Institutional theory refers to managing an organization's environment. Phillip Selznick initially defined its fundamental concept in the year 1949. Subsequently, John Meyer and Brian Rowan, as well as Paul J. DiMaggio and Walter W. Powell advanced it in 1977 and 1983, respectively (27). This theory not only explains how certain institutions influence the organization's behavior and decision-making process in which it operates but also why organizational structures and practices change (28, 29). Six key concepts make the basis of this theory: legitimacy, isomorphism, rational myths, loose coupling, diffusion, and the infusion of value (27). Collectively, these concepts ensure the survival of an organization in the institutional environment with both technical (i.e., capital and labor) and social (i.e., legitimacy and status) perspectives. Therefore, legitimacy and isomorphism are considered central assumptions of this theory.

As an essential organizational component, OHS implementation is seen as a social and technical aspect of managing an organization's environment. It reflects how an organization promotes health and safety at the workplace, which impacts several dimensions of performance, such as productivity, absenteeism, and employee satisfaction (30). Jilcha et al. (31) stated that implementation of the OHS requirements is not the only responsibility of internal stakeholders, but external stakeholders such as governing bodies, unions, insurance agencies, and other higher institutions are also equally responsible. For this reason, the institutional theory could result in better OHS implementation due to the effective involvement of both internal and external stakeholders.

Furthermore, various past research work [e.g., (32–38)] in occupational health and safety have employed this theoretical ground to understand the organizational behavior in responding to the OHS issues. The other strength of this theory is that it supports organizational transformation, even if the technical or economic advantages are lacking (39, 40). Therefore, this is also the reason that the present study employs the institutional theory to explain the influence of different institutions (i.e., factors) on OHS implementation.

Implementation style is a process of putting a plan into practice. Pollitt and Bouckaert (41) emphasized that understanding the implementation dynamics of a program requires an awareness of implementation styles. Similarly, Andrews et al. (22) stated that the successful execution of a plan or program depends on a particular implementation style, which, as a result, has substantial significance for the performance of an organization. Several styles of implementation exist in the literature. However, researchers explicit them into two main categories: rational style and incremental style (22, 42). The rational style is a systematic and well-planned approach that prioritizes organizational changes and gets people to follow the precise procedure to adopt these changes. Whereas incremental style is a fluid nature style of organizational change which encourages modification step by step based on the ground realities (22). Past studies revealed mixed findings on implementation styles. For example, Andrews et al. (43) used these implementation styles to check service performance in public organizations. The study revealed that no implementation style by itself is effective, but that style is contingent on organizational strategic orientation. Similarly, Andrews et al. (22) studied the individual and combined effect of rational and incremental implementation styles in public sector organizations in Turkey. It found that the organization performs better that use a combined implementation style more than those that emphasize a single style. Whereas the absence of a style may lead to worse performance. On the contrary, Balogun and Jenkins (44) explain that the rational style gives better results in reactive situations or crises. In contrast, the incremental style best suits managers to adopt in proactive situations. As the OHS program not only provides controls in proactive ways but also enables organizations to handle hazardous situations by remedial measures. Therefore, analyzing both implementation styles is important to assess which style leads to better OHS implementation. Thus, the following hypotheses are proposed:

Safety accountability is essential for OHS implementation. It creates and establishes in each employee a sense of obligation to perform assigned tasks effectively and efficiently. For example, management is responsible to design policies and procedures, clarify roles and responsibilities, provide appropriate training and infrastructure, and granting employees sufficient authority to do their part in OHS implementation well (45). Similarly, it is the responsibility of safety personnel to encourage management and all other employees to OHS implementation. It is because OHS implementation is an organization-wide effort. The safety department or safety personnel cannot ensure its success on their own alone. Therefore, everyone in the organization is responsible for its success (46). However, without safety accountability, a safety program implementation is likely to be unsuccessful or only ceremonial. Kim et al. (46) argued that an organization cannot achieve safety excellence until its management and all other employees are not accountable for their measurable responsibilities. Moreover, Mlynek (47) stated that safety accountability is not a punitive strategy but rather a proactive means of fostering a culture of responsibility. As such, the following hypothesis was developed:

Mimetic motives refer to the extent to which an organization implements a program by mimicking the best practices of leading organizations in an attempt to gain legitimacy. This trend is most common in developing countries because it saves both cost and time (48). Moreover, Ansari et al. (49) argued that mimicking behavior consist of two perspectives; laws & regulations, and societal norms & values. The first perspective stems at to achieve legitimacy, which is vital in securing resources for continued existence over the long run (29). Whereas the second perspective is of moral origin, which emboldens ethical and responsible behavior regardless of the outcomes (28, 50).

McBain-Rigg et al. (51) identified peer-to-peer networking, industry representatives, government, and media as the most influential agents for safety implementation in the agriculture and fishery sector of Australia. Similarly, Iatridis et al. (52) found that governmental authorities, NGOs, and the local community are mimetic motives that influence them to commit fully to certify management standards, such as OHSAS 18001. Another perspective proposed by DiMaggio and Powell (39), is that mimetic motives are especially relevant in those organizations where technology is poorly understood, the environment is uncertain, and goals are ambiguous. Thus, based on institutional theory, this study understands mimetic motives as institutional motives that push organizations to adopt OHS implementation as strategy implementation. Furthermore, this study expects that the organizations that have a strong influence from mimetic motives, their decision-making styles for implementation, and safety accountability would be high for OHS implementation. This indicates moderating effects of mimetic motives on the relationships between exogenous and endogenous variables. Therefore, it is necessary to explore the role of mimetic motives in implementing the OHS. Thus, the following hypotheses were developed.

This study focused on the province of Punjab (31.1704° N, 72.7097° E). It is most prominent in agriculture production and has the most significant share in the national economy compared to other provinces. Therefore, the targeted population was all agricultural farms in this province. The latest provincial government statistical record showed that there are 5,249,800 agriculture farms in the whole province (53). Although all districts in the province of Punjab are notable for different varieties of agricultural products. However, the district Bahawalnagar, Bahawalpur, Bhakkar, Sargodha, Multan, Pakpattan, Okara, Jhang, and Layyah were chosen purposively as the study area because these districts contain large farms, and it is assumed that large agriculture farms have dedicated safety personnel and resources to cope with OHS issues.

The agriculture farm was chosen as the unit of analysis because this study is aimed at providing a holistic view of OHS implementation in the agriculture sector of Pakistan. Moreover, the respondents of this study participants were safety person-in-charge (i.e., owners, OHS managers, OHS officers, or OHS representatives). These individuals were selected as respondents for three main reasons. First, they are linked with all activities of other departments and speak for OHS at the executive level. Second, they have sufficient practical and professional knowledge to implement the latest OHS programs. Third, these people provide reliable information due to their key role in OHS implementation and continuous monitoring.

The G^*Power software of version 3.1 was used to calculate the minimum sample size, as recommended by Joseph et al. (54) and Memon et al. (55). A sample size of 153 was calculated with the settings of statistical power of 95%, probability error of 5%, and with a medium effect size of 0.15 for four predictors and three interaction terms. The agricultural farms were selected through purposive sampling. This technique was applied because a comprehensive and up-to-date list of agriculture farms was not readily available. Kumar (56) described that the purposive sampling technique is more suitable, where a complete list of the total population is unavailable to researchers.

The questionnaire was divided into two sections. The first section contains the demographics of the respondents, such as gender, age, qualification, designation, working experience, nature of business, farm type, and number of employees. Whereas the second section includes the measures of study constructs. The measures of both rational style (5-items) and incremental style (5-items) were adapted from Andrews et al. (22). The items of safety accountability were adapted from Molenaar et al. (57), which contains four items in total. Similarly, the three items for measuring mimetic motives were adapted from Hillebrand et al. (58). Finally, the intention for OHS implementation was measured with four items adapted from Hossain et al. (59). Moreover, the data of the second section was collected on a 7-point Likert scale from 1 (“strongly disagree”) to 7 (“strongly agree”). Keeping the education level of respondents in mind, the questionnaire was translated into Urdu and sent to one academic person and two OHS professionals (i.e., OSH consultants) for content validity. After their feedback, a pre-test with 37 actual respondents was also conducted to confirm the comprehension and reliability of the questionnaire. No major amendments surfaced, aside from a slight change in wording. Moving forward, a paper-based questionnaire was personally handed over to respondents. The questionnaire distribution with this method was aimed at generating a higher response rate (60). A cover letter, stating the confidentiality of data and the nature of the study, was also attached to each questionnaire. A total of 400 farms were targeted through different channels. Moreover, data was collected in a single sitting, specifically from January 2022 to April 2022.

The PLS-SEM technique was adopted to examine the theory-based conceptual model that is specified in Figure 1. More importantly, this technique provides robustness in analysis and does not rely on data distributional assumptions (54). Two steps are strictly followed. In the first step, the reliability and validity of constructs were assessed through outer loadings, composite reliability (CR), and average variance extracted (AVE). HTMT was also examined to confirm the establishment of discriminant validity. This is then followed by structural model estimation. This step includes the assessment of the coefficient of determination (R²), path coefficient (β), effect size (f²), predictive relevance (Q²), model fit, and PLS prediction. Furthermore, a bootstrapping procedure with 5,000 iterations was used to test the significance of hypothesized relationships. Besides, moderation was evaluated using an orthogonalization approach (54). Data were analyzed using SmartPLS of version 3.3.7 (61) and SPSS of version 25.

Most of the participants were men (97.91%) with an age bracket of 31–40 years (47.04%) and had a bachelor's qualification (72.82%). Those who responded as safety in-charge persons (i.e., safety manager, safety officer) accounted for 47.73%, while those who responded as safety representatives or owners were 52.27%. The majority of farms had employees below 50 (79.10%) and livestock nature of business (41.11%). Moreover, private agriculture farms participated in this survey the majority (95.47%). The detailed findings of demographics are illustrated in Table 1.

The data on all study constructs were collected in a single sitting and from the same respondent, common method bias could be a potential problem. Three remedies were used to alleviate this problem. First, the participants were informed that there were no wrong or correct answers and that their responses would be anonymous and would not use for their performance evaluation. Additionally, the readability and comprehension of all the items were improved by keeping the question specific and concise by avoiding ambiguous terms (62). Second, Harman's Single Factor test was carried out to check the common method bias in the data. In this approach, all eighteen items were loaded to a single factor using principal component analysis with the varimax rotation method. The single factor explained variance of 28.271%, far below the 50% threshold. Third, a full collinearity test for both endogenous and exogenous variables was applied to examine this problem. The pathological (inner) VIF for all constructs ranged from 1.091 to 1.121, which is less than 3.30 (63), confirming again that CMV is an unlikely threat to this study.

The framework of the current study contained three exogenous, one endogenous, and one moderating construct. All were proposed as first-order reflective constructs. Therefore, the recommendations of Joseph et al. (54) were followed for measurement model assessment. Convergent validity and internal consistency reliability were assessed in the first step, while discernment validity was assessed in the later step. For convergent validity, the factor loading of all items must be higher than 0.703, and the average variance extracted (AVE) should be equal to or higher than the threshold of 0.500 for each scale. Whereas Cronbach's alpha and the composite reliability (CR) of all the reflective constructs must exceed 0.700 to show internal consistency reliability (54). The findings are tabulated in Table 2. It revealed that Cronbach's alpha, CR, and AVE values met the threshold criteria. Similarly, the factor loadings of all indicators, excluding RS5, were also found to be above 0.703. However, RS5 was retained due to three reasons. First, this indicator was vital and of absolute importance for the study. Second, the value of factor loading was very close to the threshold. Third, the AVE value of the relevant construct was meeting the threshold criteria without excluding it. Thus, the constructs of the proposed model were found to have satisfactory convergent validity and internal consistency reliability.

Henseler et al. (64) suggest that the Heterotrait-monotrait (HTMT) ratio of correlation should be tested for discriminant validity. It reflects the distinction of the framework's constructs (63, 64). According to recent literature, the HTMT ratio is more precise and preferred than the Fornell and Larcker criterion and cross-loading method (54). The maximum threshold value of 0.85 indicates sufficient discriminant validity of the constructs (65). The results presented in Table 3 demonstrate that HTMT < 0.85, meaning that the study's constructs are distinct. Hence, sufficient discriminant validity exists.

Moreover, multicollinearity was also checked through the outer variance inflation factor (VIF) of all items before testing the structural model. Joseph et al. (54) recommend that checking VIF reduces the biasness in the hypothesized relationships. The values of outer VIF ranged from 1.443 to 2.689 (see Table 2), which were far below 3.3 (63), showing no multicollinearity problem in the data.

Following that, the model's path quality, exploratory power model, and strength of the proposed hypotheses were examined in the structural model assessment. The path quality was estimated through PLS blindfolding procedure using the Q² value. The Q² value greater than 0 shows a predictive accuracy of the model (65). Additionally, the Q² value of 0, 0.25, and 0.50 represents small, medium, and large predictive relevance of the path model. Table 4 revealed a Q² value of 0.260 which means that the model had a medium-size predictive accuracy.

The exploratory power of the model was estimated using coefficients of determination (R²) of the endogenous construct (i.e., OHS implementation). The R² value tells the total variance explained in the endogenous construct by exogenous constructs (66). The R² value for OHS implementation indicated a medium exploratory power without a moderator and substantial with a moderator (see Table 4). It means that mimetic motives as a moderator increase the 6% exploratory power of OHS implementation, which is accounted for by the organization's decision-making style for implementation (i.e., rational and incremental) and safety accountability. Next, the strength and statistical significance of the proposed hypothesized relationships were tested. For this purpose, t-statistics and path coefficients (β) were evaluated using a bootstrapping procedure with 5,000 subsamples (63). Table 5 revealed that rational style (H1: β = 0.353, t = 6.069, p < 0.001), incremental style (H2: β = 0.319, t = 5.085, p < 0.001), and safety accountability (H3: β = 0.120, t = 2.102, p < 0.05) are having positive significant impact on OHS implementation. Thus, all three direct hypotheses were supported.

Regarding moderation, the interaction terms were created in SmartPLS and analyzed using an orthogonalization approach. Joseph et al. (54) argued that the orthogonalization approach is an extension of the product indicator approach and gave better results than other approaches (i.e., two-stage, and product indicator). The results tabulated in Table 6 showed that mimetic motives (H4: β = 0.138, t = 2.113, p < 0.05) positively influenced the relationship between rational style and OHS implementation. Whereas mimetic motives (H5: β = −0.170, t = 2.275, p < 0.05) negatively moderate the relationship between incremental style and OHS implementation. However, contrary to hypothesized relationship, mimetic motives (H6: β = 0.094, t = 0.682, p > 0.05) were not found to have a moderating effect on the relationship between safety accountability and OHS implementation in either a positive or negative direction and hence rejected. Thus, two moderating hypotheses were found to be supported, and one was rejected. In addition, moderating effect of mimetic motives was also analyzed using a simple slop test. Figure 2A shows that the interaction effect of mimetic motives is consistent with the prediction of H4. The upward increasing slope showed that rational style and OHS implementation are positive for both low and high mimetic motives. However, OHS implementation is higher when there is a low rational style but for those with low mimetic motives. On the contrary, OHS implementation increased through rational style increased for those agriculture farms with higher mimetic motives. Similarly, Figure 2B shows that when the involvement of mimetic motives is low, the impact of incremental style is higher on OHS implementation. However, when mimetic motives are high, then the relationship between incremental style and OHS implementation is weakened.

More to this, effect size (f²) was also calculated based on the recommendation of Chin (67). It measures the strength of a specific exogenous construct on an endogenous construct by mean change in the coefficient of determination (R²). Cohen (66) described the threshold of 0.35, 0.15, and 0.02 for substantial, medium, and small effect sizes. The statistics in Table 5 are indicative of small to medium effect size.

Likewise, the goodness of the model was also assessed in this study. A long list of fit indices such as standardized root-mean-square residuals (SRMR), unweighted least square discrepancy (d_ULS), geodesic discrepancy (d_G), Normed Fit Index (NFI), RMSEA, and CFI have been proposed in the literature. However, the standardized root-mean-square residuals (SRMR) index proposed by Hu and Bentler (68) is dominant as an approximate model fit criterion in the PLS-SEM context (69). Hu and Bentler (68) suggested a cut-off value of SRMR ≤ 0.08 for a good fit. The SRMR = 0.055 was estimated for this study, well below the threshold, indicating a good fit for the proposed model. It is worth mentioning that the absolute application of any fit measure remains not fully developed, and GoF assessment is unnecessary in general in PLS-SEM (65).

In the end, PLS Predict was used to check the out-of-sample predictive relevance (power) of the model with a default setting of k = 10 (70). As QPredict2 > 0 and the prediction errors in this study are highly symmetrically distributed, a comparison of RMSE values of PLS-SEM and linear regression model (LM) was held as per the recommendation of Shmueli et al. (70). The results in Table 6 show that all indicators of OHS implementation hold true for PLS-SEM < LM, which confirmed the medium out-of-sample predictive power of the model.

This study has fourfold theoretical implications. First, it adds to the body of OHS literature by combining organizational decision-making styles, safety accountability, and mimetic motives for OHS implementation through the lens of institutional theory. Second, although the importance of safety management practices for OHS implementation has been proven in past studies, these constructs, to the best of our knowledge, have been brought together into a single model for the first time. Third, the organization's style of implementation and safety accountability are important predictors in achieving OHS implementation. However, this study contributes to the sense that the choice of implementation style depends on the organizational competencies and priorities. Finally, previous research such as Iatridis et al. (52) and Hillebrand et al. (58) have examined the interactive effect of mimetic motives in the context of the certified management system and customer relationship management implementation, respectively. This research has looked at its moderating effect in the field of OHS and has discovered that mimetic motives play a significant role in OHS implementation in the Pakistani agriculture sector.

Overall, the findings of this study have a number of practical implications for OHS professionals, OHS practitioners, and national authorities for better OHS implementation. First, the findings of this study might enable OHS professionals in designing and developing an OHS implementation program by keeping in mind the organizational competencies and priorities. Second, the findings may help OHS practitioners to execute OHS plans and procedures in accordance with the duties and responsibilities that have been assigned to them. Third, it may enable the national authorities in making better policies and decisions about OHS training, field demonstrations, and technology selection for OHS implementation. Fourth, it may help agriculture organizations to be vigilant about safety-related market developments, especially their competitors' best practices. This could help them to stay in business and gain legitimacy in the long run. Finally, the findings provide the liberty to agriculture organizations to choose an implementation style for OHS implementation as per their competencies, environment uncertainties, and their desire to be like other organizations.