4.1. Findings from Symmetric Modeling
In this study, variance-based structural equation modeling (PLS-SEM) was employed for symmetric modeling. The assessment of construct validity, encompassing convergent and discriminant validity, as well as reliability, involved several indicators including item loadings, weights, Cronbach’s alpha (CA), composite reliability (CR), and average variance extracted (AVE), as presented in
Table 2. Notably, all retained item loadings exhibited significant
t-values. To ensure convergent validity, CA, CR, and AVE estimates for each construct were examined, surpassing the conventional thresholds of 0.70, 0.70, and 0.5, respectively, as recommended by Hair Jr. et al. [
74].
In
Table 3, discriminant validity was assessed employing Fornell and Larcker’s criterion in conjunction with the heterotrait–monotrait (HTMT) ratio. Following the Fornell–Larcker criterion, it was observed that the square root of the average variance extracted (AVE) for each construct surpassed its correlation coefficients with other constructs [
75]. Moreover, all HTMT ratio values were below 0.90 [
71]. Consequently, the findings of this study affirm both convergent and discriminant validity. For a comprehensive breakdown, refer to
Table 3.
In
Table 4, the direct effects of EMA and green financing on ESG performance are thoroughly examined. The results offer valuable insights into the nuanced relationships between these variables. Environmental cost tracking (ECT) demonstrates a statistically significant positive impact on ESG performance, as indicated by a coefficient (β = 0.165,
p < 0.01). This suggests that organizations that effectively track environmental costs tend to exhibit higher levels of ESG performance. Similarly, CEM and environmental reporting transparency (ERT) both show significant positive effects on ESG performance, with coefficients of β = 0.259 and β = 0.182, respectively (both
p < 0.01). These findings highlight the importance of actively managing carbon emissions and maintaining transparent environmental reporting practices to enhance overall ESG performance (see
Figure 2).
Conversely, eco-efficiency improvement (EEI) is found to have an insignificant but positive impact on ESG performance, with a coefficient of β = 0.024 and p > 0.10. While the effect is not statistically significant in this analysis, the trend suggests that efforts to improve eco-efficiency may still contribute positively to ESG performance outcomes. Furthermore, green financing (GF) emerges as a significant driver of ESG performance, with a positive coefficient of β = 0.280 (p < 0.01). This indicates that organizations leveraging green financing initiatives tend to achieve higher levels of ESG performance, underscoring the importance of sustainable financing practices in driving overall environmental, social, and governance objectives.
Table 5 presents a detailed examination of the direct effects of EMA and green financing on sustainable production. The analysis reveals intricate insights into the various components’ contributions to sustainable production outcomes. Notably, eco-efficiency improvement (EEI) emerges as a significant driver, with a coefficient (β = 0.301,
p < 0.01) signifying a substantial positive impact on sustainable production. This underscores the importance of enhancing eco-efficiency within organizational processes to achieve tangible improvements in sustainable production outcomes. Similarly, ECT demonstrates a robust positive influence, with a coefficient (β = 0.558,
p < 0.01) indicating a significant impact on sustainable production. This highlights the critical role of accurately monitoring environmental costs in driving sustainable production initiatives within organizations. However, CEM and ERT exhibit positive coefficients (β = 0.013 and β = 0.037, respectively), albeit statistically insignificant (
p > 0.10), suggesting potential contributions to sustainable production that lack statistical robustness in this analysis. Additionally, while GF shows a positive coefficient (β = 0.008) on sustainable production, it is statistically insignificant (
p > 0.10), indicating a potential but unconfirmed role in supporting sustainable production practices.
In
Table 6, the mediating effects of sustainable production on the relationship between EMA and ESG performance, as well as between GF and ESG performance, are explored. The findings illuminate the nuanced dynamics within these relationships. Sustainable production emerges as a significant mediator between eco-efficiency improvement and ESG performance, with a coefficient of β = 0.063 (
p < 0.01). This implies that enhancements in eco-efficiency lead to improved sustainable production practices, subsequently positively influencing ESG performance outcomes. Similarly, sustainable production acts as a mediator between environmental cost tracking and ESG performance, with a coefficient of β = 0.117 (
p < 0.01). This suggests that the effective monitoring of environmental costs contributes to enhanced sustainable production practices, consequently leading to heightened levels of ESG performance.
Conversely, sustainable production does not act as a mediator in the relationship between carbon emission management and ESG performance, as indicated by a coefficient of β = 0.003 and p > 0.10. This suggests that while the management of carbon emissions may directly impact ESG performance, its effect is not significantly influenced by sustainable production practices within this context. Similarly, sustainable production does not mediate the relationship between environmental reporting transparency and ESG performance, with a coefficient of β = 0.008 and p > 0.10. This implies that although transparent environmental reporting may positively contribute to ESG performance, its impact is not mediated through sustainable production practices. Furthermore, the analysis reveals that the relationship between green financing and ESG performance is not mediated by sustainable production, with a coefficient of β = 0.002 and p > 0.10. This suggests that the influence of green financing initiatives on ESG performance is not significantly affected by sustainable production practices within this study’s framework.
4.2. Findings from Asymmetric Modeling
Asymmetric modeling using fuzzy sets (fsQCA) integrates fuzzy sets and fuzzy logic, offering a nuanced approach to understanding complex relationships. This methodology holds significance for several reasons. Firstly, traditional methods like correlation and beta coefficients may inadequately capture the association between variables, particularly when the relationship is nonlinear. The application of fuzzy sets, as advocated by scholars such as Rahman et al. [
76], provides a framework that allows for multiple solutions, all leading to the same outcome. This flexibility is valuable in situations where traditional regression analysis may fail to identify independent measures that impact the outcome in only a subset of cases, as highlighted by Olya and Altinay [
77]. Therefore, asymmetric modeling with complexity theory, facilitated by fuzzy sets, offers a more comprehensive understanding of complex relationships that may not be fully captured by conventional statistical methods [
72].
Secondly, the reliability of symmetric approaches like multiple regression analysis (MRA) and structural equation modeling (SEM) is questioned when testing models featuring numerous independent variables with high correlation among them. This skepticism stems from concerns regarding the confounding impact of collinearity, as emphasized by Olya and Altinay [
77]. In MRA and SEM, the presence of collinearity complicates result interpretation, potentially leading to inaccurate conclusions. Furthermore, despite the utilization of large sample sizes, these methods may not effectively control for the effects of confounding variables such as gender, age, and education, as noted by Olya and Altinay [
77]. Hence, the limitations of symmetric approaches highlight the necessity for alternative methodologies like asymmetric modeling with fuzzy sets to offer more robust and accurate analyses, especially in complex scenarios with multiple interrelated variables.
Thirdly, in real-world scenarios, achieving a favorable outcome often relies on various antecedents collectively forming what is termed an algorithm within the asymmetric method, as discussed by Woodside [
78]. Unlike symmetric relationships, where high values of an independent variable (X) are both sufficient and necessary for predicting high values of a dependent variable (Y), in asymmetric approaches, high levels of X are sufficient but not necessarily required for predicting high levels of Y. This distinction underscores the nuanced nature of asymmetric modeling, where outcomes are determined by combinations of antecedents rather than strict linear relationships [
72]. Therefore, asymmetric modeling offers a more flexible and realistic framework for comprehending complex relationships and predicting outcomes in real-world contexts.
Fourthly, the asymmetric approach incorporates both positive and negative logic, recognizing that relying solely on one logic can lead to fallacious conclusions. Advocates, as highlighted by Woodside [
78], argue against the notion of “net effects” in asymmetric methods, contending that observed net effects may overlook cases where contradictory outcomes occur. This underscores the importance of examining all combinatory conditions where an independent variable (X) may have either a positive or negative influence on the outcome variable (Y), as emphasized by Olya and Altinay [
77]. In our study, we explore the various combinatory conditions through which CEM, ECT, EEI, ERT, GF, and SP interact to predict ESG performance, as depicted in
Figure 3. This approach allows for a comprehensive understanding of how different factors collectively influence the outcome variable, contributing to a more nuanced analysis of complex relationships.
The initial step in the fsQCA analysis involved calibration, given that the variables in our study were measured on a 7-point Likert scale, necessitating rescaling [
8,
72]. We utilized the latent variable scores as inputs for the fsQCA analysis. To ensure the suitability of our uncalibrated data, we evaluated skewness and kurtosis, confirming that they fell within acceptable ranges (skewness less than ±1 and kurtosis less than ±2), indicative of a normal distribution. Following recommendations from prior research [
8,
72,
73], the measures were calibrated into fuzzy sets with values ranging from 0 to 1. Specifically, 0.95 represented full set membership, 0.5 denoted the crossover point, and 0.05 indicated no set membership. The calibration process involved transforming variables into calibrated sets using the fsQCA program, where the maximum value represented full membership, the average value denoted the crossover point, and the minimum value indicated full non-membership. The results of this transformation, alongside other descriptive statistics of the causal conditions under investigation, are presented in
Table 7. This comprehensive approach ensures that variables are appropriately prepared for subsequent fsQCA analysis, facilitating a rigorous examination of causal relationships.
The second step of the analysis involved necessity analysis, also referred to as configurational element assessment. A condition is deemed necessary when its consistency score exceeds 0.9, as outlined by Ragin [
79]. Necessity analysis entails evaluating the proportion of fuzzy set scores within a condition (across all cases) that are equal to or lower than the corresponding scores in the outcome [
73]. The consistency scores presented in
Table 8 indicate that carbon emission management, environmental reporting transparency, and green financing are all necessary factors for achieving a high level of ESG performance. However, it is important to note that they are not individually sufficient. This assertion was corroborated by negation analysis, which revealed that the absence of these conditions resulted in lower scores for ESG performance (consistency score < 0.90). This comprehensive examination of configurational elements provides valuable insights into the complex interplay between various factors and their impact on ESG performance outcomes.
The third step in the fsQCA analysis involves applying the fsQCA truth table algorithm to generate a truth table comprising 2
k rows, where k represents the number of outcomes considered in the analysis [
79]. Each row in the truth table represents every possible combination among the causal conditions. For instance, in a truth table involving two causal conditions, there would be four logical combinations between them. In our study, the truth table was evaluated based on frequency and consistency values, following the guidelines established by Ragin [
79]. Frequency refers to the number of observations for each possible combination, with a suggested threshold of 3 for samples exceeding 150 [
72]. Meanwhile, consistency measures the degree to which cases correspond to the set-theoretic relationships expressed in a solution [
72], with a recommended threshold of 0.75 [
79]. This rigorous evaluation process ensures that the truth table accurately captures the interplay between causal conditions and outcomes, facilitating a comprehensive analysis of complex relationships. In fsQCA software (v4.1), three sets of solutions are typically generated: complex, parsimonious, and intermediate. These sets are distinguished based on the presence of “easy” and “difficult” counterfactuals. “Easy counterfactuals” occur when an unnecessary causal condition is added to a set of causal conditions that already predict the focal outcome. Conversely, “difficult counterfactuals” arise when a condition is removed from a set of causal conditions, resulting in an outcome under the assumption that this condition is unnecessary [
72,
73].
The complex solution encompasses all possible configurations of conditions or elements and includes neither easy nor difficult counterfactuals [
77]. However, this solution tends to be excessively complex and impractical, offering limited insights into causal configurations. In contrast, the parsimonious solution identifies vital conditions that can be either easy or difficult counterfactuals [
72]. It provides essential insights into the causal relationships by highlighting the key factors influencing the outcome. On the other hand, the intermediate solution focuses on vital conditions based on easy counterfactuals [
77]. It represents a compromise between complexity and simplicity, incorporating essential elements from both the parsimonious and complex solutions [
8]. By doing so, it offers a more balanced and manageable approach to understanding causal configurations, providing valuable insights while maintaining practicality.
These solutions are distinguished by necessary and sufficient conditions, categorized into core and peripheral conditions [
77]. “Core” conditions, or essential elements, are indispensable and exhibit a strong causal relationship with the outcome, commonly found in parsimonious and intermediate solutions. Conversely, “peripheral” conditions are less critical and may be interchangeable, typically present only in the intermediate solution [
72]. These conditions may manifest as present, absent, or deemed irrelevant (“do not care”), where their presence or absence does not significantly affect the outcome. Using ESG performance as the outcome variable, the fsQCA analysis comprehensively examines the relationships between causal conditions and their impact on the outcome, shedding light on factors influencing ESG performance. The principles of mediation and indirect effects elucidate the direct impact of independent and mediating variables on dependent variables. In our study, the measurement model identified a mediatory path from the symmetry analysis. Complementing the asymmetry analysis, tests assessed the direct effects of conditions on outcome variables. The results in
Table 9 detail how CEM, ECT, EEI, ERT, GF, and SP are all necessary conditions, yet individually insufficient for achieving higher ESG performance scores.
Moreover, our analysis reveals specific combinations of these conditions that are both necessary and sufficient for predicting higher scores of ESG performance. For instance, Solution S1 highlights that a combination of ECT, EEI, and SP is necessary and sufficient for predicting higher ESG performance. Similarly, Solution S2 indicates that CEM, ECT, EEI, and GF together constitute a necessary and sufficient combination for predicting higher ESG performance. Additionally, Solution S3 suggests that EEI, ERT, and SP in combination are necessary and sufficient for predicting higher ESG performance. Finally, Solution S4 demonstrates that ERT, GF, and SP collectively form a necessary and sufficient combination for predicting higher ESG performance. These findings underscore the complex interplay between various factors in influencing ESG performance outcomes and provide valuable insights for developing comprehensive sustainability strategies. All solutions demonstrated high consistency, indicating the reliability of the findings. Consistency measures the degree to which the observed cases correspond to the set-theoretic relationships expressed in the solution. Additionally, coverage signifies the extent to which a particular solution can explain variations in the outcome, like the concept of R-square in regression and structural equation modeling (SEM). The overall solution coverage suggests that the causal conditions (CEM, ECT, EEI, ERT, GF, and SP) collectively account for 89.60% of the membership in the solution associated with very high ESG performance. This high coverage indicates that the identified combinations of causal conditions offer a comprehensive explanation for the observed variations in ESG performance outcomes. It underscores the robustness of the solutions and the effectiveness of the selected variables in predicting and understanding ESG performance levels.
Table 10 presents the outcomes of negating the conditions, which further corroborate the findings from
Table 9 and the SEM analysis. These results provide additional support for the hypotheses put forward in this study. Notably, the analysis reveals that specific causal conditions, including CEM, ECT, EEI, and GF, along with ERT, GF, and sustainable production (SP), are both necessary and sufficient for achieving higher scores of ESG performance. This implies that as the combination of CEM, ECT, EEI, and GF increases, a firm’s ESG performance also increases. Similarly, an increase in the combination of ERT, GF, and SP corresponds to an increase in a firm’s ESG performance. Additionally, a predictive validity test was conducted to assess the reliability and validity of the research model, affirming the robustness and accuracy of the findings. These results provide valuable insights into the factors influencing ESG performance and enhance our understanding of the predictive power of the research model.