducting inspections, collecting hazard control ideas, implementing hazard controls, addressing emergencies, seeking input on workplace changes and making continuous improvements (OSHA, 2024).

Both the U.S. and the EU regulations emphasize the employer’s role in these best practices but allow for some flexibility in who manages safety. In other words, who is tasked with managing OHS in the establishment is not specified in regulation. Safety rules and regulations do not specify or differentiate by size when it comes to whether a dedicated safety manager is required. European directives, such as the EU Framework Directive 89/391/EEC, only require employers to take appropriate measures (due diligence) to ensure safety, making sure that safety is managed by “competent personnel” (Eur Lex, 2024). The directive does not specifically mandate a dedicated safety manager. Likewise, U.S. regulation requires firms to meet certain safety standards and to appoint a responsible person to handle safety issues. However, the regulation does not specify further on whom should be appointed, and whether the person appointed to handle safety issues should also be the person managing safety. Hence, firms can have different interpretations of both the formal titles for safety and who manages OHS. This degree of flexibility that firms have in the existing institutional frameworks might result in increased accidents and is the focus of this study.

Previous research has repeatedly identified various operational reasons as to why some workplaces remain unsafe. However, the flexibility that regulations and standards grant to firms in managing safety might itself lead to ineffective safety management and subsequently put operational workers at risk of being involved in an accident. Therefore, we explore if having occupational safety managed by owners as opposed to dedicated safety managers and employee representatives, increases accidents, and examine if establishment size moderates this relationship.

To analyze our research questions, we utilize stakeholder-agency theory (Hill and Jones, 1992). Stakeholder-agency theory builds on the foundations of both agency and stakeholder theories. Agency theory primarily examines the relationship between principals such as shareholders and agents such as managers (Bosse and Phillips, 2016). It defines an agency relationship as one in which one or more individuals (the principals) engages another person (the agent) to perform services on their behalf, which involves delegating some decision-making authority to the agent. Stakeholder-agency theory, extends this idea, suggesting that principal-agent relationships, as defined by agency theory, are a subset of the broader class of stakeholder-agent relationships. Stakeholders are defined as a group of people with an exchange relationship with the firm, such as employees (Chen et al., 2023; Hill and Jones, 1992).

Stakeholder-agency theory differentiates between mangers and non-managerial employees. Managers have a unique role in that they are hired as agents by the firm, but they also have responsibilities to all other stakeholder groups such as shareholders, the community, operational workers and so on (Hill and Jones, 1992). This responsibility comes from their having direct control over decision making. However, non-managerial employees such as operational workers do not have direct control over decision making so they are stakeholders but not agents (Hill and Jones, 1992). For simplicity we refer to non-managerial employees as employees or workers for the remainder of the paper.

Managers (as agents) provide the firm with time, skills, and human capital commitments. In exchange, they expect a fair income and decent working conditions. Owners or partners (as principals), on the other hand, have an equity stake in the business, sharing in its profits and losses and using personal assets to cover business debts if necessary. The cornerstone of agency theory is the assumption that the interests and objectives of principals and agents diverge (Bosse and Phillips, 2016). Stakeholder-agency theory expands on this by considering the implicit contracts not only between principles and agents, but also between the owners of a firm and other interest groups or stakeholders, including employees (Chen et al., 2023; Hill and Jones, 1992). This theory posits that power differentials in these relationships (e.g. employees have power vis-à-vis their employers, because of their ability to resign) limit the owners’ ability to strictly enforce implicit or explicit contracts on what workers deem unacceptable behavior/decision making. Consequently, if the interests of the owner and employees diverge, it is difficult for the owner or manager (as an agent) to manage the decision making process to effectively meet the interests or objectives of the workers.

It is reasonable to assume that the interests of owners, managers and employees regarding workplace safety might not align. For example, research on family businesses has shown that profitability goals rank high on owners’ agendas (Chrisman and Patel, 2012), because achieving financial targets is crucial for a firm’s viability, growth and long-term survival (Gómez-Mejía et al., 2007). Employees, on the other hand, have a vested and primary interest in their safety that comes before considering the financial wealth and growth of the firm they are working for (Nahrgang et al., 2011). Managers who act as agents for the owners and also need employees to respect and follow their guidance, likely fall in the middle.

Stakeholder-agency theory is appropriate for this research due to the unique role of managers in the network of stakeholders. Wiseman et al. (2012), for example, argued that it is reasonable to expect that wealth maximization is not the only interest of managers as agents, advocating for a social theory of agency. Bosse and Phillips (2016) proposed that the factors of reciprocity and fairness have a key effect on decision-making by agents. Consequently, we assume that if the person managing safety is a manager, they will place a higher emphasis on employee safety than an owner would in the same role.

Therefore, satisfying owners’ claims likely involves maximizing profitability and growth, while satisfying employees’ claims requires significant investments in safe working conditions. Campbell (2007) argued that firms primarily driven by profit and efficiency motives often disregard their social responsibilities towards workers, as there is a common belief that worker safety is primarily a cost factor. This aligns with traditional operations management thinking that often suggests that safe working conditions can hinder operational efficiency. Recent studies support this claim (Pagell et al., 2020; Wiengarten et al., 2021). Traditionally, research in our domain has viewed sustainability objectives as a hindrance to efficiency and profitability. This argument is particularly relevant when considering the potential tradeoff between social sustainability in the guise of OHS and profitability, especially when operations management is the focus (Wu and Pagell, 2011).

Satisfying employees’ claims for safety involves using resources that might otherwise be invested by owners in maximizing profits and growth, leading to an inherent conflict between owners and employees. Our logic closely aligns with recent theoretical arguments by Mitchell et al. (2016), who used the stakeholder-agency perspective to propose that when decision-making is no longer centralized with one person (e.g. the owner) but delegated, corporations automatically start integrating more of the objectives of all stakeholders, including operational workers, thereby increasing the social welfare of the firm.

In sum, agency conflict arises when job functions, other than the owner, manage OHS. The divergence between owners’ preferences and those of designated managers regarding resource allocation can lead to decisions that diverge from the owners’ key objectives. By definition, these roles are closer to the workers and their safety needs. Due to power differentials, owners cannot fully control these workers’ decisions, as these employees can exit the contractual relationship they have with the firm. We therefore propose the following hypothesis.

The question then arises: do owners and senior managers, such as managing directors or branch managers, exhibit the same outcomes in terms of accidents when they manage OHS? Senior managers are contracted agents, although they may also own shares in the firm. However, unlike owners, they have the immediate option to exit. They receive salaries, bonuses, and other benefits for their role in managing the firm, but they can relatively easily leave if they choose.

According to agency theory, if an agent does not like the terms of a contract, they can seek a better alternative (Bosse and Phillips, 2016) – although such alternatives may not always be available. It is reasonable to assume that the alignment of senior managers interests with the long-term financial viability and growth of the firm is less strong compared to owners. Supporting this assumption, literature has found that manager-controlled firms exhibit less profit-maximizing behavior compared to owner-controlled firms (Bothwell, 1980). Therefore, we do not expect that the effect postulated in H1, for owner managers, will hold for senior managers managing OHS.

Next, we consider whether the effectiveness of having OHS managed by people in different job functions depends on the size of the establishment. Our arguments consider two different job functions that typically manage safety: a dedicated OHS officer, or an employee representative. Using stakeholder theory, Darnall et al. (2010) showed that changes in firm size can explain differences in the adoption of proactive environmental strategies. Larger firms have more resources that need allocation, experience more external pressure, and require more rigidity in the form of standard operating procedures. We assume that larger firms face similar allocation and management complexities regarding safety. Those in charge of managing safety in larger firms must deal with such complexities.

How well different people manage OHS in firms and deal with complexity likely depends on their training. While the details are country-specific, becoming an OHS officer or specialist in Europe usually requires specific training focused on OHS practices, regulations, risk assessment and management, as well as typically improving regulatory knowledge (OHSA Europe, 2024). Employee representatives usually do not require and consequently do not receive such training.

Employee representatives effectively have a double agency role. First, by taking on managerial responsibility, even if just for safety, they take on decision-making authority and are acting as an agent for the firm. Second, as implied in their job title, they are an agent for the workers they represent. Hence, employee representatives, will have direct and frequent communication with the operational workers. These workers not only perform the operational tasks, but they are also exposed to most of the firm’s safety risks; risks that will be mainly a function of the way the operations are managed (e.g. Das et al., 2008). Operational workers know the work best and have the most to lose when safety is poorly managed.

Smaller firms are not as complex and typically have simpler operations and fewer employees. Therefore, it may be feasible for an employee representative to manage safety effectively in smaller establishments, even if they do not have specific safety training. The double agent status of employee representatives may even increase employee ownership and participation, relative to an OHS manager, in improving safety and other operational practices, fostering a collaborative approach.

Therefore, we propose that having a specialized OHS officer manage the task in larger firms might be more beneficial than having an employee representative manage OHS. In smaller firms, the employee representative might have the advantage of being more aware of the safety issues. Given the scarcity of research in this area, we propose and test a general theoretical prediction.

Figure 1 below graphically illustrates our research model and its two hypotheses.

To test our theoretical framework, we required data that captured both changes in the assignment of managing OHS within establishments, as well as subsequent changes in accidents. Such data is typically difficult to obtain through secondary databases such as Compustat or Bloomberg, while the alternative, cross-sectional data from self-administered surveys, often suffers from issues like limited temporal observations, representativeness concerns, and small-sample bias. Therefore to test our hypotheses, we utilized representative data from the ESENER surveys I and II conducted by EU-OSHA ^[1].

To test our theoretical assumptions, we utilized establishment-level data collected in the 2014 and 2019 waves of the ESENER survey. These surveys are designed to monitor OHS across Europe. Although EU-OSHA conducted ESENER surveys in 2009, 2014, and 2019, the methodology of the 2009 survey differed significantly from those of 2014 and 2019 in terms of the sectors observed, the establishments covered, and the respondents interviewed. The 2014 and 2019 surveys consistently targeted establishments with five or more employees and covered all NACE Rev.2 industry sectors in Europe from A-S, interviewing the “person most knowledgeable about health and safety in the establishment”. The 2009 survey excluded industry sector A (“Agriculture, Forestry and Fishing”), covered only establishments with ten or more employees, and varied in whom it surveyed, in terms of management, employee representatives responsible for OHS, or both. Due to its methodological inconsistencies, we excluded the 2009 data from our analysis.

Our theoretical framework is based on observing changes in the role of the person managing OHS and their effects on changes in accidents. Therefore, we capitalized on the consistent methodology and survey content of the 2014 and 2019 waves, which allowed for the observation of changes in the job roles of people managing OHS between 2014 and 2019, and the changes in accident-related outcomes reported in 2019.

The objective of the sampling strategy for the ESENER survey was to collect statistically representative samples of the population of establishments within each surveyed country. The sampling for the 2014 and 2019 surveys was based on a stratified random sampling procedure, using a sampling matrix defined by 28 cells (four size classes × seven sector groups) in each country (refer to Table 1). For each cell in this matrix, specific net sample targets were established by EU-OSHA.

EU-OSHA coordinated and administered the survey, but also relied on local contractors in each country to carry out the fieldwork. These contractors are usually well-established survey or research firms with a proven track record in conducting interviews and managing data collection on a large scale. Details on the selection and training of interviewers and supervisors of the data collection process can be obtained from the ESENER II and ESENER III overviews. ^[2].

The 2014 wave encompassed 49,320 establishments across 36 countries, while the 2019 wave covered 45,420 establishments, spread across 33 countries. Across both waves, 32 countries were consistently surveyed, comprising 44,354 establishments in 2014 and 43,927 establishments in 2019. For an overview of the countries and the number of observations within each country, per year, used in the current study, refer to Table 2.

After the data was collected, structural discrepancies between the net sample and the population of each country remained. Therefore, to make the data statistically representative of the population of each country, EU-OSHA calculated the inclusion probabilities of each establishment for each country ex post and provided weightings along with the survey data. We use these weights in our analysis.

EU-OSHA’s ESENER surveys provide longitudinal, representative, and large-scale data from entities across Europe. These surveys track changes in the role of the person tasked with managing OHS in each entity and measure changes in workplace accidents. Despite these unique advantages, using this dataset to answer our research questions presents two key challenges, which are also often encountered in studies of labor market dynamics (Ribas, 2022).

First, while ESENER sampled the person tasked with managing OHS to respond to the survey, it does not explicitly indicate whether this individual is also responsible for OHS, which would have been interesting to explore further.

Second, while the ESENER surveys offer a representative overview of the state of OHS in the surveyed countries, the cross-sectional firm-level data from the 2014 and 2019 surveys cannot be directly used as panel data. This is because EU-OSHA does not provide a unique identifier for the establishments in its datasets, making it impossible to create a panel data structure from the 2014 and 2019 datasets. Furthermore, a composite of data points within the datasets that would allow tracking repeated observations was also not possible. Other considerations, such as propensity score matching or fuzzy matching, also did not promise reliable results, given that firms likely changed in terms of observable variables between 2014 and 2019.

Consequently, we devised an alternative econometric strategy based on the survey data being representative of each country’s economy. This allowed us to develop a pseudo-panel model at the country level (Ribas, 2022). Pseudo-panels have been widely used in labor market dynamics research (Antman and McKenzie, 2007; Cuesta et al., 2011; Juodis, 2018), where individual-level transitions are inferred from cohort-based trends. While such a model cannot fully assess establishment-level effects (components of establishment effects such as OHS manager age, gender or within-establishment context can only be measured using true panels), the approach allows inferring establishment-level outcomes from country-level data. This approach also comes with reduced variance in the data, which may prevent the detection of certain effects (type II error), which we address in the limitations section.

Even though individual firms are not tracked over time in our pseudo panel, the representative nature of the survey means that the aggregated data at the country level reflects the average characteristics and outcomes of firms within that country. Given the representative nature of the survey data for each country, this approach allows us to explore how changes in the roles of the people managing OHS from 2014 to 2019 affected changes in accidents reported in 2019. Furthermore, within a single country, firms operate under a relatively homogeneous policy and regulatory environment, especially in terms of OHS regulations (Bonafede et al., 2016). Changes in OHS management responsibilities and practices at the country level are likely to impact all firms within that country in a uniform manner. Thus, by observing country-level changes, we can deduce the typical impact of these changes on individual firms. Essentially, this approach compares the same country against itself over time, thereby focusing on internal changes rather than cross-country differences.

Another advantage of this approach is that it inherently mitigates the single-respondent issue (Flynn et al., 2018) . At the country level, measures can be assumed to be robust to variations among individual responses, especially with sample sizes as large as in the ESENER waves (see Table 2). As the data represents the broader economic and regulatory environment rather than the specifics of individual firms, peculiarities and biases of single respondents are less likely to skew the overall picture because the responses are now representative of a wider array of business operations within the country.

To examine how changes in the role of the person managing OHS affects accidents, we calculated the mean of each measure relevant to this study at the country level, thereby weighting each observation based on the country-specific probability weights provided by EU-OSHA. As a result, we obtained mean values of each measure, representative of the economy in each country within our sample. This strategy provides a balanced pseudo-panel dataset of 64 country-year observations derived from 88,281 establishment-year observations.

By applying the country-specific probability weights provided by EU-OSHA, we can ensure that our aggregated data accurately reflects changes at the average firm level within each country. This weighting allows us to capture representative changes in “OHS responsibility assignment” despite the lack of a consistent firm-level panel.

Dependent Variable: Worker safety has traditionally been operationalized and evaluated in operations management either through accidents (De Koster et al., 2011; Pagell et al., 2020) or safety breaches (Pagell and Gobeli, 2009; Wiengarten et al., 2017). A breach is typically identified during inspections and may not necessarily have caused substantial harm. Accidents can result in employee injuries, diseases, or fatalities. However, the more general term, “OHS performance” used in the safety literature (Bonafede et al., 2016; Ramli et al., 2011), is not limited to worker safety but also considers employees’ well-being and mental health. While not diminishing the value and importance of these perspectives, given our data and to stay consistent with previous operations management research, this study focuses on accidents as a typical measure to evaluate worker safety. Specifically, we employed question Q160 from the 2019 survey, which asks, “Has absence due to work-related accidents rather increased, rather decreased, or stayed about the same over the last 3 years?” as the dependent variable. We treated the answers “rather increased,” “stayed about the same,” and “rather decreased” by assigning ordinal values of 1, 2, and 3 to these categories, respectively. Similar to the use of Likert-scale data in other studies, this approach assumes equal distance between categories, thereby representing an approximation. We then assessed the probability-weighted means of this measure for each country, which reflect the dependent variables in our estimation.

Explanatory Variables and Controls: Our key explanatory variables explore changes in the roles of people (indicated with the symbol ∆) managing OHS in firms between 2014 and 2019. We employed question Q100 from the 2014 survey and question Q113 from the 2019 survey, which asked whether it is the (1) owner or a partner of the firm, (2) managing director, site, or branch manager, or (4) the health and safety officer, (5) safety representative responding to the survey. The survey also asks respondents, “Is health and safety your main task or just one of a number of tasks you have at this establishment?” with the options “main task” or “one of a number of tasks” (see Q101 in the 2014 survey and Q113 in the 2019 survey), indicating that the person sampled by ESENER to respond, manages OHS. Please note any error in this assumption should downward bias our results and increase p-values. Therefore, the explanatory variables of changes in the job role managing OHS for each of these job categories is then the change in the probability-weighted adjusted means of these categories between 2014 and 2019 for each country. It should be noted that the question provided additional options, (3) another manager, (6) another employee in charge of the subject, or (7) an external health and safety consultant. The first two options are unspecific and the latter delegates responsibility outside the firm. All three forms (options 3, 6 and 7) remain untheorized in our model. Yet, for robustness, we integrate them in our analysis along with the four theorized options (1, 2, 4 and 5).

Next, the moderating variable of establishment size was measured through variable Q102gr, which asked, “How many employees are currently on the payroll of this establishment?” with categorical answer possibilities of 5–9, 10–19, 20–49, 50–99, 100–149, 150–249, and 250+. We again assessed a probability-weighted mean for this measure per country.

Furthermore, as the economic situation of an establishment may influence the importance of OHS in the establishment, and therefore responsibilities as well as accidents, we controlled for the economic environment of each establishment, using question Q400, “How would you rate the current economic situation of this establishment? Is it very good, quite good, neither good nor bad, quite bad, or very bad?” Again, this was assessed as a weighted mean per country. We also controlled for the age of the establishment, as age might affect role definitions and responsibilities in firms, as well as the routines in working processes and resulting accidents. It was assessed using question Q112gr “In about which year did this establishment start to operate? Please include time at previous locations or under a different ownership.” With answer options “before 1990” “1990 to 2015”, “After 2015”.

Further, the nesting of the data within countries controls for all variables at the country level, as it allows for random intercepts for each country, capturing all country-specific characteristics. This approach also allows us to focus more directly on the variables of interest without over-controlling for the country effect. Given that we have 32 country-level observations, the use of only a limited number of controls is a necessity to ensure a robust and meaningful statistical analysis, given the limited degrees of freedom and the risk of overfitting. Therefore, a simple model with limited controls is likely to have a slightly higher bias, but lower variance than a model with many controls and small degrees of freedom, leading to more reliable predictions and inferences.

Summary statistics of the variables and correlations amongst the variables at the country level are reported in Table 3.

The analysis employed instrumental variables regression using a two-stage least squares (2SLS) estimation approach, given concerns over endogeneity due to omitted variables, measurement error, and simultaneity. The OM literature points toward the 2SLS estimation as “the most popular statistical solution to the endogeneity problem” (Ketokivi and McIntosh, 2017, p. 12).

We began by conducting the Breusch-Pagan/Cook-Weisberg test for heteroskedasticity on the models without and with all interactions. The results are reasonably compatible with the null hypothesis of homoscedasticity in our data (p = 0.102 and 0.971) (see section 4.1.). Therefore, we employed conventional standard errors for our main estimations – see robustness test (section 4.2) for estimation results with robust standard errors. The average variance inflation factor (VIF) of the variables is 1.50, and the maximum VIF is 1.80. Thus, increased variance of the estimated regression coefficients due to multicollinearity should not be a serious concern, a conclusion also supported by the correlations (see Table 3).

Next, we explored the risk of endogeneity leading to biased or inconsistent OLS regression results (Lu and Ding, 2018). First, the risk of not having controlled for important omitted variables, that influence both the dependent and independent variables, is plausible in the present study. Second, given the interview-based nature of the data, measurement error might also exist, potentially leading to attenuation bias, where the effect of predictors on the outcome is biased towards zero. Also, the simultaneity risk that arises when work-related accidents have increased – a firm might respond by changing the person who manages OHS – implying that the direction of causality is opposite to what is modeled. This risk is less likely to affect the estimates, yet, as the dependent variable was observed between 2016 and 2018 and the independent variable between 2013 and 2018 (note, the data was collected the year before the survey was published, so the 2019 survey includes data collected in 2018), it cannot be ignored either.

Therefore, we retrieve instrumental variables from within the dataset, as has been put forth in the marketing domain (Van et al., 2013; Becker et al., 2019). Specifically, as instruments, we used the changes in the job roles managing OHS among countries other than the focal country. We distinguish among countries within the EU27 and those outside the EU27. Changes in job roles managing OHS of other countries may be due to the same underlying reasoning, but this reasoning is likely unrelated to the disturbance term of the accident-related absences in the focal country. Therefore, the seven variables for the job roles managing OHS of other countries inside and outside the EU27 should be approximately exogenous to our model specification (note, here we still considered all seven possibilities for our explanatory variable), offering a just-identified model. The Stock and Yogo F-statistic for the seven instruments is estimated at 24.98, suggesting that the set of instruments collectively has enough explanatory power for the endogenous variable. The results of the instrumental variables regression using 2SLS estimation indicate that managing OHS by job role is exogenous for the data at hand (Durbin [p = 0.835]; Wu-Hausman [p = 0.965]). We conclude that endogeneity is not a major concern and followed the approach in Wiengarten et al. (2021) using ordinary least squares (OLS) regression for the main analysis.

To test the hypotheses, we estimated a series of models using OLS. Table 4 (to explore H1) and 5 (to explore H2) present the results of our tests. In this study, we adhere to the recommendations of McShane et al. (2024) regarding statistical reporting by moving beyond binary significance thresholds. Rather than relying on arbitrary cutoffs for statistical significance (e.g. p < 0.1), we report exact p-values in our Tables 4 and 5 to provide a more nuanced interpretation of our results. This approach acknowledges that a single study is never definitive and helps mitigate selective reporting biases. We thereby aim to contribute to a more cumulative and evidence-based understanding of the underlying phenomena.

An initial assessment of the model fit indicates a satisfactory in-sample fit, with values ranging between 0.517 and 0.781. Table 4, Model 1 is the controls-only model. In Model 2 only the explanatory variable “∆Owner/Partner” is added. In Models 3 through 8, we successively include and then exclude alternative changes in employee functions managing OHS. Finally, Model 9 presents the results when the changes in all functions are included simultaneously. We employ this stepwise build-up to examine the robustness of our models and measures, as well as exploring the risk of overextending the data given the small sample.

The negative estimates obtained for “∆Owner/Partner” as observed in all models indicate that, when controlling for establishment size and other factors, economies where more owners/partners have started managing OHS, since 2014, have also experienced more accidents (Model 9, -2.289) with a p-value of 0.078. Overall, we conclude at the establishment level, that having owners/partners manage OHS seems to lead to increased accidents. To further explore hypothesis 1, we examine whether there is a notable difference between the estimates obtained for “∆Owner/partner” in Models 3–8 versus changes in the other functions (Table 4). Formal t-tests for equality of the estimates obtained in Models 3–8 suggest that, with the exception of comparing the “∆Owner/Partner” coefficient estimates to the estimates of “∆Managing Director” (Model 3) (and the untheorized option “∆Ext. OHS Consultant”; Model 8) we find reasonable evidence for differences in estimates (p = 0.313 [Model 3], 0.037 [Model 4], 0.027 [Model 5], 0.048 [Model 6], 0.029 [Model 7], and 0.605 [Model 8]). In other words, increasing the number of owners/partners managing safety, relative to all other theorized job titles except managing directors, seems to have increased the average accidents experienced per establishment in a given country, supporting Hypothesis 1.

In Hypothesis 2, we assume that the efficacy of employees other than the owner managing OHS depends on establishment size. To test this, we successively include and then exclude the interaction effects between size and the various changes in functions managing OHS in Models 10–16 (Table 5). Finally, Model 17 presents the results when all interactions are included simultaneously. H2 only considered changes in OHS officers and employee representatives, but we again opted to explore all answer possibilities for completeness. As before, we employ the stepwise build-up to explore the risk of overextending the data given the small OLS sample.

We find that Models 13 and 14 indicate strong interaction coefficients with opposing signs and small p-values, and the estimates and p-values appear robust when explored in the full model 17. For ease of interpretation, we plotted these interactions using predictive margins with 90% confidence intervals (see Figure 2). The results suggest that in smaller establishments, it is beneficial to have employee representatives manage OHS, while in larger firms, a dedicated OHS officer is preferable. These findings support Hypothesis 2. An alternative estimation approach is reported the  supplementary material, yielding qualitatively similar results.

Worker safety has become an integral part of operations management research and practice. We started this study with the simple observation, that despite all of the research, managerial advancements and policy developments, occupational accidents and injuries are still common across industries and nations. Previous research linking safety to operations largely focused on finding managerial solutions and practices that affect occupational safety at the shopfloor level (e.g. ISO 45001). This is understandable and these efforts have identified important levers to make workplaces safer. However, we followed numerous calls and investigated factors at the management level that affect safety. Specifically, we draw on stakeholder-agency theory and developed and tested a framework to explore how changing the role of the person that manages safety, impacted accident rates.

Our results provide evidence that having owners and partners managing safety is not effective. The results suggest that when more owners or partners manage safety, establishments experience more accidents. This simple finding conveys a powerful message that has important implications from a management, policy and theoretical perspective.

Safety policy and regulation is clear that its intent is to prevent harm to workers (e.g. the Occupational Safety and Health Act of 1970). Equally, it would be the rare manager who claimed anything other than wanting to protect their operational workforce and certainly no manager sets out to harm their workforce. Yet poor safety is a persistent problem, even in our sample from mainly (very) rich countries with extensive safety regulations.

Our research shows that one of the simplest things that firms can do, and regulators and policy makers should encourage or mandate, is to make sure that owners are not managing safety. Our results show that regardless of firm size, firms should not place the owner in the position of managing operations and worker safety at the same time.

As to who should manage safety, our results suggest that this depends on the size of the establishment. In smaller establishments a reduction in accidents occurs when an employee representative manages safety. In larger establishments, a dedicated OHS officer should manage OHS. These differences are likely because the increased complexity of large firms and their processes means that specialist safety training is useful in these contexts. However, these results also suggest that worker representatives, because they are close to the people doing the work and the people most likely to get harmed at work, should be involved in ensuring an operation is safe (much in line with arguments offered in section 5.4 of ISO 45001), even if they should not have final responsibility to manage OHS in larger firms.

Current safety regulations do not specify who should manage safety in an establishment. Having this freedom may help some establishments, especially smaller and younger ones, implement a minimum safety standard as they are likely to have flexible job roles and their expertise is both quickly evolving and may be less formalized from what is typical in large firms.

However, our results suggest that owners and partners should not manage safety. This may seem counterintuitive since safety is on the agenda of most owners, and owners claim to be committed to safety in most firms. Hence, it might seem intuitive for owners who are committed to safety to manage it themselves, as a visible indicator of their commitment. Our results suggest that no matter how well intentioned such an action might be, that it will be counterproductive. It seems to be more effective to have someone else manage OHS, give them the resources they need, and back them up when safety priorities come in conflict with other operational priorities (e.g. Johnston et al., 2020).

These results have two main implications for policy makers and regulators. The first is the fundamental question of how prescriptive safety regulation should be regarding who manages safety. Our results suggest that the choice matters and that as firms become larger, they are best served by having a dedicated safety officer. Hence, policy needs might be best achieved by retaining the flexibility for small firms, but being more prescriptive and mandating a dedicated safety manager for larger establishments.

The second implication for policy makers and regulators is simpler and requires no change in current regulations. Specifically, one of the leading indicators to track which firms may need more attention, because they have increased their likelihood of harming workers, should be a change in who manages safety. When firms change this person or role, regulators should be aware that the establishment’s safety trajectory is likely to change going forward. And when the change is to give an owner responsibility for safety they did not have previously, the trajectory is bound to get worse, indicating that more regulatory oversight is needed. Said differently, our results provide all regulators, even those not looking to change their policy, another way to predict which establishments need more (or less) attention to ensure that workers are protected.

Our study makes multiple theoretical contributions that are particularly important for the social sustainability stream of research within the operations management domain. Social sustainability in operations management research is increasingly concerned with well-being and occupational health and safety (Wiengarten et al., 2021). Our research thus contributes to the ongoing debate as to how to make operations and supply chains a safe workplace. As Longoni and Cagliano (2015) suggested; “social sustainability refers to actively supplying the preservation and creation of skills as well as the capabilities of future generations, promoting health and supporting equal and democratic treatments that allow for good quality of life both inside and outside of the company context” (p. 218). We emphasized that OHS is a critical aspect of social sustainability and has typically been studied through shopfloor-focused aspects such as improvement programs, process design, job design, resource allocation and resources availability (Wiengarten et al., 2021). Expanding on this, our research naturally extends this focus by examining who manages workplace safety, an area that has been largely overlooked in prior research. By identifying patterns that show which job roles are more effective at managing workplace safety, depending on the size of the establishment, we add an important nuance that benefits our theoretical understanding.

Moreover, our study aligns with Corbett’s (2024) exploration of well-being in the operations management context, which organizes operational impacts across individual, group, and societal levels. Specifically, Corbett highlights the significance of “prevention” in promoting sustainable operations and worker well-being. Our findings resonate with this by demonstrating that the assignment of OHS responsibility not only influences individual health and safety, but also extends to the well-being of the workforce as a collective. Thereby, our findings respond to calls for a broader view of social sustainability that incorporates the operational and managerial processes influencing well-being, as described in Corbett’s work. We invite future research to build upon our findings and further explore how the roles and management of OHS impact not just workplace safety but also contribute to other aspects of organizational sustainability.

Further, our predictions were framed using stakeholder-agency theory. The results support the theory’s predictions in the present context. They also suggest an interesting, albeit, small contribution to the theory itself. A foundation of the theory is the unique role of managers, as both agents for the owners and having responsibilities to all other stakeholders (Hill and Jones, 1992). Yet the theory is also clear that they are not agents of the other stakeholders to whom they have responsibilities, since they are not hired by these other stakeholders. In our context, employee representatives take on a double agent role, which calls into question the uniqueness of managers in having responsibility to multiple stakeholders. An employee representative’s control of decision making related to safety makes them an agent for the owners, while they are also explicitly representing (and likely selected by) the workforce. That they can be effective in this context suggests there might be other scenarios where employees could take on a double agency role to improve outcomes; such as reducing absenteeism or increasing workers motivation to engage in continuous improvement, that should matter to both the workforce and owners.

More generally, our results also suggest that rather than using the traditional stakeholder theory, which is common in sustainable operations management research, that the more advanced form of stakeholder-agency theory could be more useful in many applications. This is because the theory simultaneously addresses a pair of fundamental issues which are critical to understanding sustainable operations.

First, sustainability outcomes are judged by a variety of stakeholders who have different opinions as to what is acceptable or not. This is in contrast to our traditional profit-based performance metrics, which are mainly judged by owners and managers. In other words, if we want to understand sustainability and have outcomes that are acceptable to all stakeholders, we will need to move away from the dominant focal firm perspective to including the views of more stakeholders. This is the usual approach to sustainable operations framed using stakeholder theory.

Second, that individual actors, even within a single firm’s operations, often have very different drivers and motivations, which they can and will act upon. This is clear in the literature on internal integration which consistently shows that even employees of the same firm, but in different functions, can have very different perspectives on the firm’s priorities, policies, and practices (e.g. Pagell, 2004). More importantly, agency theory makes it clear that individuals acting in these roles can and will make decisions that are optimal for them, but which may not be for the organization, other stakeholders, or the wider social-ecological system.

It is the second point, on taking actions, that leads us to suggest that typical stakeholder framing used in research on sustainable operations management would be better served by using stakeholder-agency framing. We suggest this because research on sustainable operations that takes the perspective of achieving the focal firm’s goals and that is framed using stakeholder theory, often concludes that achieving sustainability goals will require engaging or cooperating with a wider range of stakeholders (e.g. Bansal, 2005), often in large networks such as multi-stakeholder initiatives (e.g. Matzembacher et al., 2021). These suggestions are fine on their own, but when they are made from a focal firm’s perspective without exploring how the other stakeholders will view the same issue, and more importantly what goals they will be trying to achieve, it gives the impression that the focal firm can dictate terms and easily achieve its sustainability goals if it just engages with others. Stakeholder-agency theory avoids this issue and can provide more nuanced predictions on how various members of the supply chain and other stakeholders will engage with each other when trying to achieve sustainability goals.

This study aimed to deepen our understanding of the impact that assigning responsibility for managing OHS has on safety outcomes. We specifically examined how entrusting safety management to different job holders, particularly comparing owners to managers and workers, influences the incidence of workplace accidents. Furthermore, we explored how the effectiveness of these assignments varies depending on the size of the establishment.

We found that having owners and partners in charge of managing safety is not beneficial. When owners are responsible for managing safety, accidents increase even when controlling for establishment size. Considering the effect of establishment size, we found that in smaller establishments, it is more beneficial to have employee representatives in charge of safety management. However, in larger firms, a dedicated safety officer should hold responsibility.

This research is not without limitations. First and foremost, while the use of the ESENER surveys provided several advantages – such as their representative nature and the large number of entities surveyed – it also limited us in the choice of variables to test our theory. The present study used accidents as its indicator of OHS performance. However, OHS performance is a multidimensional construct that also includes breaches, as well as considerations of employees’ well-being and mental health. We encourage future research to consider and integrate these other dimensions of OHS performance to offer a more comprehensive picture in the field.

Second, we recognize that our study opens avenues for future research to explore the distinction between being tasked with OHS and formal titles for OHS more deeply. Currently, our theory and analysis do not address the impact of formal OHS titles on OHS performance. We recommend that future studies, if possible, collect data that can directly measure formal OHS assignments.

Relatedly, this research was based on a major study over two waves (2014 and 2019) looking at how European workplaces manage safety and health (ESENER). A major limitation of the analysis of existing data sets is that the researchers who are analyzing the data are not the same individuals as those involved in the data collection process. Therefore, they are probably unaware of study-specific nuances in the data collection process that may be important to the interpretation of specific variables in the dataset. Sometimes, the amount of documentation is daunting (particularly for complex, large-scale surveys conducted by government agencies), so users may miss important details unless they are prominently presented in the documents. Succinct documentation of important information about the validity of the data (by the provider) and careful examination of all relevant documents (by the user) can mitigate this problem.

In the case of the ESENER surveys this documentation exists. The ESENER project team provides detailed documentation on the overall population, the respondents, the sampling strategy, the questionnaire design (consisting of a cognitive pre-test with 36 in-depth face-to-face interviews), a translatability assessment of the English master questionnaire and a pilot field test in all countries surveyed, with 30 interviews per country. While we did not collect the data, we are confident that we have the necessary information to interpret the variables in the data set. Furthermore, ESENER surveys were conducted in Europe, which places considerable emphasis on the regulatory framework to support safe working practices. Future research needs to explore if the results hold outside of this region and are relevant to other jurisdictions, where OHS regulatory practices are likely to be less extensive. In addition, in the ESENER studies to date, data on firm costs, firm performance and return on assets has not been collected making it difficult to tie safety practices to other firm outcomes. Similarly, the lack of a means to identify firms makes it impossible to link the ESNER data to any other data, let alone to the firm-level data over time. A future wave of the survey which included other firm outcomes and the identity of the firms participating would provide additional insights and allow researchers to triangulate the OHS measures with other outcomes and data.

Next, while this study focuses on the impact of the role of the person managing OHS on workplace accidents, we acknowledge that other factors – such as local management practices, and regulatory frameworks (Bonafede et al., 2016; Ramli et al., 2011) – also influence OHS performance. These broader variables fall outside the scope of this research and were not covered in the ESENER surveys. Future studies should, if possible, consider integrating these factors to provide a more comprehensive view of how they interact with OHS management.

Finally, the ESENER research design made an explicit trade-off to capture the firm level of analysis but provides no information on how individuals experience their jobs and the outcomes of those experiences. Future research that captures both firm and worker perspectives simultaneously will be needed to more fully understand the OHS measures, as well as the impact of health and safety practices on both worker and firm outcomes.

Despite these limitations, the study seeks to make an important contribution in advancing our understanding of how to effectively enhance safety in organizations by providing initial insights into the question of who should be made responsible for managing safety to reduce accidents. This perspective contributes to the operations management literature by introducing organizational job assignments as a relevant construct, complementing existing explorations in our domain that have primarily focused on operational-level traits. The findings not only enhance scholarly and managerial knowledge but also offer clear guidance to policymakers for improving future guidelines aimed at creating safer and sustainable workplaces.

As reported in section 4 of the manuscript, we conducted the Breusch-Pagan/Cook-Weisberg test for heteroskedasticity on the models without and with all interactions. The results did not reject the assumption of constant variance. Thus, we employed conventional standard errors for our main estimations (Tables 4 and 5). However, a check of Model 11 rejected this assumption (p = 0.045). Therefore, we also explored our estimation results with robust standard errors. The results, listed in Table A1, remain qualitatively consistent, generally exhibiting smaller standard errors and, consequently, smaller p-values.⁶