Research on the Path of Green Total Factor Productivity Improvement in Resource-Based Enterprises—Empirical Evidence from China

Abstract

The enhancement of green total factor productivity of resource-based enterprises has a crucial impact on building an environmentally friendly society, promoting a circular economy, and guaranteeing the security of national resource supplies. This study takes the green transformation of resource-based enterprises as the main line, and then constructs the research framework. A random forest algorithm is used to discover the effects of different characteristics of green transformation on green total factor productivity. Further, using panel data-based qualitative comparative analysis and necessary condition analysis, it clarifies what kind of green transformation group path promotes green total factor productivity from the group perspective. This study shows that (1) environmental regulations, green cognition, and green investment are important factors in condition configuration and there exists a complex nonlinear relationship among them, indicating the need for further research from a configurational perspective; (2) green technological innovation, green investment, green development strategies, green governance level, green cognition, environmental subsidies, and environmental regulation are not necessary conditions for a high level of green total factor productivity; (3) there are three grouping paths for green total factor productivity improvement in resource-based enterprises; and (4) there are time and case differences in the explanatory strength of the three grouping paths. Finally, based on the results of this study, and in conjunction with China’s green transformation practice, countermeasures are proposed from both the enterprise and government perspectives.

1. Introduction

With the deepening of China’s supply-side structural reform, the structure of domestic resource consumption is changing, with the proportion of traditional fossil energy consumption declining and the consumption of rare earths, lithium, cobalt, tungsten, and nickel, which are needed for strategic emerging industries, growing rapidly [1]. However, resource-based enterprises mostly focus on mining, smelting, and processing as their main business. These enterprises have long continued the extensive production and operation model of exhausting resources to catch fish. These “savage growth period” bad habits have led to a series of ecological problems such as high resource dependence and severe environmental pollution. This has caused enormous pressure on the ecological environment and also constrained its own sustainable development. Ecological and environmental issues do not exist in isolation; they are deeply rooted in the economic development model. Therefore, the green transformation of the development mode has become an inevitable choice to alleviate ecological and environmental pressure and promote sustainable economic development. From 2012 to 2021, China’s industrial added value soared from CNY 20.9 trillion to CNY 37.3 trillion, while energy consumption increased by nearly 70%, and the emissions of sulfur dioxide and nitrogen oxides accounted for approximately 80% and 60% of the total emissions, respectively [2]. To improve this situation, the Ministry of Industry and Information Technology of China has further clearly put forward the carbon peaking action in the industrial sector in the “14th Five-Year Plan for Green Development of Industry”, with the main goal of promoting the green and low-carbon transformation of the industrial structure and production methods. Therefore, in the context of the green transformation of economic development, how should we promote the green transformation of resource-based enterprises to facilitate the formation of resource-saving and environmentally friendly industrial structures and production modes? Through what combination of green transformation strategies will we help to improve the green total factor productivity of resource-based enterprises? What is the mechanism behind this? These are the questions that need to be studied urgently. Meanwhile, studying the improvement path of green total factor productivity in the context of the green transformation of resource-based enterprises can promote the transformation and upgrading of related industries. It can also contribute to the development of green industries, break away from reliance on traditional production and operation methods, and achieve high-quality development of the resource economy. In addition, this research can also help enterprises improve the efficiency of resource utilization while reducing negative externalities. This helps enterprises break through resource bottlenecks and environmental constraints, achieving coordinated development of economic benefits and environmental benefits. Thus, a unique green competitive advantage for resource-based enterprises can be constructed and the sustainable development level of resource-based enterprises can be enhanced. At the same time, a green corporate image can be established and market competitiveness can be enhanced.

Compared with previous studies, the potential marginal contributions of this research are as follows: (1) This research adopts PD-QCA and further combines the random forest algorithm and NCA method to explore the configuration path for enhancing green total factor productivity. (2) This study reveals the complex nonlinear characteristics of the improvement of green total factor productivity in resource-based enterprises from different perspectives and analytical dimensions. A more comprehensive revelation of the internal mechanism for the improvement of green total factor productivity in resource-based enterprises against the background of green transformation is conducive to scientifically grasping the laws of the improvement of green total factor productivity. (3) This study provides empirical evidence and insights from the enterprise level to help resource-based enterprises reduce negative externalities while improving resource utilization efficiency and production efficiency. At the same time, it provides a decision-making basis for the government to promote the green transformation of the economic development mode.

2. Literature Review

The main manifestation of the green transition is green total factor productivity gains [3]. The core objective of promoting high-quality economic development and facilitating a green transformation of the development approach lies in the continuous improvement of green total factor productivity.

Looking at the drivers of the green transition, first, there is the external environmental policy guidance: for example, on the macro-policy front, the Paris Climate Agreement, the “dual carbon” target [4,5], and the Belt and Road Initiative, etc. [6]. However, the role of policies for the environment is not always entirely positive. Some scholars, such as [7], examined the impact of firm-level environmental policy uncertainty on firms’ green transformation and mechanisms by measuring its impact, and found that environmental policy uncertainty has a significant negative impact on green innovation and total factor productivity of Chinese firms. Second, it is at the internal level where companies’ own perceptions are raised. As the constraints of the external environment become increasingly stringent, the internal awareness of greening within the enterprise is also increasing, and to a certain extent this affects the green transformation of the enterprise [8,9].

Looking at the empirical study of green total factor productivity, many scholars have used it to measure the green transition performance of different regions [10,11]. A number of scholars, too, have used green total factor productivity to study the extent of industrial green transformation [12,13,14]. Some scholars use green total factor productivity to measure the degree of green transformation of enterprises [15,16]. In addition, some scholars have further explored how the green transition process can enhance green total factor productivity. Examples include green technology innovation [17], green finance [18,19], environmental regulations [20], and others.

The existing related studies provide valuable research ideas for the path of green total factor productivity enhancement under the green transformation of resource-based enterprises, but there are still some places worth exploring. First, there are fewer studies on green total factor productivity enhancement from the perspective of group linkage. Existing studies have discussed a number of drivers of green transformation, such as policies, corporate strategies, capital, innovative technologies, and so on. In addition, a number of scholars have tried to study the relationship between these factors and green total factor productivity enhancement. However, most of the past research focuses on single-factor perspectives, such as in [21,22]. Green transition is a complex system involving multiple factors, and it is not enough to analyze it only from one perspective or through one factor. Second, there is insufficient empirical evidence from enterprises at the micro level. It can be seen that there are quite a few existing studies on green transitions or green total factor productivity, but these studies mainly focus on the regional and industrial economy [23,24], and few studies focus on resource-based enterprises. Therefore, this study constructs a research framework for green transformation and considers different factors in an integrated way, adopting the random forest algorithm and PD-QCA and NCA methods and considering the intertwined roles of different factors from the single-factor and group perspectives. This allows it to better study the path mechanism of green total factor productivity enhancement in the process of green transformation of resource-based enterprises and grasp its rules.

3. Research Design

3.1. Research Framework

Drawing on the definition of the concept of green transformation by the Institute of Industrial Economics of the Chinese Academy of Social Sciences [25], it is considered that green transformation is oriented towards the intensive use of resources and environmental friendliness, with green innovation at its core, adhering to the road of new industrialization, realizing the greening of the entire process of industrial production, sustainable development, and obtaining a win–win situation in terms of both economic and environmental benefits. Green transformation is not a simple static process, but a complex dynamic process involving innovative technologies, production factors, product structure, and other dimensions, the essence of which is to adopt new technologies, new ideas, and new systems to promote the overall improvement of resource allocation, innovation level, and organizational efficiency [26]. At present, many views in the academic community suggest that the key to promoting green transformation of enterprises lies in improving their ability and willingness for green transformation [15]. Therefore, combined with previous research [27], with green development as the goal, green innovation as the driving force, and green regulation as the guideline, this research framework is constructed from the three dimensions of green development strategy, green innovation inputs, and external guidance for green transformation. Among these, green innovation input is divided into green technology innovation (Green_Innov) and green investment (Green_Invest) from the perspectives of capital and technology; green development strategy is divided into green governance level (Green_gov) and green cognition (Green_Congni) from the perspectives of management and cognition; and green transformation guidance is divided into environmental subsidies (Enviral_Subsidy) from the perspectives of incentives and regulation, divided into environmental subsidies (Envir_Subsidy) and environmental regulations (Envir_rule). The research framework of green total factor productivity improvement in the context of green transformation of resource-based enterprises is shown in Figure 1.

3.2. Research Methodology

(1)

Panel Data QCA

Panel data-based dynamic qualitative comparative analysis (PD-QCA), also called dynamic QCA, is a more cutting-edge research method within the field of QCA methodology. Distinguished from traditional non-dynamic QCA and econometric methods based on statistical theory as the mathematical foundation, PD-QCA takes set theory as the mathematical foundation, introduces the idea of ecological evolution and coupling, and runs in the R language environment to realize dynamic analysis. Green total factor productivity enhancement of resource-based enterprises is a multifactor concurrent coupling problem. Introducing the perspective of grouping through the PD-QCA method can allow us to explore the grouping effect formed by the intertwined action of multiple factors, and find the minimum sufficiency condition by running in the R language environment. Using panel data to analyze the dynamic evolution process of group states can break the shackles of QCA methodology cross-section data, creatively demonstrate the different changes in multiple group states in continuous time from the three dimensions of between, within, and pooled, empirically study the panel data, and, through the “Between Consistency Adjustment Distance and Within Consistency Adjustment Distance”, reflect individual and temporal differences in consistency.

(2)

Necessary condition analysis

Necessary condition analysis (NCA) is a methodology for assessing causality and determining the significance of the impact of certain variables in a given outcome. QCA, although it involves necessary condition analysis (NCA), is unable to answer questions such as the degree of necessity of conditions, i.e., “to what extent are the conditions of the green transition binding green total factor productivity” or “constraining green total factor productivity”. NCA not only identifies the necessary conditions, but also further analyzes the intensity of the conditions’ impact on the outcomes and the thresholds required to achieve the goals, which can help researchers to more accurately grasp the key drivers and enhance the causal inferences. NCA employs Ceiling Regression (CR) and Ceiling Envelopment (CE) analyses to deal with sample variables, combining the necessity effect value and its significance (p-value) to determine the necessary conditions, and measuring the degree of necessity of the conditions in terms of the bottleneck level. Therefore, in studies involving complex social phenomena and policy evaluation, NCA can serve as an effective complement to QCA necessity conditions, providing new insights that cannot be found by traditional QCA methods.

(3)

Random Forest Algorithm

Random forest (RF) is a proposed machine learning algorithm based on decision trees and integrated learning that has the ability to influence factor importance assessment and identify factors that significantly affect output results. The random forest model realizes the assessment of the importance of influencing factors by constructing multiple decision trees and training them based on different subsets of features. The influence mechanism of the green transformation process of resource-based enterprises on the improvement of green total factor productivity is complex. With the help of the random forest model, research and analysis can scientifically and accurately explore the mechanism of different factors and capture the nonlinear relationship between independent variables and dependent variables, which is crucial in the study of the green transformation process involving multiple complex internal and external factors. In addition, in reality, data are often subject to various disturbances, while random forests are robust to outliers and noise, meaning that errors and fluctuations in the actual data will not significantly affect the final results. Random forests can effectively handle high-dimensional data, and discover a large number of variables and complex nonlinear interactions among them. Thus, this model effectively supplements the QCA method.

3.3. Variables and Measures

Green cognition (Green_Congni) refers to knowledge systems and cognition of resources and the environment formed by enterprises on the basis of their understanding of resource and environmental issues [28]. Such cognition reflects managers’ perception and understanding of resource and environmental issues based on their own knowledge structure and values when assuming responsibility for resource conservation and environmental protection. The green cognition of an enterprise determines, to a certain extent, whether the enterprise is able to make a sustainable green transformation, and it also promotes the innovation of green management methods within the enterprise, thus promoting the improvement of green total factor productivity. The wording of “Management Discussion and Analysis” in the annual report can largely reflect the future outlook of the enterprise and the strategic characteristics it implements, as well as the business philosophy it upholds and the development path guided by this philosophy. We referred to [29,30], who conducted a study using machine learning text analysis to analyze and measure the “management discussion and Analysis” in annual reports.

Environmental subsidies (Envir_Subsidy) are compensatory or incentivized financial support provided by the government to promote the active participation of enterprises in environmental governance, controlling pollutant emissions, and reducing environmental pollution, or encouraging them to improve their products and processing techniques to enhance resource efficiency [31]. Referring to the previous study [32], the government environmental subsidies were measured by dividing the number of government environmental subsidies by the proportion of the total assets of the enterprise and then multiplying it by 100 based on the details of government subsidy items in the notes of the annual report disclosed by the enterprise.

Green technological innovation (Green_Innov) refers to technological innovation that effectively reduces environmental loads, reduces resource consumption, and improves resource utilization efficiency [33]. In order to promote the international exchange and transformation of green technology patents and provide a standardized basis for domestic green patent identification, four departments of China’s National Development and Reform Commission (NDRC), the Ministry of Science and Technology (MOST), the Ministry of Industry and Information Technology (MIIT), and the Ministry of Ecology and Environment (MOE) issued the Catalogue for the Promotion of Green Technology (CPGT) in 2020, and the State Intellectual Property Office (SIPO) issued the Classification System of Patents on Green Technology (CSGT) in 2023. Green technology innovation not only optimizes production processes, reduces costs, improves product quality, and enhances market competitiveness, but it also helps enterprises achieve the unity of economic, social, and environmental benefits [34]. Drawing on [35], this study measures firms’ green innovation capabilities through their number of green patent applications. The reason for using the number of patent applications rather than the number of patent authorizations is that patent authorizations need to go through testing and require paying annual fees, and there are many non-technical factors such as administrative approval that interfere. However, patent technology is very likely to have an impact on enterprise performance at the application stage. Therefore, patent application data can reflect the level of innovation more reliably and promptly than authorization data.

Green governance (Green_gov) refers to the management capability of enterprises in environmental protection and sustainable development, including monitoring and assessing the impact of production activities on environmental quality, responding quickly to environmental emergencies, and taking measures to mitigate damage. The level of green governance reflects an enterprise’s ability to respond to environmental emergencies, and its level of disposal and its strength or weakness have a bearing on the effectiveness of environmental protection and the realization of sustainable development [36]. The impact of an enterprise’s level of green governance on green total factor productivity is mainly focused on environmental risk management and reducing compliance costs. A higher level of green governance can help enterprises to identify and assess environmental risks, reduce potential environmental damages through preventive measures, avoid increased costs due to environmental problems, and reduce fines due to non-compliance, thus enhancing green total factor productivity. Referring to [37], the Janis–Fadner coefficient (J-F coefficient) was applied to measure the level of green governance based on the positive and negative scores obtained by the company in green governance. The specific formula is as follows:

$$Green\_gov=\left\{\begin{array}{c}(p^2-p\times |q|)/r^2,\mathrm{if} \mathrm{p}>\left|q\right|\\ (p\times |q|-q^2)/r^2,\mathrm{if} \mathrm{p}<\left|q\right|\\ 0,\mathrm{if} \mathrm{p}=\left|q\right|\end{array}\right.$$

(1)

where $p$ is the positive score for the level of green governance, which is scored based on the number of items that the sample company meets in the positive score criteria, with each item counting for one point; $q$ is the negative score for the level of green governance, which is based on the number of items that the sample company meets in the negative score criterion, with each item being scored as −1; and $r$ is the sum of the absolute values of $p$ and $q$; that is, $r$ = $p$ + $\left|q\right|$.

Environmental regulation (Envir_rule) mainly refers to command-and-control environmental regulation, which is a type of government intervention based on administrative rather than market instruments [38] and considers that the number of policies does not directly reflect the intensity of implementation and that there are differences in per capita income levels between regions, which have a nonlinear relationship with environmental control. In addition, the unit pollution emissions of different products are also affected by their heterogeneity. This study focuses more on environmental control, so we chose to measure the intensity of environmental control by the proportion of funds invested in controlling three types of waste pollution in the region where the listed company is located in that year to the industrial output value of that year [39].

Green investment (Green_Invest) refers to the investment in environmental governance, environmental protection inputs, green technological transformation, innovation, etc. by enterprises in order to reduce environmental costs and improve environmental performance [40]. Referring to the study by [41], in the annual reports of listed companies in resource-based enterprises, the expenditure items directly related to environmental protection, including desulfurization, denitrification, wastewater treatment, exhaust gas treatment, dust removal, and energy conservation, were extracted as the detailed items of the construction-in-progress account, and the data were summarized to obtain the increase in the enterprises’ environmental protection investment in the current year. In order to control for the effect of differences in company size, the total assets of enterprises at the end of the year are standardized to facilitate the analysis of their environmental protection investment. At the same time, in order to enhance the readability of the subsequent regression coefficients, the standardized environmental protection investment data are multiplied by 100 for processing.

Green total factor productivity (GTFP) makes up for the shortcomings of TFP in environmental pollution and ensures that the evaluation results are more in line with the concepts of sustainable development and green development. In analyzing the relationship between inputs, desired outputs, and undesired outputs, the concept of green total factor productivity has clear advantages over traditional total factor productivity [42]. This paper draws on [17] to solve the directional distance function for the non-radial slack-based measure (SBM) in conjunction with the non-expected output SBM model. The Undesired Output SBM model is a Data Envelopment Analysis (DEA) modeling approach for measuring relative efficiency and is suitable for situations where undesired outputs (e.g., environmental pollution) are considered. The basic form of this model is set up with n decision units (DMUs). The inputs and outputs of each DMU are defined as, respectively, input vectors: $x=(x_1,x_2,\dots ,x_m)$; the expected output vector: $y=(y_1,y_2,\dots ,y_s)$; and the non-expected output vector: $z=(z_1,z_2,\dots ,z_t)$. In the mathematical expression of the SBM model for the kth decision-making unit (DMU_k), its efficiency can be solved by the following linear programming:

$$\begin{array}{c}\mathrm{Maximize}\theta \\ {\displaystyle \sum _{j=1}^n}{\lambda }_j{x}_{ij}\le x_{ik}(1+s_i),\hspace{1em}\forall i\in 1,2,\dots ,m\\ {\displaystyle \sum _{j=1}^n}{\lambda }_j{y}_{rj}\ge y_{rk}(1-s_r),\hspace{1em}\forall r\in 1,2,\dots ,s\\ \mathrm{s}.\mathrm{t}.{\displaystyle \sum _{j=1}^n}{\lambda }_j{z}_{tj}\le z_{tk}(1+s_t),\hspace{1em}\forall t\in 1,2,\dots ,t\\ {\displaystyle \sum _{j=1}^n}{\lambda }_j=1\\ {\lambda }_j\ge 0,\hspace{1em}\forall j\\ s_i,s_r,s_t\ge 0\end{array}$$

(2)

When $\theta$ < 1, or $s_i,s_r,s_t$ are not all 0, decision-making units are weakly effective; there is a loss of efficiency, which suggests that there is room for improvement in the input–output ratio at this point. Input inefficiencies and output inefficiencies are expressed as

$$\mathrm{Input} \mathrm{inefficiencies}: I{E}_x={\displaystyle \frac{1}{m}}\sum _{i=1}^m{\displaystyle \frac{s_i^i}{x_{i0}}}, (i=1,2,\dots ,m)$$

(3)

$$\mathrm{Expected} \mathrm{output} \mathrm{inefficiencies}: I{E}_g={\displaystyle \frac{1}{s_1}}(\sum _{r=1}^{s_1}{\displaystyle \frac{s_r^r}{y_{r0}^g}}),(r=1,2,\dots ,s_1)$$

(4)

$$\mathrm{Undesired} \mathrm{output} \mathrm{inefficiencies}: I{E}_b={\displaystyle \frac{1}{s_2}}(\sum _{r=1}^{s_2}{\displaystyle \frac{s_t^r}{y_{r0}^b}}),(t=1,2,\dots ,s_2)$$

(5)

where $\theta$ represents the efficiency level of the DMU, with a value between 0 and 1; ${\lambda }_j$ denotes the weight of each DMU on the target DMUs; and $s_i,s_r,s_t$ AAA denotes slack variables for inputs, desired outputs, and undesired outputs, respectively. The input and output indicators of green total factor productivity are shown in

Table 1. Input–output indicators.

Input and Output	Essential Factor	Norm
Input element	Labor force	Number of employees in the enterprise
	Capital	Net fixed assets
	Energy	Industrial electricity consumption
Expected outputs	Economic output	Revenues
Non-expected outputs	Environmental pollution	Industrial sulfur dioxide
		Industrial wastewater
		Industrial fumes

, where the undesired outputs are converted using the industrial sulfur dioxide, industrial wastewater, and industrial soot emissions of the enterprise in the city where it is located [43].

3.4. Sample Data Sources

The main characteristic of resource-based enterprises is their dependence on specific natural resources for exploration, extraction, processing, and utilization. Referring to the research by [44], the research object is targeted at resource mining, washing, and processing manufacturing enterprises according to the National Economic Industry Classification and the 2019 CSRC industry classification standard, as shown in

Table 2. Industry classification codes of the China Securities Regulatory Commission.

Code	Name	Code	Name
B06	Coal mining and washing industry	C25	Petroleum processing and coking industry
B07	Oil and gas extraction industry	C26	Chemical raw material and chemical product manufacturing industry
B08	Mining and selection industry of ferrous metal ores	C30	Nonmetallic mineral products industry
B09	Non-ferrous metal mining and selection industry	C31	Ferrous metal smelting and rolling processing industry
B10	Nonmetallic mineral mining and selection industry	C32	Non-ferrous metal smelting and rolling processing industry
D44	Power and heat production and supply industry	C33	Metal products industry

. The data on industrial waste emissions used for the calculation were obtained from the China Environmental Statistics Yearbook. Data on industrial output value and labor force in the region where the enterprise is located are from the China Industrial Statistical Yearbook and China Urban Statistical Yearbook. Patent application data are from the China Research Data Service Platform (CNRDS). Corporate and institutional investor data are from CSMAR and social responsibility score data are from Hexun.com. Due to the changes in the statistical scope and items of some data in the statistical yearbook, and considering the availability and completeness of the data, the data from 2012 to 2021 are adopted. After eliminating the missing data and outliers, a total of 1746 observations from 193 companies were obtained.

4. Results and Analysis

4.1. Feature Importance Analysis Based on Random Forest Algorithm

First, random feature selection is performed. Since the random forest model is not sensitive to the data magnitude, the data are directly randomly sampled, then divided by the ratio of 70% training set and 30% test set, and then the data in the 70% training set are randomly sampled by bootstrap sampling, and several sample features are selected for constructing each decision tree, and multiple decision trees are generated after repeating this process several times. Next, the decision tree base learner is constructed. A number of generated feature subsets are combined to form a binary tree structure of multiple decision trees, each decision tree is trained one by one, and then node splitting and pruning operations are performed on each decision tree, in which each node represents a feature, and each leaf node represents a category or value. Each tree is constructed by recursively dividing the training data. The decision tree feature vectors are calculated as follows:

$$\underset{j,m}{\min }\left[\sum _{q_i\in R_1(j,m)}{(q_i-\widehat{q_{R_1}})}^2+\sum _{q_i\in R_2(j,m)}{(q_i-\widehat{q_{R_2}})}^2\right]$$

(6)

In the equation, $\widehat{q_{R_1}}$ and $\widehat{q_{R_2}}$ are the means of the two groups of samples after division. Finally, the above process is repeated several times to build multiple decision trees that converge into a random forest. The random forest model is constructed as shown in Figure 2.

Influence factor importance assessment is an important step in random forest modeling. Two commonly used methods are Increasing Mean Squared Error (IncMSE) and Increasing Node Purity (IncNodePurity). The main difference between these two computational methods is that IncMSE evaluates feature importance by perturbing the features and observing the change in model performance, while IncNodePurity calculates feature importance based on the change in node purity when the decision tree splits. In order to visualize the results, the dplyr package in RStudio 4.3.1 was utilized to convert the feature importance into percentages, and the results are shown in Figure 3.

By comparing the results of both IncMSE and IncNodePurity calibers, it is found that Envir_rule, Green_Congni, and Green_Invest, regardless of the caliber of the algorithm, are at the top. This suggests that changes in the three factors, which perturb the whole model, are of high importance. This importance cannot be understood simply as correlation in linear regression or as causality. Based on reality, there are two main reasons for resource-based enterprises to actively engage in green transformation to improve green total factor productivity: one is based on the active pursuit of interests, and the other is based on the forced adaptation of regulatory constraints. Government environmental control, as an important policy instrument, can to a certain extent constrain the behavior of resource-based enterprises, thus reducing non-desired outputs such as pollutants and improving the efficiency of resource allocation, which is the most important factor affecting the green total factor productivity of resource-based enterprises. Resource-based enterprises are prone to negative externalities in the production and operation of natural resources and the environment. In the absence of environmental rules, companies follow a profit-oriented approach and will not proactively transfer the cost of green governance to themselves. On the contrary, enterprises are influenced by the regulatory role of green awareness and will consider the long-term interests of the company. If influenced by the concept of sustainable development, they reduce pollution through green investment, increase green output, or develop environmental strategies to address stakeholders’ environmental concerns.

4.2. Analysis of Necessary Conditions

The necessity analysis of individual conditions is conducted to determine whether there is a logical relationship of necessity between the six condition variables and the green total factor productivity. At the same time, a single conditional necessity analysis can also ensure that the subsequent conditional grouping sufficiency analysis will not omit important conditional variables.

(1)

Necessity of QCA conditions

Calibration is also required prior to conditional necessity analysis, and this study used a direct calibration method to calibrate the pooled affiliation at the 0.95, 0.5, and 0.05 percentiles, representing full affiliation, intersection, and incomplete affiliation, respectively, as shown in

Table 3. Variable calibration.

Variable		Total Affiliation	Intersection Point	Completely Unaffiliated
Outcome variable	GTFP	1.122	1.012	0.879
Conditional variable	Green_Congni	19.000	5.000	0.000
	Envir_Subsidy	0.251	0.006	0.000
	Green_Innov	2.565	0.693	0.000
	Green_gov	1.000	0.222	0.000
	Envir_rule	7.275	1.600	0.300
	Green_Invest	380.733	78.962	5.692

.

In the conditional necessity analysis of QCA, a conditional variable with pooled consistency > 0.9 and pooled coverage > 0.5 can be basically regarded as a necessity condition for the outcome variable. If the within- or between-consistency-adjusted distance is ≥0.2, it is still necessary to judge the necessity of the condition by further observing its distribution.

As can be seen from

Table 4. QCA necessary condition analysis.

Conditional Variable	GTFP				~GTFP
	Pooled Consistency	Pooled Coverage	Between-Consistency-Adjusted Distance	Within-Consistency-Adjusted Distance	Pooled Consistency	Pooled Coverage	Between-Consistency-Adjusted Distance	Within-Consistency-Adjusted Distance
Green_Congni	0.547	0.637	0.407	0.947	0.533	0.653	0.421	0.894
~Green_Congni	0.702	0.588	0.258	0.454	0.704	0.621	0.163	0.480
Envir_Subsidy	0.581	0.128	1.613	0.841	0.528	0.728	0.352	0.881
~Envir_Subsidy	0.792	0.615	0.109	0.307	0.696	0.569	0.240	0.454
Green_Innov	0.597	0.667	0.247	0.627	0.487	0.572	0.440	0.787
~Green_Innov	0.616	0.533	0.269	0.467	0.045	0.652	2.601	0.414
Green_gov	0.761	0.578	0.109	0.480	0.678	0.542	0.218	0.494
~Green_gov	0.396	0.539	0.465	0.814	0.678	0.542	0.218	0.494
Envir_rule	0.356	0.805	0.538	1.027	0.282	0.671	0.781	0.961
~Envir_rule	0.854	0.531	0.098	0.200	0.918	0.600	0.040	0.107
Green_Invest	0.672	0.545	0.127	0.454	0.564	0.481	0.280	0.694
~Green_Invest	0.360	0.439	0.440	0.774	0.466	0.599	0.283	0.774

Note: “~” denotes “not” in a logical rule, which is a negative proposition in a logical proposition.

, in the single conditional necessity test, the consistency level and coverage of ~Envir_rule for ~GTFP reached 0.918 and 0.600, respectively, and the between- and within-group adjustment distances were both less than 0.2. This suggests that ~GTFP can be deduced from ~Envir_rule, a set of necessary conditions. By the logical rules of counterfactual analysis, its inverse negation also holds, which means that Envir_rule may be a sufficient condition for GTFP. In addition, there are quite a number of variables whose adjusted distances between groups are within [0.2, 1], which indicates that the consistency of the cross-sectional data fluctuates from year to year, and there may be a time effect, so they need to be further analyzed.

Further analysis of the cross-section of each condition found that the consistency between the six conditional variables in their effect on the outcome variables showed a downward trend with time, while in the combination of the negative proposition of the conditional variables and the negative proposition of the outcome variables, the consistency between the groups showed an upward trend with time, as shown in Figure 4.

In addition, this pattern of change also reflects the decreasing necessity and increasing sufficiency of the conditional variable for the outcome variable. Although the QCA method is mathematically based on “set theory”, focusing on causal asymmetry and multifactor concurrent aggregation effects, and does not have the endogeneity problem of traditional quantitative methods [45], the above characteristics of the situation, from the point of view of set theory, also indicate that there may be a certain degree of unidirectional causality between the variables and the outcome. This indicates that green cognition, environmental subsidies, green technological innovation, environmental control, and green investment all have different degrees of influence on green total factor productivity enhancement, but the specific mechanism of action as well as the path of influence need to be verified by conditional group state sufficiency analysis.

(2)

Necessity of NCA conditions

NCA necessity analysis can complement QCA necessity analysis, which can only qualitatively determine the necessity of a condition and cannot further elucidate the extent to which this necessity affects the outcome. Instead, NCA can be further explained by effect sizes as well as bottleneck levels.

Using Rstudio 3.3.3 software, the NCA program package was invoked and 1000 Monte Carlo simulation permutation tests were performed using Ceiling Regression (CR) and Ceiling Envelopment (CE), and the results were obtained, as shown in

Table 5. NCA necessary condition analysis.

Condition Variable	Method	Precision	Space Ceiling	Effect Size	p-Value
Green_Congni	CR	100.00%	0.000	0.000	1.000
	CE	100.00%	0.000	0.000	1.000
Envir_Subsidy	CR	100.00%	0.000	0.000	0.421
	CE	100.00%	0.000	0.000	0.409
Green_Innov	CR	100.00%	0.000	0.000	1.000
	CE	100.00%	0.000	0.000	1.000
Green_gov	CR	99.10%	0.070	0.120	0.320
	CE	100.00%	0.084	0.140	0.389
Envir_rule	CR	100.00%	0.007	0.000	0.837
	CE	99.40%	0.005	0.000	0.905
Green_Invest	CR	99.00%	0.003	0.000	0.195
	CE	100.00%	0.000	0.000	0.372

.

The effect size is in the range of 0 to 1, and the closer to 1 means the stronger the necessity. According to [46], when the effect size is >0.1 and the p-value test result is significant (p-value < 0.05), the condition can be judged as a necessity condition. From the results in the table, no conditions were found necessary to increase green total factor productivity.

Bottleneck level results for different variables are reported in

Table 6. NCA bottleneck level.

GTFP	Green_Congni	Envir_Subsidy	Green_Innov	Green_gov	Envir_rule	Green_Invest
0%	NN	NN	NN	NN	NN	NN
10%	NN	NN	NN	NN	NN	NN
20%	NN	NN	NN	NN	NN	NN
30%	NN	NN	NN	NN	NN	NN
40%	NN	NN	NN	NN	NN	NN
50%	NN	NN	NN	NN	NN	NN
60%	NN	NN	NN	NN	NN	NN
70%	NN	NN	NN	NN	NN	NN
80%	NN	NN	NN	25.10%	NN	NN
90%	NN	NN	NN	53.70%	NN	0.50%
100%	NN	2.00%	NN	82.30%	3.10%	4.50%

Note: Unit is %, NN means not necessary.

. Since the condition variables are mostly continuous rather than discrete and the number of variables is more than five, it is customary to use the CR method in the reporting of the results for the bottleneck level of the NCA. Bottleneck level refers to the fact that, for the outcome variable to satisfy a certain level in the maximum observed range, the antecedent condition level needs to be achieved within the maximum observation range. Combining the results of the bottleneck level, it is found that the green total factor productivity from the 0% to 70% levels is not limited by the condition variables, while the levels of Envir_Subsidy, Green_gov, Envir_rule, and Green_Invest increase to different degrees when moving from the 80% to 100% level, with the green governance capacity being the most obvious.

4.3. Configuration Adequacy Analysis

Referring to the study of [47], the PRI must not be lower than 0.5, otherwise it indicates that there is non-significant consistency in the grouping, and the PRI threshold is set to 0.5. Reference [48] proposed that consistency higher than 0.8 has better explanatory strength, and [49] considered that as long as the consistency is greater than 0.75 it is sufficient. Based on the previous studies and the data, a higher consistency threshold of 0.8 was chosen, and the frequency threshold was set at 2.

Running the Rstudio 5.3.2 software QCA program package, the summary results were obtained, as shown in

Table 7. Configuration path.

	Group 1	Group 2	Group 3
Green_Congni		◎	●
Envir_Subsidy	◎		◎
Green_Innov	◐	◐	●
Green_gov		●	●
Envir_rule		●
Green_Invest	◐	◐	◐
Consistency	0.837	0.840	0.794
PRI	0.653	0.652	0.550
Original coverage	0.318	0.237	0.248
Unique coverage	0.055	0.022	0.065
Between-consistency-adjusted distance	0.258	0.193	0.211
Within-consistency-adjusted distance	0.120	0.160	0.227
Pooled consistency	0.793
Pooled PRI	0.611
Pooled coverage	0.405

Note: ● denotes the core existence condition, ◐ denotes the existence condition, ◎ denotes the core missing condition, and a blank space represents that the condition may or may not exist.

. The pooled consistency is 0.793, which is greater than the set consistency threshold of 0.75, and the overall PRI is also greater than the set PRI threshold (0.5), indicating that the overall explanation is stronger. In addition, the between-group adjusted distance and within-group adjusted distance were also around 0.2, with relatively stable and steady data fluctuations, but some were still slightly higher than 0.2, suggesting the presence of certain time effects or case differences.

The consistencies of the three configuration paths generated are 0.837, 0.840, and 0.794, respectively, which are greater than the initial consistency threshold of 0.75, indicating that these conditional configurations can fully lead to a high level of green total factor productivity. The experimentally discovered groupings are classified and named according to the grouping theorizing process [50].

① Enterprise independent innovation investment type: This type corresponds to group 1 and group 3. Group 1 is characterized by Green_Innov and Green_Invest as presence conditions and Envir_Subsidy as the core missing condition. It indicates that resource-based enterprises can rely on their own technological innovation to improve the efficiency of energy and resource utilization without relying on government environmental subsidies, thus reducing production costs and becoming a key way for heavily polluting enterprises to enhance their green total factor productivity. However, enterprises’ green technology innovation and research and development cycles are long, with large inputs and high uncertainty of results, so green investment is also needed as strategic support. From a short-term perspective, green investment may increase enterprise costs in the short term, and investment in pollution control and energy efficiency may have a crowding-out effect on the transformation and application of green technology, with a certain negative impact on technical efficiency. However, from a long-term perspective, green investment can promote the adoption of cleaner production technologies by enterprises and enhance productivity and profitability, thereby offsetting the increase in costs brought about by environmental pollution control and promoting green total factor productivity. Group 3 is a conditional grouping with Green_Congni, Green_Innov, and Green_gov as the core presence conditions, Envir_Subsidy as the core absence condition, and Green_Invest as the presence condition. Group 3 has similarities with group 1, but on the basis of group 1, the Green_Congni and Green_gov capabilities are added as the core existence conditions, and the status of Green_Innov innovation is also elevated, which can be regarded as an enhancement of group 1. Comparing group 3 and group 1, it is easy to see that group 3 reflects that enterprises are more proactive in production technology innovation and pollution control in order to improve green total factor productivity. However, it is worth noting that the consistency of group 3 is 0.794, which is lower than the consistency of group 1 of 0.837, and the explanatory strength of group 3 is lower than that of group 1, indicating that the enterprises’ environmental protection awareness enhancement and their investment in the innovation of green production technology, as well as the management of negative externalities in the process of industrial production, may not be covered by the effective output due to the rising cost, which affects the green total factor productivity.

② Government environmental control—passive governance type: This type corresponds to group 2, which is characterized by Green_Congni as the core missing condition, Green_Innov and Green_Invest as the presence conditions, and Green_gov and Envir_rule as the core presence conditions. Group 2 embodies the fact that when enterprises lack the initiative for green cognition and green production, the government can force the enterprises to improve their green governance ability through a series of environmental controls; according to “Porter’s hypothesis”, when firms are forced to improve their green governance capacity under policy pressure, it will stimulate their innovation motivation to develop new technologies or improve the existing production process to meet more stringent environmental standards. In this process, green innovation may enable firms to offset the economic costs incurred by environmental governance, which in turn has a positive effect on green total factor productivity. In addition, group 2 is more consistent than the remaining two groups. Due to their inherent industry characteristics, resource-based enterprises may generate negative externalities on the environment during resource production and processing, and there are certain environmental governance costs involved. If there is a lack of environmental control constraints, few companies will proactively shift the cost of environmental governance onto themselves. Therefore, relying more on government environmental control to guide green behavior is consistent with the significant impact analysis results under the random forest algorithm. In addition, the consistency of group 2 is higher than those of group 1 and group 3, indicating that among the three grouping paths, group 2 is the strongest explanation of high-level green total factor productivity, which also reflects the relative lack of green behavior initiative in resource-based enterprises.

(1)

Analysis of Between-group results

Figure 5 shows the change trend of the three groups. Taking 2013 as the time point, the consistency level of the three groups from 2012 to 2013 declined to varying degrees, which indicates that the explanation of the three groups for the green total factor productivity enhancement is not strong enough in this period. The consistency level of the groups after 2013 has risen rapidly, and then in 2015, it started to overlap and stabilized and finally approached 1, indicating that all three groups can better explain the realization path of green total factor productivity improvement after 2015. Overall, the explanatory power of the three conditional configurations is gradually increasing and tending towards stability, but there are also stage-specific characteristics. From 2012 to 2013, perhaps due to resource-based enterprises still remaining in the extensive development model of the past, neglecting environmental protection in pursuit of economic benefits, the problems of excessive resource exploitation and environmental pollution became increasingly prominent. During this period, the explanatory power of various configuration paths for the improvement of green total factor productivity decreased. Since 2015, the explanatory power of the three configurations of green transformation for green total factor productivity has steadily increased and remained stable, possibly due to the revision and promulgation of China’s Environmental Protection Law in 2015. As a key industry for environmental pollution, resource-based enterprises face stricter external environmental regulations and need to bear greater environmental responsibilities. At the same time, the Chinese economy has entered a new normal, with a slowdown in economic growth and resource-based enterprises entering a period of transformation. The traditional extensive development model of high pollution and high consumption is difficult to sustain. Driven by the transformation of industrial structure and technological upgrading, resource-based enterprises have begun to attach importance to green development, actively explore green production technologies and management models, strive to improve resource utilization efficiency, and reduce pollution emissions.

(2)

Analysis of Within-group results

Observing the scatter (Figure 6) distribution status of the three groupings, it is found that there are still a small number of scatters falling sporadically in the lower part of the picture, showing a certain degree of disorder. The intra-group consistency adjustment distances of the three conditional groupings in the summarized results indicate that there are some differences in the cross-sectional data of different case consistency levels. Further analyzing the cross-sectional data and representative cases of the three groups, it is found that the typical cases of group 2 mainly focus on the resource extraction industries in the upstream of the industrial chain, such as the oil and gas extraction industry and the nonmetallic mineral extraction industry, whereas group 1 and group 3 mainly have better explanatory strength for the downstream resource processing and manufacturing industries, such as the nonmetallic mineral products industry. This may be due to the different profit models and benefit-driven mechanisms, leading to significant differences in the assessment of costs, benefits, and risks of green transformation. Resource extraction enterprises mainly rely on the extraction and sale of resources, and their profits are directly related to resource prices and production. Under the traditional model, economies of scale and low costs are at the core of competitiveness, and investment in environmental protection is seen as increasing costs and reducing profits. Resource processing and manufacturing enterprises rely on resources provided upstream for processing, manufacturing, and sales, and their profits are affected by the price of upstream resources, their own production efficiency, market demand, and the added value of their products, and they may face pressures from various parties, including environmental protection requirements from consumers, environmental protection regulations from the government, and green trade barriers in the international market.

5. Conclusions

This article uses resource-based enterprises as the research sample and conducts an in-depth study on the characteristics and mechanisms driving the improvement of GTFP from both a single-factor and a configurational dual perspective, utilizing machine learning and set theory across different data analysis dimensions. The findings are as follows: (1) Environmental regulation, green cognition, and green investment are important factors in condition configuration and there exists a complex nonlinear relationship among them, indicating the need for further research from a configurational perspective. (2) No necessary conditions for GTFP were found; however, the higher the level of GTFP, the more it is constrained by the green governance capacity of resource-based enterprises, suggesting the presence of causal asymmetry. (3) There are three configurational pathways for the improvement of GTFP in resource-based enterprises, and the explanatory power varies with time and case differences. Based on the above research results and conclusions, combined with the management practices of green transformation in our country, suggestions for countermeasures are proposed on both the enterprise and government levels.

6. Discussion and Future Directions

Resource-based enterprises lack the initiative for green transformation, and the input costs of green transformation come from the enterprises themselves, while the green benefits have to be shared with other stakeholders, which further reduces the willingness of enterprises to participate in green transformation. To this end, government departments can improve the market mechanism, improve the environmental rights and interests trading market, and explore other environmental rights and interests trading modes besides carbon trading, such as water pollution rights trading and solid waste emissions trading, etc., so as to promote the effective allocation of resources and environmental protection, and actively guide enterprises to actively participate in green transformation and enhance green total factor productivity. At the same time, establishing a green behavior point system for enterprises, allowing them to exchange green behavior points for subsidies or tax incentives, can increase their motivation. In addition, the government can, while strengthening environmental supervision, guide banks to participate in green credit and further provide financial support for enterprises.

Green innovation is an important factor in driving high levels of green total factor productivity. For this reason, enterprises should emphasize green innovation inputs and improve the efficiency of results transformation. In terms of green investment, enterprises can strive for more government subsidies while meeting the government’s requirements for environmental protection. At the same time, they can take green responsibility as an opportunity to attract external capital to invest in green innovation projects for social organizations, such as civil non-profit environmental protection organizations, in order to reduce the financial burden on enterprises, while responding to the environmental concerns of the public in a market-oriented manner, while promoting the formation of a pluralistic common governance situation to further expand the sources of funding. Second, in terms of green technology innovation, the upstream and downstream of the industry can establish corporate partnerships, build green technology innovation transformation platforms, share green technology and experience, avoid incorrect green behaviors due to inconsistent parameter specifications, reduce “uneconomical behavior” in different industrial production links, promote the green transformation of the entire resource-based industrial chain, and improve overall green total factor productivity.

This study examines the influencing factors and configuration paths of green total factor productivity in resource-based enterprises from different dimensions, which brings a certain marginal contribution to the advancement of related research. In the future, the application of methods should be further implemented, and the integration of method application should be made to more closely enhance scientificity. For instance, the random forest algorithm can be utilized for the initial screening of influencing factors to identify key elements and reduce redundant conditions, making up for the deficiency in the existing QCA methods in terms of “minimum sufficiency”, promoting their cross-application in multiple fields, and forming a scientific research paradigm.