TBL theory, given by Elkington (1998), suggests that enterprises should commit to delivering economic, social, and environmental performance. This theory supports the continued adoption of green practices, which is the environmental dimension, promoting practices that lessen ecological interference (Jum’a et al., 2022). GMPs diminished harm to the environment within the mining industry, thereby entrenching the TBL’s environmental goals as noted by Hilson (2008) and Tost et al. (2018). CSR aligns with the social dimension, promoting the positive contribution of business to society (Pan et al., 2021). CSR activities are not exclusively linked to the social dimension but also to the environmental dimension since CSR promotes green innovation and GOC by integrating sustainability issues into corporate strategy, as underlined by Aguinis and Glavas (2017) and Carroll (2021). The economic dimension is represented by green innovation and the GOC of employees, which influences both the dimensions of sustainable business growth together with efficiency (Fernando et al., 2019; Hendarjanti, 2022). According to Cainelli et al. (2015) and Cheng et al. (2014), green innovation is essential in minimizing environmental impacts and enhancing competitive advantage. It is also because GOC, highlighting a firm’s commitment to sustainability, creates a culture of environmental care that fuels green innovation and leads to CSD. Moreover, GOC, emphasizing a firm’s dedication to sustainability, cultivates a culture of environmental responsibility that drives green innovation and contributes to CSD, as noted by Dyllick and Muff (2015) and Lozano (2015). Together, these elements foster CSD, which aims to achieve a balanced integration of TBL principles (Liu et al., 2024). Ultimately, CSD is simply a junction of economic performance with environmental and social responsibilities; it can be placed in line with the holistic approach to sustainability advanced by Dyllick and Muff (2015). According to Bansal (2005), firms that apply TBL to their business conduct have the ability to contribute toward long-term sustainability by reducing risk and tapping new market opportunities. Business integrations through GMPs, CSR, green innovation, and GOC can result in a synergistic effect promoting comprehensive sustainable development.

Only a quantitative methodology was used to conduct this study. This study was conducted based on the fact that nearly 1,500 coal mining companies are operating in China (Textor, 2024a) and they employ more than 1.5 million workers (Caixin, 2023). Since the number of organizations is not large, therefore to ensure impartiality, probability sampling was used in the form of simple random sampling, due to which generalization to the whole population could be asserted (Azhar et al., 2024). The sample size followed Hair et al.’s (2011) “10-times rule,” which states that the sample size should exceed ten times the number of model connections to any latent variable. Pesämaa et al. (2021) recommended 10 responses per item. Therefore, our study aimed for a minimum of 340 responses, which corresponds to 10 responses per item across 34 items among 5 variables.

The data collection period lasted 4 months, from July to October 2023, in multiple waves. Data collection for independent construct GMPs was done in Wave 1, mediators green innovation and GOC in Wave 2, moderator CSR in Wave 3, and dependent constructs CSD in Wave 4. Each wave was 1 month in duration, with two reminders after 10 days. To maximize the response rate and generalizability, the study distributed 600 questionnaires through email, WeChat, and personal visits. Among the top 5 companies which are listed on the Fortune China 500 ranking, China Shenhua Energy, China Coal Energy, Yankuang Energy Group, Shaanxi Coal Industry, and Huaibei Mining Holdings Co., Ltd. (Textor, 2024b). The study distributed 50 questionnaires each; among other companies, a single questionnaire was shared to reduce the probability of bias issues. Despite some limitations, self-reported data and self-administered questionnaires remain very common and appropriate in data collection for behavioral research (Masud et al., 2023).

A structured, closed-ended questionnaire was administered with assurances that the information supplied would be kept confidential and used for matching purposes only and not given to any third party. A web-based survey was created using Google Forms; after that, links for filling out the survey were shared via email and WhatsApp, quite a common method these days (Fan et al., 2024). Reasons for using an online questionnaire included that it was cost-effective, had a reputation as being a useful tool, ease of distribution, its interactive nature, the increased response rate, and was faster compared to traditional tools for conducting survey research (Qalati et al., 2023). The IP address of each participant was recorded to prevent one respondent from clicking on links in each wave, and respondents were informed about this.

Of the 1000 approached participants, 539 responses were received after the end of Wave 4. However, 28 mismatched IP addresses were identified, leaving 511 valid responses. According to Malhotra and Grover (1998), response rates below 20% are unsatisfactory, while rates between 30% and 70% are acceptable (Masud et al., 2023). Therefore, the response rate for this study is satisfactory. Among the respondents, 398 were male and 113 were female.

This questionnaire was in two different parts: parts 1 and 2. Part 1 was on the descriptive information relating to the respondents, and part 2 was on the research variables. Part 2 had 8 items: 3 items for each dimension of GMPs adapted from Onubi et al. (2020a) and Onubi et al. (2020b); 4 items measuring green innovation adapted from Al-Swidi et al. (2024); 3 items assessing the employees’ GOC adapted from Sharma et al. (2021); 5 items measuring CSR adapted from Masud et al. (2023); and 8 items evaluating CSD covering three dimensions—economic, environmental, and social—adapted from Afum et al. (2020) and Bansal (2005). Items were all rated on a five-point Likert scale, ranging from “1” (totally disagree) to “5” (totally agree).

Even though data for the constructs were collected from the same respondents simultaneously, it was argued that there could be a possibility of common source bias (Onubi et al., 2021). To avoid this, scholars supplement that data should be collected in different waves, as we did in our research since data for the constructs is collected from the same respondents simultaneously (Qalati et al., 2023). In this regard, Harman’s 1-factor test has been performed, as suggested by (Podsakoff et al., 2003), and recently used by several studies (Masud et al., 2023). However, this approach has been criticized by Pesämaa et al. (2021). According to this method, the single factor derived from all major constructs should account for less than 50% of the variance. In this study, it has been found that only one component explained 34.40% dispersion which is below the 50% threshold. Finally, the study used the variance inflation factor which is recommended while using partial least square structural equation modeling. The retained values were <3.33 recommended standard suggested by Cao et al. (2024).

A partial least squares structural equation modeling was estimated using SmartPLS 4.1 for several reasons, which have many advantages over covariance-based SEM, especially in the case of an exploratory nature of research and when dealing with complex models or smaller sample sizes (Cao et al., 2024; Qalati et al., 2023). In particular, PLS-SEM is of interest because it supports formative measurement models and has fewer requirements for data distribution and sample size compared to the other covariance-based approaches (Hair et al., 2019). It is a method of variance-based approach that maximizes the explained variance of the dependent constructs, hence quite relevant for predictive modeling and theory development (Henseler et al., 2015). While CB-SEM focuses on model fit, thus requiring a larger sample size for reliable results, PLS-SEM can become more flexible and robust to the issues of complex relationships and non-normal data distribution, therefore able to arrive at more accurate and reliable insights in practical research scenarios (Hair et al., 2019; Onubi et al., 2021; Qalati et al., 2023; Simmou et al., 2023).

First, in PLS-SEM using SmartPLS 4.1, assessment of the reflective measurement model is necessary to ensure the reliability and validity of the constructs. This is done in a number of ways: First, internal consistency reliability needs to be checked using Cronbach’s alpha (CA) and composite reliability (CR). The acceptable values for both are larger than 0.7 (Hair et al., 2019). Composite reliability is often preferred since it takes account of different outer loadings of the indicators. Second, the convergent validity is checked through the average variance extracted (AVE), which needs to be above 0.5 (Cao et al., 2024; Hair et al., 2019). This is for the simple reason that the construct explains more than half of the variation of its indicators. Third, the respective outer loading of each indicator needs to be significant, but also ideally above 0.65, ensuring that each item reliably measures the intended construct (Hair et al., 2019; Onubi et al., 2021). This research removed two items of the GMPs construct due to a value below 0.65 Table 1 results evidenced that our research reflective model retained values for AVE, CA, CR, and factor loadings.

Another important part of the measurement model assessment is discriminant validity. It is suggested to be checked according to the Fornell-Larcker criterion, where the square root of the AVE for each construct has to be larger than the highest correlation with any other construct (Fornell and Larcker, 1981). A more recent approach to this is the Heterotrait-Monotrait ratio with a threshold value below 0.9, to reflect adequate discriminant validity (Hair et al., 2019; Henseler et al., 2015) (see Table 2). Tables 1, 2 outcomes ensure a reliable and valid reflective measurement model, thereby supporting the robustness of the subsequent structural model evaluations.

After the reflective measurement model has been validated, the structural model is estimated to test the hypothesized relationships between constructs and the predictive power of the model. Some important criteria suggested and used to evaluate a structural model are path coefficients, coefficient of determination, effect sizes, and predictive relevance (Cao et al., 2024). First, the study used bootstrapping features with 5000 sub-samples in SmartPLS 4.1 to check that the hypotheses of the study were tested correctly and that the model path coefficients were significant (p < 0.05) (see Table 3; Figure 2). According to Table 3, GMPs had a significantly positive influence on green innovation (β = 0.493, t = 8.879, p = 0.000), CSD (β = 0.195, t = 3.522, p = 0.000), and GOC (β = 0.476, t = 9.858, p = 0.000), respectively; hence, supported the H1a–c. Besides, green innovation significantly affected CSD (β = 0.262, t = 5.772, p = 0.000) and GOC (β = 0.293, t = 5.944, p = 0.000); in this respective confirming H2a and H2b. The findings also confirmed that GOC significantly impacts CSD (β = 0.233, t = 5.186, p = 0.000); thus, validating H3. Mediation and moderation effects hypotheses were also supported given their level of significance.

On the other hand, while the R ² value indicates the amount of variance explained by the independent variables in the dependent variable, a weak relationship is when it equals 0.25, a moderate one is when it is 0.50, and a substantial one if it is 0.75, according to Hair et al. (2019). The results are shown in Table 3; Figure 2, which indicate R ² values of 63.5% in green innovation, 52.7% in GOC, and 67.2% in CSD. Typically, 67.2% state that GMPs, green innovation, GOC, and CSR are responsible for 67.2% of changes in CSD. It furthers that they have moderate relationships with CSD (Hair et al., 2019).

The effect sizes, f ² , are also suggested to be used to assess the impact of each exogenous construct on the endogenous construct (Hair et al., 2019). The f ² value is interpreted as follows: 0.02 ≤ f ² ≤ 0.15 represents a small effect; 0.15 ≤ f ² ≤ 0.35 represents a moderate effect, and f ² ≥ 0.35 represents a large effect. Table 3 shows that GMPs have a moderate effect on green innovation with f ² = 0.224 and GOC with f ² = 0.196 and a small effect on CSD with f ² = 0.031. In addition, green innovation has a small effect on CSD with f ² = 0.074 and GOC with f ² = 0.074. Moreover, GOC also has a small effect on CSD with f ² = 0.066. Lastly, CSR also has a small effect on the relationship of GMPs with green innovation (f ² = 0.025) and CSD (f ² = 0.021).

Besides, predictive relevance Q ² is assessed with a blindfolding procedure; values above zero would mean that the model has predictive relevance for the respective endogenous constructs (Hair et al., 2019; Qalati et al., 2023). Our study Q ² values range from 0.326 to 0.411, which is above the benchmark of zero (see Table 3). Moreover, these estimates assure that the structural model is not only statistically significant but has practical meaning, too, and therefore, the insights gained into the nature of the relationships between constructs are robust.

The assessment of the green innovation and GOC mediation effect between GMPs and CSD includes measuring their indirect effect. This procedure involves testing the significance of direct paths from GMPs to green innovation, GOC, and CSD, and then the paths from green innovation and GOC to CSD. Bootstrapping yields confidence intervals and p-values for these indirect effects; a p < 0.05 is considered to be a significant mediation effect. Table 3, indicates that all indirect hypotheses H4a–H4c were supported. The green innovation significantly and indirectly affected the relationship of GMPs with CSD (β = 0.129, t = 5.101, p = 0.000). Besides, GOC also indirectly impacted the connection of GMPs with CSD (β = 0.111, t = 5.130, p = 0.000). In addition, green innovation and GOC also sequentially–partially mediated GMPs link with CSD (β = 0.034, t = 3.459, p = 0.001).

Moreover, the variance accounted for (VAF) is computed to quantify the extent of mediation; the VAF ranging from 20% to 80% represented partial mediation, above 80%, it is referred to as full mediation, and below 20% indicates no mediation (Cao et al., 2024; Qalati et al., 2023). Specifically, this refers to whether the addition of green innovation and GOCC significantly increases the explanatory power of the model, as expressed by changes in the R ² value of CSD. V A F G I N N = I n d i r e c t   e f f e c t T o t a l   e f f e c t = 0.129 0.622 = 20.7 % V A F G O C = I n d i r e c t   e f f e c t T o t a l   e f f e c t = 0.111 0.344 = 32.6 %

The above results evidenced the confirmation of mediation that GMPs have direct and indirect effects on CSD through improvements in green innovation and increasing employee GOC to sustainability. Given the VAF test values which fall in the range of 0.20–0.80, it is concluded that these factors partially mediated the relationships.

The moderation effect of CSR on all the links between GMPs and green innovation, CSD, and GOC can be evaluated as follows: To begin with, it is required to create the interaction terms by multiplying the standardized scores of the moderator CSR with the independent variable GMPs. The next step is that these interaction terms need to be added to the structural model. The moderation effects are estimated with their significance level, indicating that the moderation effect is significant. Table 3 shows that the interaction effect is significant; therefore, H5a and H5b are validated.

As suggested by Hair et al. (2019), the change in R ² values (∆R ²) of dependent variables green innovation, CSD, and GOC before and after including the interaction terms is checked, which will help to determine how strong the moderation effect is. The difference or ∆R ² can be noticed between Figure 2 with the moderator and Figure 3 without the moderator. A large increase in R ² would suggest that CSR significantly moderates the GMP effects on respective dependent variables green innovation, CSD, and employee GOC.

The f ² for moderation is also checked in Table 3, where the values 0.02, 0.15, and 0.35 reflect small, medium, and large effects, respectively (Chin et al., 2003). The comprehensive assessment ensures that the moderating role of CSR in enhancing the impact of GMPs on green innovation and CSD is robust and meaningful.

The results also provide some theoretical implications. First regarding the TBL theory, which requires a balance between social, environmental, and economic performance (Jum’a et al., 2022; Pan et al., 2021). The study results identify GMPs as positively and directly influential in green innovation, CSD, and GOC, hence reaffirming that environmental practices are part of the integral quest for sustainable development (Albloushi et al., 2023; Salim et al., 2019). Indeed, the adoption of GMPs enables mining companies to innovate and develop a committed pool of employees who contribute towards sustainability performance in all dimensions of the triple bottom line. It supports the literature that views environmental sustainability as a driver of long-term economic performance and social wellbeing (Fernando et al., 2019; Hendarjanti, 2022; Jum’a et al., 2022). These results also support the resource-based view that had identified green practices as offering not only unique resources but also acting as a catalyst for innovation and commitment to enhance overall corporate sustainability (Al-Hakimi et al., 2022; Al-Swidi et al., 2024; Suleman et al., 2024).

Second, a substantial impact of green innovation on both CSD and GOC, as well as GOC on CSD, strengthens the theoretical assumption that employee involvement and innovation are the main drivers of CSD; therefore, it is obliged to the stakeholder theory, which states that satisfying the interests of several stakeholders is important for long-term success (Fosu et al., 2024; Murtaza et al., 2024; Silva et al., 2023; Simmou et al., 2023).

Third, the study added scale to the coal mining industry which lacks literature (Onifade et al., 2024) and intermediating impacts of green innovation and GOC further underscore the pathways through which green practices are channelized to evoke sustainable outcomes (Le and Tham, 2024; Le et al., 2024; Omar et al., 2024; Rana and Arya, 2024; Ren et al., 2022), thus placing appropriate emphasis on the focus that TBL theory places on integrated performance (Dyllick and Muff, 2015; Pan et al., 2021).

Last, the positive moderating effect of CSR signifies that CSR activities help enhance the effectiveness of green practice; hence, this finding may be regarded as evidence for the assertion that social responsibility is a constituent element in the attainment of holistic sustainability (Awa et al., 2024). These findings enrich the literature by providing empirical data that CSR and green practices are not only compatible but also mutually enforcing to promote the overall sustainability of the mining industry.

The findings also have some practical implications for managers and policymakers in the mining industry. First of all, the positive outcomes of GMPs on green innovation, GOC, and CSD mean that mining companies have to pay great attention and spend valuable investment in GMPs (Onifade et al., 2024; Xiao et al., 2024). For that, managers must integrate green practices into the mainstream of business to comply with environmental regulations, stimulate innovation, and enhance employee engagement (Le and Tham, 2024; Rana and Arya, 2024; Suleman et al., 2024). The training programs, green behavior incentives, and investments in eco-friendly technologies may facilitate the establishment of a culture of sustainability. This may improve organizational performance and raise stronger commitments from employees towards the goals set regarding the environment by the company, thus contributing to CSD.

For policy-decision makers, evidence of the significant mediation effects of green innovation and GOC between GMPs and CSD, as well as the positive moderation effect of CSR, may suggest that supportive policies and frameworks are called for. Policies encouraging and supporting GI and CSR activities would further amplify the benefits of GMPs. Incentives in the form of tax breaks, research funding for green technology, and sterling examples recognition programs will attract and enhance green practices for mining companies. Second, several CSR reporting regulations can be enacted to ensure that CSR reporting will not just become a performative exercise but actually usher in sustainable development. This could be actualized through policymaking that would foster the environment for green practices and CSR in creating a more sustainable and responsible mining industry.