Unlocking the Digital Dividend: How Does Digitalization Promote Corporate Carbon Emission Reduction?

Abstract

Although digitalization offers new pathways for carbon reduction, its underlying mechanisms have not been fully explored. Unlike previous studies, this research investigates the impact of digitalization on corporate carbon performance through both technological and structural effects while also revealing the boundary conditions under which digitalization contributes to carbon reduction in the context of corporate financing constraints. We conducted an empirical analysis using a fixed-effects model and a partially linear functional-coefficient model based on data from A-share listed companyies in China from 2008 to 2023. The results show that digitalization is significantly and positively associated with corporate carbon performance, confirming its potential for emission reduction. Mechanism tests indicated that digitalization improves corporate carbon performance by enhancing technological absorptive capacity, promoting factor substitution, and optimizing resource allocation. Further analysis revealed that, under financing constraints, the marginal effect of digitalization on corporate carbon performance follows an “inverted U-shaped” curve. Our study enriches the literature on the digital economy and carbon emissions and provides both theoretical and practical insights for promoting the coordinated transformation of enterprises toward digitalization and low-carbon development.

1. Introduction

The fast-growing digital economy has emerged as a key catalyst of worldwide economic transformation and industrial advancement, and it is also a key means whereby countries around the world can enhance their international competitiveness [1]. In particular, digital transformation has become a strategic choice allowing nations to seize new opportunities brought by the latest technological revolution and industrial changes. At present, China’s economy is at a critical stage of structural transformation. The early extensive economic development model has led to increasingly severe environmental problems. The pursuit of high-quality economic growth in China is increasingly challenged by ecological constraints and the legacy of energy-intensive expansion [2]. According to CEADs, China consumed 2.3 billion tons of standard coal in 2022, which resulted in carbon emissions of around 10.523 billion tons, representing 31.2% of total global emissions. In response to the continuous increase in carbon emissions, the Chinese government set the goals of carbon peaking and carbon neutrality (the dual carbon goals) in 2020, designed to serve as a benchmark in promoting sustainable global economic development. As the main source of greenhouse gas emissions, enterprises’ carbon reduction efforts are crucial to whether entire industries and even the country as a whole can achieve the dual carbon goals [3]. Therefore, promoting carbon reduction in enterprises is key to China’s current economic transition toward low-carbon development.

Digital technologies have made significant breakthroughs and are becoming increasingly integrated with the real economy. The Global Digital Economy White Paper (2024) reports that, in 2023, the digital sector contributed 58% to China’s GDP, maintaining a growth pace faster than the national average. Driven by the digital economy boom, enterprises are increasingly undergoing digital transformation, which has emerged as a critical path for business evolution. While digital transformation is well underway, its precise effects on corporate carbon emissions are still not fully understood. The extensive literature suggests that digital economic advancement may help lower carbon emission intensity and accelerate urban low-carbon transitions [4]. Some micro-level studies also suggest that the application of digital technologies enables the dynamic monitoring of product markets, reduces firms’ search and transaction costs, and thus decreases resource consumption [5]. In addition, as a production management tool, digital technology can optimize firms’ product manufacturing processes, reduce resource waste during production, and thereby achieve carbon reduction goals [2]. However, some studies argue that the widespread application of digital technologies increases energy consumption, thereby having a negative impact on the environment [6]. Research to date has investigated the relationship between digitalization and carbon emissions from diverse perspectives, but the conclusions remain inconsistent, and micro-level evidence is limited. There are still questions about what kind of impact digitalization has on corporate carbon performance: Is it positive or negative? How does digitalization affect corporate carbon performance? What are the conditions under which it operates? Addressing these research questions holds substantial importance for both academic research and practical applications, as it contributes to harnessing digital advantages, meeting carbon neutrality targets, and fostering sustainable economic growth.

Based on this, we first conduct a theoretical examination of how digitalization affects corporate carbon performance. Then, using financial statement data, patent text data, and carbon emission data of A-share listed manufacturing companies in China from 2008 to 2023, we measure the levels of corporate digitalization and carbon performance. On this basis, we examine the impact and underlying mechanisms of digitalization on corporate carbon performance from the perspectives of technology absorption and factor substitution. Furthermore, we implement a partially linear functional-coefficient model to analyze how financial constraints exert non-uniform moderating influences.

Relative to prior research, this study provides the following contributions: First, this paper conducts empirical tests using the latest micro-level data. Because the China Industrial Enterprise Pollution Emissions Database only covers the period up to 2014 and the wave of digitalization began to rise around 2012 [7], this study adopts the most recent data from Chinese listed companies to provide up-to-date empirical insights. Second, this paper explores how digitalization affects corporate carbon performance through the lens of financing constraints. The financing environment is crucial to the process of digitalization. Examining this relationship through the lens of corporate financing constraints helps to uncover the underlying logic behind how digitalization drives carbon reduction in firms. Third, this study utilizes a partially linear functional-coefficient model to examine the moderating effect. Considering the nonlinear role of financing constraints in the relationship between digitalization and corporate carbon performance, the use of this model allows for a more nuanced examination of the underlying mechanism. This not only enriches the understanding of how digitalization affects corporate carbon performance but also provides a practical pathway for unlocking the potential benefits of the digital dividend.

The remainder of the paper is organized as follows: Section 2 reviews prior research, highlighting key developments in studies linking the digital economy with carbon emissions. Section 3 presents the theoretical framework and proposes the hypotheses. Section 4 describes the model construction and details the measurement of relevant variables. Section 5 reports the findings of the empirical analysis. Section 6 concludes the paper by summarizing the key insights and proposing related policy implications.

2. Literature Review

Theoretical explorations and empirical analyses of the relationship between digitalization and carbon emissions have yielded notable results. However, due to differences in research perspectives and sample selection, the conclusions vary [8,9]. The main viewpoints can be categorized into three types.

(1) Most studies believe that digitalization can promote carbon emission reductions [10,11,12]. According to Romer’s [13] endogenous growth theory, economic growth is closely related to technological progress. The technological progress driven by digitalization can improve production efficiency and optimize resource allocation. Furthermore, advanced digital machinery can supplant high-emission equipment, lowering energy use and emissions per production unit, thus supporting sustainable development. For example, empirical research by Lyu et al. [14], based on Chinese manufacturing data, demonstrates that digital adoption boosts energy efficiency and facilitates the shift toward greener industrial practices. Wang et al. [15], based on city-level data in China, found that digital economic growth drives better resource distribution and innovation, leading to decreased carbon emissions both locally and in adjacent regions. Research by Ma et al. [16] on Chinese firms shows that digitalization fosters eco-friendly innovation, highlighting its role in advancing corporate sustainability. Lan and Zhou [17] believe that digital factors can not only replace traditional factors as new production factors, exerting the factor substitution effect, but also can be used as a means to optimize resource allocation, exerting the factor reorganization effect. Both effects contribute to enterprises’ shift toward low-carbon development.

(2) However, other studies argue that the digital economy may have limited effects on emission reductions or could even lead to increased carbon outputs [18,19,20]. Neoclassical economic growth theory posits that technological progress leads to economic scale expansion and increases energy consumption. From the perspective of firm production theory, technological advancement improves firm efficiency and shifts the cost curve downward, and the resulting scale effect increases a firm’s energy consumption. Some studies have confirmed the existence of the scale effect. Sadorsky [21] observed that while digital technologies heightened electricity usage, they also indirectly drove up the consumption of fossil fuels like coal. Hittinger and Jaramillo [22], through the quantitative measurement of carbon emissions from cloud computing, confirmed that data centers and cloud servers are key sources of carbon footprints, with data center operations consuming a large amount of energy. Tang and Yang [23] used panel data from Chinese cities and concluded that digital infrastructure development may, to some extent, contribute to higher regional carbon emissions. Rahnamay Bonab et al. [24] argue that the production of digital devices increases carbon emissions through the generation of electronic waste and electricity consumption, which, if left unchecked, may hinder sustainable development. Moreover, the technological progress brought by digitalization is also considered to have a rebound effect, meaning that efficiency improves, but more consumption is generated [25]. For example, Liu et al. [26] observed that, although digital transformation boosts energy efficiency, it can also drive up energy demand, producing a rebound effect that ultimately raises carbon emissions.

(3) Moreover, certain research indicates that digitalization and carbon emissions may be connected through a nonlinear relationship [27]. The nonlinear relationship between digitalization and carbon emissions largely stems from the combined effects of technological progress, scale effects, and rebound phenomena [28]. When the technological effect outweighs the scale effect and rebound effect, digitalization suppresses carbon emissions; otherwise, it does not. Luan et al. [29] developed a theoretical model examining the relationship between automation and environmental pollution, finding that the connection between the application of industrial robots and environmental pollution depends on the relative strengths of technological effects, scale effects, and rebound effects. Huang and Zhang [30] identified an “inverted U-shaped” link between digitalization in manufacturing and carbon emissions: at lower levels, digitalization contributes to emission increases, but beyond a specific threshold, it helps reduce them. Similarly, Li et al. [31] revealed that the digital economy and carbon emission intensity are related in an “inverted U-shaped” manner and found that the spatial spillover effects exhibit a “U-shaped” trend. Wang et al. [32] analyzed data at the city level in China and discovered that digitalization’s effect on carbon emissions shows a threshold pattern, with carbon reduction advantages becoming evident only once a city’s digital transformation surpasses a specific point.

Considering the growing academic interest in the environmental effects of corporate digitalization, we summarize the relevant literature on the relationship between digitalization and carbon emissions, as shown in

Table 1. Relevant literature on the relationship between digitalization and carbon emissions.

Reference	Effect	Independent Variable	Key Findings
Huang and Lin [33]	Technological and structural effects	Digital economy	Digital economy promotes regional carbon reduction through energy efficiency and energy structure optimization.
Wang et al. [12]	Agglomeration economies	Digitalization	Digitalization reduces urban carbon emission intensity by increasing urban population density and promoting green technological innovation.
Wang et al. [15]	Spillover effect	Digital economy	The digital economy can reduce not only local carbon emission intensity but also that of neighboring regions.
Liu et al. [26]	Energy rebound effect	Digital technology	Digital technology innovation can generate an energy rebound effect, which may partially offset the impact of emission reductions.
Rahnamay Bonab et al. [24]	Scale effect	Digitalization	The production of digital devices increases carbon emissions through the generation of electronic waste and electricity consumption.
Tang and Yang [23]	Scale effect	Digital infrastructure	Digital infrastructure has significantly increased carbon emission intensity in Chinese cities.
Wang et al. [34]; Salahuddin and Alam [6]	Scale effect	ICT	The widespread expansion of ICT has led to a sharp increase in energy demand.
Qian et al. [35]	Rebound effect	Digital economy	The digital economy increases local carbon intensity while reducing emissions in neighboring regions.
Bai et al. [36]; Huang and Zhang [30]	Hybrid effect	Digital economy	There is an “inverted U-shaped” relationship between the digital economy and carbon emission intensity.
Wang et al. [32]	Threshold effect	Digital transformation	Digital transformation exerts a carbon reduction effect after reaching a certain threshold.
Li et al. [31]	Nonlinear effect	Digital economy	The digital economy exhibits a nonlinear impact on urban carbon emission efficiency.
Our research	Technological–structural–scale framework	Digitalization	Digitalization improves corporate carbon performance through technology adoption, factor substitution, and optimized resource allocation. Under the influence of financing constraints, the carbon reduction effect of digitalization exhibits an “inverted U-shape”.

. Previous studies have mainly focused on the impact of information and communication technology (ICT), the digital economy, and digital transformation on carbon performance.

The above studies indicate that researchers have examined the link between digitalization and carbon emissions; however, further exploration in this area remains necessary. First, most of the existing literature focuses on the regional and industry levels, with limited attention paid to the micro-level of individual enterprises. Second, although certain research points to a beneficial effect of digitalization on carbon reduction, other studies highlight a nonlinear relationship between digitalization and carbon performance, resulting in mixed findings. These discrepancies are often related to differences in variable measurement methods and the inadequate handling of endogeneity. Finally, the existing research on the mechanisms through which digitalization affects carbon emissions remains incomplete.

To address the above research gaps, this study first analyzes the impact mechanism of digitalization on corporate carbon performance from both technological and structural perspectives. Second, by collecting financial data and carbon emission data of A-share listed companies in China, we measure firm-level digitalization and carbon performance indicators. Third, using fixed effects models and partially linear functional-coefficient models, we reveal the mechanisms and boundary conditions through which digitalization affects corporate carbon performance. Finally, based on the research findings, we provide practical guidance for enterprises to achieve sustainable development through coordinated digital and low-carbon transformation.

3. Theoretical Analysis and Research Hypothesis

3.1. Direct Mechanism

Digitalization refers to the process by which enterprises incorporate digital production factors into their production, operations, and services, and apply advanced digital technologies to transform their management and production activities [37]. Digitalization has multiple impacts on corporate carbon performance. On the one hand, enterprises adopt digital technologies like artificial intelligence and big data in their production processes, which enhance operational efficiency while also promoting energy saving and emission reductions. Specifically, big data technology allows for the real-time tracking of energy use throughout different production stages, effectively improving energy efficiency [38]. On the other hand, digitalization significantly improves operational efficiency and reduces search and transaction costs for enterprises. Specifically, big data technology enables intelligent decision-making, helping firms avoid unnecessary resource consumption. Digital platforms can assist enterprises in identifying and monitoring carbon emissions during the manufacturing process, thereby supporting targeted emission reductions. Moreover, with the adoption of new technologies and factors of production, traditional production inputs are gradually being replaced, leading to improved resource allocation efficiency and, to some extent, reduced corporate carbon emissions [39]. Accordingly, this paper formulates Hypothesis 1:

Hypothesis 1.

Digitalization can improve corporate carbon performance.

3.2. Indirect Mechanism

Digitalization promotes corporate carbon reduction by enhancing technological absorptive capacity. Based on the endogenous growth theory [13], a firm’s production activities require not only the input of physical capital and labor, but also rely on the accumulation, absorption, and reinvention of technological knowledge—factors that are key to long-term improvements in productivity and sustainable growth. Digital technologies are embedded into traditional production equipment in an interactive and mutually reinforcing manner, endowing machines with cognitive and learning capabilities. Intelligent devices learn from past experiences and optimize production processes to achieve maximum efficiency. The application of big data facilitates the continuous accumulation of data resources within enterprises, providing a material foundation for the absorption of new technologies. On the one hand, digital technologies improve firms’ capacity to access and leverage external data, thereby speeding up the adoption and implementation of advanced low-carbon technologies. On the other hand, they reduce the adaptation friction costs associated with production technologies, thereby strengthening firms’ capacity to acquire and absorb new technologies and knowledge in the production process [40]. Moreover, digitalization also encourages firms to hire employees who are proficient in technology, thereby enhancing the level of human capital within the organization and promoting green technological innovation. Therefore, this study proposes Hypothesis 2:

Hypothesis 2.

Digitalization enhances firms’ carbon performance by strengthening their capacity to absorb.

Digitalization promotes corporate carbon emission reduction by exerting a factor substitution effect. From a producer’s perspective, digital equipment can substitute for labor engaged in simple production tasks, thereby increasing the proportion of skilled technical workers within a firm. Digitalization also allows employees more time to learn new technologies, thereby enhancing their intellectual capital. This is crucial for improving a firm’s carbon performance. On the one hand, the improvement in employee skill levels drives the clean production cost curve downward, reducing the unit cost of clean products [28]. On the other hand, capable managers tend to have stronger environmental awareness and are more proficient in supply chain management, which facilitates the green transformation of enterprises. Moreover, digitalization also reduces the input of traditional capital factors. For example, the digitalization of production management not only optimizes production processes but also enables the precise scheduling of energy consumption, thereby achieving energy saving and emission mitigation. Therefore, this study puts forward Hypothesis 3:

Hypothesis 3.

Digitalization enhances corporate carbon performance by exerting a factor substitution effect.

From the perspective of corporate resource allocation, digitalization can improve energy efficiency by optimizing the allocation of production factors, thereby enhancing corporate carbon performance [41]. First, digitalization improves firms’ capacity to obtain and analyze information, thereby lowering information asymmetry and transaction costs in the allocation of factors, and promoting the flow of resources toward efficient and low-carbon areas. According to transaction cost theory, information asymmetry often leads to inefficient resource allocation [42]. Through the implementation of technologies like big data and artificial intelligence, companies can effectively process cross-departmental data, enabling a holistic view of market, customer, and product insights, thereby reducing transaction costs [43]. Second, by leveraging the Internet of Things (IoT) and artificial intelligence technologies to track equipment performance and predict demand in advance, firms can reduce unplanned downtime, thereby lowering energy consumption and carbon emissions. Finally, digitalization can also enhance supply chain management capabilities, enabling firms to more effectively match supply with demand, thereby reducing logistics and inventory costs. Accordingly, the study puts forward Hypothesis 4:

Hypothesis 4.

Digitalization enhances corporate carbon performance through enhancing the efficiency of resource allocation.

3.3. The Role of Financing Constraints

Digitalization can effectively enhance corporate carbon performance by promoting intelligent production processes, improving resource allocation efficiency, and strengthening environmental monitoring capabilities. However, digitalization is often accompanied by substantial initial investments, such as the construction of information infrastructure, the purchase of intelligent equipment, and technological research and development, leading to strong capital demands. Therefore, corporate financing constraints become an important moderating factor affecting the effectiveness of digitalization [44]. When a firm faces high financing constraints, its resources available for digital transformation are limited, resulting in a slower digitalization process and thus weakening the carbon reduction capability of digitalization. In contrast, firms with a relaxed financing environment and smooth access to funding are more likely to achieve digital upgrades, thereby unlocking the benefits of digitalization. Moreover, financing constraints also affect a firm’s risk-taking capacity in green innovation. For firms facing financing constraints, their willingness to invest in green digitalization projects will significantly decrease, thereby hindering the advancement of green low-carbon transformation paths [45]. Accordingly, this paper puts forward Hypothesis 5:

Hypothesis 5.

Financing constraints have a nonlinear moderating effect on the relationship between digitalization and corporate carbon performance.

4. Research Design

4.1. Model

4.1.1. Baseline Model

To examine the impact of digitalization on corporate carbon performance, this paper specifies the baseline model as follows:

$$\mathrm{E}\mathrm{G}\mathrm{l}\mathrm{c}{\mathrm{p}}_{\mathrm{i}\mathrm{j}\mathrm{t}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}{\mathrm{l}}_{\mathrm{i}\mathrm{j}\mathrm{t}}+\sum {\mathsf{\alpha }}_{\mathrm{x}}\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}{\mathrm{l}}_{\mathrm{i}\mathrm{j}\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\mu }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{j}\mathrm{t}}$$

(1)

where $\mathrm{i}$, $\mathrm{j}$, and $\mathrm{t}$ represent firm, industry, and time, respectively. $\mathrm{E}\mathrm{G}\mathrm{l}\mathrm{c}{\mathrm{p}}_{\mathrm{i}\mathrm{j}\mathrm{t}}$ is the dependent variable, representing the carbon performance of manufacturing firm $\mathrm{i}$ in time $\mathrm{t}$. $\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}{\mathrm{l}}_{\mathrm{i}\mathrm{j}\mathrm{t}}$ is the independent variable, representing the level of digitalization of manufacturing firm $\mathrm{i}$ in time $\mathrm{t}$. $\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}{\mathrm{l}}_{\mathrm{i}\mathrm{j}\mathrm{t}}$ represents the set of control variables included in this model. To control for time-invariant firm-level confounding factors and time-varying confounding factors at the year level, this study incorporates firm fixed effects and time fixed effects into the model, denoted as ${\mathsf{\mu }}_{\mathrm{i}}$ and ${\mathsf{\mu }}_{\mathrm{t}}$, respectively. ${\mathsf{\alpha }}_0$ represents the intercept term of the model. ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$ represents the random error term, assumed to be independently and identically distributed with finite variance. If ${\mathsf{\alpha }}_1$ is statistically significant and positive, it suggests that digitalization enhances corporate carbon performance.

4.1.2. Mediation Effect Model

This paper conducts a mechanism test by examining the impact of the independent variable on the mediating variable. The model is specified as follows:

$$\mathrm{M}\mathrm{e}\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{i}\mathrm{s}{\mathrm{m}}_{\mathrm{i}\mathrm{j}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}{\mathrm{l}}_{\mathrm{i}\mathrm{j}\mathrm{t}}+\sum {\mathsf{\beta }}_{\mathrm{x}}\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}{\mathrm{l}}_{\mathrm{i}\mathrm{j}\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\mu }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{j}\mathrm{t}}$$

(2)

where $\mathrm{M}\mathrm{e}\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{i}\mathrm{s}{\mathrm{m}}_{\mathrm{i}\mathrm{j}\mathrm{t}}$ represents the mediating variable, which refers to the technological absorption effect, factor substitution effect, and resource allocation effect. The remaining variable specifications align with those in the baseline model.

4.1.3. Partially Linear Functional-Coefficient Model

The partially linear functional-coefficient model is specified as follows:

$$\mathrm{T}\mathrm{l}\mathrm{c}{\mathrm{p}}_{\mathrm{i}\mathrm{j}}={\mathsf{\beta }}_0+\mathrm{G}\left(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}\right)\times \mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}{\mathrm{l}}_{\mathrm{i}\mathrm{t}}+\mathsf{\rho }{\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\delta }}_{\mathrm{t}}+{\mathsf{\xi }}_{\mathrm{i}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(3)

where $\mathrm{G}\left(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}\right)$ is an unknown function representing the moderating effect of the moderating variable on the marginal effect of digitalization. $\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}$ is the firm’s financing constraint index, measured by the Size–Age index [46]. The remaining variables are consistent with those mentioned above. This paper follows the sequential estimation method proposed by An et al. [47] and Zhang and Zhou [48] to estimate the unknown function. First, a varying coefficient function is expressed as a linear combination of a set of basic functions:

$$\mathrm{h}{(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}})}^{\prime }\mathsf{\eta }=\left[{\mathrm{h}}_1\left(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}\right),\dots ,{\mathrm{h}}_{\mathrm{k}}\left(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}\right)\right]\left[\begin{array}{c}{\mathsf{\eta }}_1\\ ⋮\\ {\mathsf{\eta }}_{\mathrm{k}}\end{array}\right]$$

(4)

where $\mathrm{h}\left(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}\right)={\left[{\mathrm{h}}_1\left(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}\right),\dots ,{\mathrm{h}}_{\mathrm{k}}\left(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}\right)\right]}^{\prime }$ represents a $\mathrm{k}\times 1$ vector of basis functions, and $\mathsf{\eta }=[{\mathsf{\eta }}_1,\dots ,{\mathsf{\eta }}_{\mathrm{k}}{]}^{\prime }$ is a $\mathrm{k}\times 1$ vector of unknown parameters. As $\mathrm{k}$ increases, the linear combination of the basis functions $\mathrm{h}\left(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}\right)$ can approximate the unknown function $\mathrm{G}\left(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}\right)$, and the approximation error tends to zero. Therefore, Equation (3) can be rewritten as follows:

$$\mathrm{T}\mathrm{l}\mathrm{c}{\mathrm{p}}_{\mathrm{i}\mathrm{j}}={\mathsf{\beta }}_0+\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}{\mathrm{l}}_{\mathrm{i}\mathrm{t}}\times \mathrm{h}(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}{)}^{\prime }\mathsf{\eta }+\mathsf{\rho }{\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\delta }}_{\mathrm{t}}+{\mathsf{\xi }}_{\mathrm{i}}+{\mathrm{v}}_{\mathrm{i}\mathrm{t}}$$

(5)

In Equation (5), ${\mathrm{v}}_{\mathrm{i}\mathrm{t}}={\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}+\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}{\mathrm{l}}_{\mathrm{i}\mathrm{t}}\times \mathrm{G}\left(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}\right)-\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}{\mathrm{l}}_{\mathrm{i}\mathrm{t}}\times \mathrm{h}(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}{)}^{\prime }\mathsf{\eta }$ represents the approximation error of the sieve basis functions. By further eliminating the fixed effects in Equation (5), the model can be expressed as follows:

$$\mathsf{\Delta }\mathrm{T}\mathrm{l}\mathrm{c}{\mathrm{p}}_{\mathrm{i}\mathrm{j}}=\mathsf{\Delta }\left[\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}{\mathrm{l}}_{\mathrm{i}\mathrm{t}}\times \mathrm{h}(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}{)}^{\prime }\right]\mathsf{\eta }+\mathsf{\rho }\mathsf{\Delta }{\mathrm{X}}_{\mathrm{i}\mathrm{t}}+\mathsf{\Delta }{\mathrm{v}}_{\mathrm{i}\mathrm{t}}$$

(6)

Assuming all explanatory variables satisfy the exogeneity condition, Equation (6) can be estimated by the ordinary least squares (OLS) method to obtain the estimates of the coefficient vectors $(\widehat{\mathsf{\eta }},\widehat{\mathsf{\rho }})$.

$$(\widehat{\mathsf{\eta }},\widehat{\mathsf{\rho }})={\left[\mathsf{\Delta }{\tilde{\mathrm{X}}}^{\prime }\mathsf{\Delta }\tilde{\mathrm{X}}\right]}^{-1}\mathsf{\Delta }{\tilde{\mathrm{X}}}^{\prime }\mathsf{\Delta }\widetilde{\mathrm{T}\mathrm{l}\mathrm{c}\mathrm{p}}$$

(7)

Therefore, the varying coefficient function can be estimated by $\widehat{\mathrm{G}}(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}}{)=\mathrm{h}(\mathrm{E}{\mathrm{c}}_{\mathrm{i}\mathrm{t}})}^{\prime }\widehat{\mathsf{\eta }}$.

Figure 1 illustrates the framework of the research and methodology that guides our empirical analysis.

4.2. Variable Measurement

The dependent variable is enterprise carbon performance ($\mathrm{E}\mathrm{G}\mathrm{l}\mathrm{c}\mathrm{p}$), calculated as the ratio of a firm’s operating revenue over its carbon emissions. Due to the lack of energy consumption and carbon emission data in the disclosures of Chinese listed companies, this study innovatively estimates firm-level carbon emissions using provincial-level carbon emission data for subdivided manufacturing industries. The estimation process for provincial-level carbon emissions by sub-industry is as follows:

$${\mathrm{C}}_{\mathrm{g}\mathrm{j}\mathrm{t}}=\sum _{\mathrm{k}}{\mathrm{e}}_{\mathrm{g}\mathrm{j}\mathrm{k}\mathrm{t}}\times {\mathrm{f}}_{\mathrm{k}}\times {\mathrm{c}}_{\mathrm{k}}$$

(8)

In Equation (8). ${\mathrm{C}}_{\mathrm{g}\mathrm{j}\mathrm{t}}$ represents the carbon emissions of industry $\mathrm{j}$ in province $\mathrm{g}$ during period $\mathrm{t}$, which are determined by energy usage and corresponding carbon emission factors. ${\mathrm{e}}_{\mathrm{g}\mathrm{j}\mathrm{k}\mathrm{t}}$ denotes the consumption of energy type $\mathrm{k}$ by industry $\mathrm{j}$ in province $\mathrm{g}$ during period $\mathrm{t}$. ${\mathrm{f}}_{\mathrm{k}}$ is the average lower heating value of energy type $\mathrm{k}$, and ${\mathrm{c}}_{\mathrm{k}}$ is the carbon emission coefficient for energy type $\mathrm{k}$. Data on total energy consumption for subdivided industries in each province, as well as the average lower heating values and carbon emission coefficients of various energy types, are obtained from the CEADs database and the China Energy Statistical Yearbook. Based on the provincial-level carbon emissions for subdivided industries, firm-level carbon emissions are further estimated as follows:

$$\mathrm{E}\mathrm{G}{\mathrm{C}}_{\mathrm{g}\mathrm{i}\mathrm{t}}={\mathrm{C}}_{\mathrm{g}\mathrm{j}\mathrm{t}}\times {\displaystyle \frac{{\mathrm{Q}}_{\mathrm{g}\mathrm{i}\mathrm{t}}}{{\sum }_{\mathrm{i}}{\mathrm{Q}}_{\mathrm{g}\mathrm{i}\mathrm{t}}}}$$

(9)

In Equation (9), $\mathrm{E}\mathrm{G}{\mathrm{C}}_{\mathrm{g}\mathrm{i}\mathrm{t}}$ represents the carbon emissions of firm $\mathrm{i}$ in province $\mathrm{g}$ during period $\mathrm{t}$. ${\mathrm{Q}}_{\mathrm{g}\mathrm{i}\mathrm{t}}$ denotes the operating costs of firm $\mathrm{i}$ in province $\mathrm{g}$ during period $\mathrm{t}$, and ${\sum }_{\mathrm{i}}{\mathrm{Q}}_{\mathrm{g}\mathrm{i}\mathrm{t}}$ denotes the aggregate operating costs of all firms in the same industry and province, g, during the corresponding period. The data on provincial-level industry operating income and operating costs are obtained from the China Industrial Statistical Yearbook. Firm-level operating income and cost data are obtained from the CSMAR database. The carbon performance of firms is calculated as follows:

$$\mathrm{E}\mathrm{G}\mathrm{l}\mathrm{c}{\mathrm{p}}_{\mathrm{i}\mathrm{t}}={\displaystyle \frac{\mathrm{E}\mathrm{G}{\mathrm{C}}_{\mathrm{i}\mathrm{t}}}{{\mathrm{W}}_{\mathrm{i}\mathrm{t}}}}$$

(10)

In Equation (10), $\mathrm{E}\mathrm{G}\mathrm{l}\mathrm{c}{\mathrm{p}}_{\mathrm{i}\mathrm{t}}$ represents the carbon productivity of firm $\mathrm{i}$ in period $\mathrm{t}$, and ${\mathrm{W}}_{\mathrm{i}\mathrm{t}}$ denotes the operating revenue of firm $\mathrm{i}$ in period $\mathrm{t}$.

The independent variable is enterprise digitalization ($\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{l}$), which is measured by the proportion of digital assets. Following the study by Chen et al. [49], intangible asset items containing keywords such as “intelligent”, “software”, “information platform”, “digital”, and “system” are classified as digital software investments, while fixed asset items containing keywords such as “computer”, “electronic equipment”, and “data equipment” are identified as digital hardware investments. The digital assets of listed companies are identified based on the notes to financial statements available in the RESSET Financial Research Database. If the name of a specific asset item contains any of the “digitalization” keywords, it is regarded as a digital-related investment. Finally, the digital components of intangible and fixed assets are aggregated to calculate the total digital assets of the firm. The digitalization level of a listed company is then computed using the following formula:

$${\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{l}}_{\mathrm{i}\mathrm{t}}={\displaystyle \frac{\sum \mathrm{W}\text{\_}\mathrm{d}\mathrm{i}{\mathrm{g}}_{\mathrm{i}\mathrm{t}}}{\sum {\mathrm{W}}_{\mathrm{i}\mathrm{t}}}}$$

(11)

In Equation (11), ${\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{l}}_{\mathrm{i}\mathrm{t}}$ represents the level of digitalization of firm $\mathrm{i}$ in year $\mathrm{t}$. $\mathrm{W}\text{\_}\mathrm{d}\mathrm{i}{\mathrm{g}}_{\mathrm{i}\mathrm{t}}$ denotes the total value of digital intangible assets and digital fixed assets of firm $\mathrm{i}$ in year $\mathrm{t}$, while ${\mathrm{W}}_{\mathrm{i}\mathrm{t}}$ is the total value of intangible assets and fixed assets of firm $\mathrm{i}$ in year $\mathrm{t}$.

For the control variables, informed by a previous study by Yang et al. [7], the following are selected: Firm size ($\mathrm{S}\mathrm{i}\mathrm{z}\mathrm{e}$): quantified as the natural logarithm of total employees. According to the Environmental Kuznets Curve (EKC) hypothesis, firm size can influence both output and pollution emissions. Firm age ($\mathrm{A}\mathrm{g}\mathrm{e}$): measured as the logarithm of the current year minus the year of establishment, plus one. Leverage ratio ($\mathrm{L}\mathrm{e}\mathrm{v}$): calculated as the proportion of total liabilities relative to total assets. Return on equity ($\mathrm{R}\mathrm{o}\mathrm{e}$): measured as the ratio of net profit to shareholders’ equity. Ownership concentration ($\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}$): measured by the percentage of total shares owned by the largest shareholder. Board size ($\mathrm{B}\mathrm{o}\mathrm{a}\mathrm{r}\mathrm{d}$): calculated as the natural logarithm of board size. CEO duality ($\mathrm{D}\mathrm{u}\mathrm{a}\mathrm{l}$): assigned a value of 1 if the chairman and general manager are the same person, and 0 otherwise. Level of economic development ($\mathrm{P}\mathrm{g}\mathrm{d}\mathrm{p}$): measured by the natural logarithm of per capita GDP of the city where the firm is located. Environmental regulation at the city level ($\mathrm{G}\mathrm{z}$): calculated as the ratio of environment-related term frequency to the total word count in the annual government work reports of the cities where the firms are located.

4.3. Data and Sample

The financial information of listed companies used in this study is obtained from the RESSET Financial Database and the CSMAR database, while the carbon emission data are sourced from the CEADs database and the China Energy Statistical Yearbook. Considering data availability and timeliness, this study selects A-share listed manufacturing companies in China from 2008 to 2023 as the research sample. Missing values in some variables are supplemented using linear interpolation. Descriptive statistics of the variables are presented in

Table 2. Descriptive statistics of variables.

Variables	Obs	Mean	Std. Dev.	Min	Max
$\mathrm{E}\mathrm{G}\mathrm{l}\mathrm{c}\mathrm{p}$	20,922	8.408	1.819	3.204	15.32
$\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{l}$	20,922	0.272	0.473	0.000	4.201
$\mathrm{l}\mathrm{n}\mathrm{S}\mathrm{i}\mathrm{z}\mathrm{e}$	20,922	5.430	1.149	2.809	9.364
$\mathrm{l}\mathrm{n}\mathrm{A}\mathrm{g}\mathrm{e}$	20,922	1.969	0.914	0.000	3.701
$\mathrm{L}\mathrm{e}\mathrm{v}$	20,922	0.390	0.192	0.040	0.923
$\mathrm{R}\mathrm{o}\mathrm{e}$	20,922	0.060	0.148	−1.152	0.387
$\mathrm{l}\mathrm{n}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}$	20,922	3.423	0.459	2.016	4.347
$\mathrm{l}\mathrm{n}\mathrm{B}\mathrm{o}\mathrm{a}\mathrm{r}\mathrm{d}$	20,922	2.116	0.190	1.609	2.709
$\mathrm{D}\mathrm{u}\mathrm{a}\mathrm{l}$	20,922	0.311	0.461	0.000	1.000
$\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{g}\mathrm{d}\mathrm{p}$	20,922	11.39	0.551	9.603	12.25
$\mathrm{G}\mathrm{z}$	20,922	0.802	0.237	0.278	1.564

.

5. Results Analysis

5.1. Baseline Regression Results

Table 3. Baseline regression results.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$
$\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{l}$	0.9047 ***	0.1207 **	0.8485 ***	0.1163 **	0.6946 ***	0.1158 **
	(15.5946)	(2.4485)	(14.8620)	(2.3801)	(12.2116)	(2.3821)
$\mathrm{l}\mathrm{n}\mathrm{S}\mathrm{i}\mathrm{z}\mathrm{e}$			0.0625	0.0386	0.0682 *	0.0389
			(1.6377)	(1.2803)	(1.8775)	(1.3002)
$\mathrm{l}\mathrm{n}\mathrm{A}\mathrm{g}\mathrm{e}$			−0.0990 **	−0.0606 **	−0.1007 ***	−0.0565 **
			(−2.5650)	(−2.1030)	(−2.7775)	(−1.9661)
$\mathrm{L}\mathrm{e}\mathrm{v}$			−1.4876 ***	−0.0326	−1.4568 ***	−0.0390
			(−7.5028)	(−0.3032)	(−7.5699)	(−0.3635)
$\mathrm{R}\mathrm{o}\mathrm{e}$			0.2798 *	0.3910 ***	0.1945	0.3900 ***
			(1.8387)	(7.1470)	(1.4191)	(7.1213)
$\mathrm{l}\mathrm{n}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}$			−0.3747 ***	0.0047	−0.3320 ***	0.0079
			(−4.6937)	(0.0849)	(−4.3630)	(0.1424)
$\mathrm{l}\mathrm{n}\mathrm{B}\mathrm{o}\mathrm{a}\mathrm{r}\mathrm{d}$			−0.9374 ***	0.0421	−0.5994 ***	0.0432
			(−6.1871)	(0.7917)	(−4.0947)	(0.8121)
$\mathrm{D}\mathrm{u}\mathrm{a}\mathrm{l}$			0.3526 ***	0.0347	0.2148 ***	0.0367
			(5.9256)	(1.3107)	(3.7791)	(1.3852)
$\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{g}\mathrm{d}\mathrm{p}$					1.0232 ***	0.2556 ***
					(16.7680)	(2.6670)
$\mathrm{G}\mathrm{z}$					−10.4878	−1.2981
					(−1.0957)	(−0.3180)
$\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}$	−1.0486 ***	−0.8353 ***	2.8302 ***	−0.8733 ***	−9.5271 ***	−3.7940 ***
	(−23.6580)	(−62.2943)	(6.2461)	(−3.9543)	(−10.5745)	(−3.3921)
Firm FE	$\mathrm{N}\mathrm{O}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{N}\mathrm{O}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{N}\mathrm{O}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$
Time FE	$\mathrm{N}\mathrm{O}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{N}\mathrm{O}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{N}\mathrm{O}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$
$\mathrm{O}\mathrm{b}\mathrm{s}$	20,922	20,922	20,922	20,922	20,922	20,922
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.0554	0.8964	0.1207	0.8973	0.2118	0.8976

Note: *, **, and *** indicate statistical significance at the 10%, 5%, and 1% confidence levels, respectively. Numbers in parentheses represent t-statistics adjusted for firm-level clustering. The same applies to the tables below.

presents the primary estimation findings regarding the relationship between digitalization and corporate carbon performance. Whether or not control variables and fixed effects are incorporated, the estimated coefficients of digitalization remain significantly positive. In other words, at the current stage, corporate digitalization contributes to carbon emission reductions. In particular, an increase of one standard deviation in digitalization leads to an 11.58% improvement in corporate carbon performance. This indicates that the continuous accumulation of intelligent elements, such as data and smart algorithms brought about by digitalization, generates a positive feedback effect, which outweighs the potential crowding-out risk of green investment. Hypothesis 1 is thus supported.

5.2. Robustness Test

5.2.1. Addressing the Omitted Variable Problem

The carbon performance of enterprises is influenced by a wide range of internal and external factors. The baseline regression model primarily considers internal economic variables and does not account for external factors such as public environmental concern or environmental policy shocks. The omission of these variables may lead to serious endogeneity issues. Therefore, this study addresses the omitted variable problem from two aspects: by adding control variables and by excluding the influence of environmental policy shocks. First, the study incorporates environmental concern ($\mathrm{E}\mathrm{n}\mathrm{v}\mathrm{i}\mathrm{r}$) into the baseline model to mitigate potential estimation bias. Environmental concern is measured as the logarithmic value of a firm’s environmental information disclosure. As shown in Column (1) of

Table 4. Results of robustness test.

Variables	(1)	(2)	(3)	(4)
	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$
$\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{l}$	0.1138 **	0.1135 **	0.0694 *	0.1040 **
	(2.3221)	(2.3558)	(1.8912)	(2.0689)
$\mathrm{l}\mathrm{n}\mathrm{S}\mathrm{i}\mathrm{z}\mathrm{e}$	0.0362	0.0431	0.0373	0.0292
	(1.2052)	(1.2905)	(1.3865)	(0.9138)
$\mathrm{l}\mathrm{n}\mathrm{A}\mathrm{g}\mathrm{e}$	−0.0583 **	−0.0283	−0.0877 ***	−0.0581 *
	(−2.0244)	(−0.9001)	(−3.0960)	(−1.8495)
$\mathrm{L}\mathrm{e}\mathrm{v}$	−0.0452	−0.0514	−0.0179	−0.0572
	(−0.4219)	(−0.4409)	(−0.1697)	(−0.4871)
$\mathrm{R}\mathrm{o}\mathrm{e}$	0.3935 ***	0.3816 ***	0.3468 ***	0.3963 ***
	(7.2176)	(6.4510)	(6.4359)	(6.9117)
$\mathrm{l}\mathrm{n}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}$	0.0039	0.0259	−0.0118	0.0175
	(0.0697)	(0.4163)	(−0.2236)	(0.2956)
$\mathrm{l}\mathrm{n}\mathrm{B}\mathrm{o}\mathrm{a}\mathrm{r}\mathrm{d}$	0.0440	0.0361	0.0419	0.0567
	(0.8241)	(0.4453)	(0.8632)	(1.0365)
$\mathrm{D}\mathrm{u}\mathrm{a}\mathrm{l}$	0.0361	0.0111	0.0427 *	0.0196
	(1.3706)	(0.3873)	(1.7536)	(0.6961)
$\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{g}\mathrm{d}\mathrm{p}$	0.2692 ***	0.4448 ***	0.1284	0.2131 **
	(2.8063)	(4.2346)	(1.2525)	(2.1874)
$\mathrm{G}\mathrm{z}$	−0.9755	−6.5877	2.4376	−1.8830
	(−0.2389)	(−1.5746)	(0.6130)	(−0.4250)
$\mathrm{E}\mathrm{n}\mathrm{v}\mathrm{i}\mathrm{r}$	−0.0007
	(−0.1057)
$\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}$	−3.9635 ***	−6.0888 ***	−2.1111 *	−3.2740 ***
	(−3.5399)	(−4.9451)	(−1.7787)	(−2.8420)
Firm FE	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$
Time FE	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$
$\mathrm{O}\mathrm{b}\mathrm{s}$	20,922	15,472	18,300	17,322
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.8979	0.8989	0.9136	0.9027

, the regression coefficient of digitalization remains significantly positive, indicating the robustness of the baseline regression results. Second, the study excludes the potential impact of environmental policy shocks. China’s national carbon emission trading market officially launched in July 2021. As one of the core policy instruments under the dual carbon goals, it is expected to have a broad influence on firms’ carbon performance. To eliminate the potential interference of this policy, samples from 2021 onward are excluded from the analysis. As shown in Column (2) of Table 4, the magnitude and direction of the digitalization coefficient remain consistent with the baseline regression, further confirming the robustness of the main findings.

5.2.2. Excluding Other Policy Interference

In 2012, China unveiled the rollout plan for the “Broadband China Strategy” for the first time. Since that time, many internet companies have emerged, and the broad uptake of mobile internet and smartphones has hastened the growth of the digital economy ecosystem. Therefore, 2012 marks a critical turning point in the development of China’s digital economy. To exclude potential interference from this policy initiative, we re-estimate the model by excluding samples prior to 2012. Column (3) of Table 4 shows that the estimation results align with the baseline regression, confirming the robustness of the initial findings.

5.2.3. Excluding Samples Without Digital Technology Innovation

The profit-seeking nature of listed companies may lead some firms to engage in conspicuous asset investment or strategic information disclosure, potentially overstating their digital asset information and disclosure levels. To control for the possible impact of such strategic behavior on the textual estimation results, this study conducts a robustness check by excluding sample firms that have not engaged in digital technology innovation. In Column (4) of Table 4, the independent variable’s coefficients continue to be significantly positive, demonstrating the robustness of the core regression outcomes.

5.3. Endogeneity Test

Despite controlling for various factors affecting manufacturing firms’ carbon performance, unobserved variables may still impact the results, causing potential omitted variable bias in the model. For instance, changes in internal corporate structure, technological progress, and industry restructuring may promote enterprise digitalization while also affecting carbon performance to some extent. To mitigate potential endogeneity issues in the model, this paper employs the instrumental variable approach.

First, following Guo et al. [50], this paper employs the average digitalization level of other firms in the same industry as an instrumental variable for the firm’s own digitalization ($\mathrm{I}\mathrm{V}\text{\_}\mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n}$). Due to the presence of peer effects within the industry, a firm’s digitalization serves as a demonstration and imitation model for other firms in the industry, thus satisfying the relevance condition. Additionally, the digitalization of peer firms within the same industry does not have a direct impact on the focal firm’s carbon performance, thus fulfilling the exclusion restriction requirement.

Second, drawing on Lewbel’s [51] approach of constructing instrumental variables based on internal data, this paper constructs an instrumental variable for digitalization ($\mathrm{I}\mathrm{V}\text{\_}\mathrm{L}\mathrm{e}\mathrm{w}\mathrm{b}\mathrm{e}\mathrm{l}$). Specifically, it uses the cube of the difference between the firm-level digitalization level and the average digitalization level calculated at the two-digit industry and provincial levels as the instrumental variable [52]. Third, referring to the study by published literature [53,54,55], this paper constructs a Bartik instrumental variable ($\mathrm{I}\mathrm{V}\text{\_}\mathrm{B}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{i}\mathrm{k}$) using the shift-share method:

$$\mathrm{I}\mathrm{V}\text{\_}\mathrm{B}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{i}{\mathrm{k}}_{\mathrm{i}\mathrm{t}}=\sum _{\mathrm{i}\notin \mathrm{j}}{\mathsf{\omega }}_{\mathrm{j}}^{{\mathrm{t}}_0}\times \mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{l}\text{\_}\mathrm{g}\mathrm{r}\mathrm{o}{\mathrm{w}}_{\mathrm{t}}$$

(12)

In Equation (12), $\mathrm{i}$ represents the firm, $\mathrm{j}$ indicates the industry, and $\mathrm{t}$ indicates time. ${\mathsf{\omega }}_{\mathrm{j}}^{{\mathrm{t}}_0}$ denotes the initial share (an exogenous variable), which here represents the average digitalization level of other firms within the same two-digit industry as the firm in 2008. $\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{l}\text{\_}\mathrm{g}\mathrm{r}\mathrm{o}{\mathrm{w}}_{\mathrm{t}}$ is the national digitalization growth rate (common shock). The initial share uses the digitalization shares of other entities in the industry, while the common shock component corresponds to demand-side shocks. Using national-level data and growth rates helps ensure that the instrumental variable fulfills the condition of exogeneity.

Table 5. Results of endogeneity test.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	$\mathbf{D}\mathbf{i}\mathbf{g}\mathbf{i}\mathbf{t}\mathbf{a}\mathbf{l}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$	$\mathbf{D}\mathbf{i}\mathbf{g}\mathbf{i}\mathbf{t}\mathbf{a}\mathbf{l}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$	$\mathbf{D}\mathbf{i}\mathbf{g}\mathbf{i}\mathbf{t}\mathbf{a}\mathbf{l}$	$\mathbf{E}\mathbf{G}\mathbf{l}\mathbf{c}\mathbf{p}$
$\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{l}$		5.3733 ***		0.0777 *		0.1144 **
		(4.5532)		(1.9189)		(2.4125)
$\mathrm{I}\mathrm{V}\text{\_}\mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n}$	0.5666 ***
	(4.5791)
$\mathrm{I}\mathrm{V}\text{\_}\mathrm{L}\mathrm{e}\mathrm{w}\mathrm{b}\mathrm{e}\mathrm{l}$			0.1902 ***
			(25.1207)
$\mathrm{I}\mathrm{V}\text{\_}\mathrm{B}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{i}\mathrm{k}$					0.9356 ***
					(117.5446)
$\mathrm{l}\mathrm{n}\mathrm{S}\mathrm{i}\mathrm{z}\mathrm{e}$	−0.0090	0.0734	0.0026	0.0387	0.0020	0.0389
	(−0.7413)	(1.1801)	(0.3314)	(1.2920)	(1.4372)	(1.2999)
$\mathrm{l}\mathrm{n}\mathrm{A}\mathrm{g}\mathrm{e}$	−0.0201	0.0075	−0.0266 **	−0.0570 **	−0.0032	−0.0566 **
	(−1.1851)	(0.1145)	(−2.3529)	(−1.9819)	(−1.5980)	(−1.9669)
$\mathrm{L}\mathrm{e}\mathrm{v}$	−0.0154	0.0971	0.0027	−0.0400	0.0014	−0.0391
	(−0.3779)	(0.4372)	(0.1015)	(−0.3726)	(0.3828)	(−0.3638)
$\mathrm{R}\mathrm{o}\mathrm{e}$	0.0031	0.3642 ***	0.0041	0.3902 ***	−0.0020	0.3900 ***
	(0.1535)	(3.1298)	(0.2930)	(7.1235)	(−0.9082)	(7.1215)
$\mathrm{l}\mathrm{n}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}$	0.0235	−0.0886	−0.0022	0.0086	−0.0011	0.0080
	(0.9865)	(−0.6487)	(−0.1346)	(0.1550)	(−0.6027)	(0.1429)
$\mathrm{l}\mathrm{n}\mathrm{B}\mathrm{o}\mathrm{a}\mathrm{r}\mathrm{d}$	0.0377	−0.1403	0.0178	0.0446	0.0038	0.0433
	(1.4594)	(−1.0071)	(1.0049)	(0.8362)	(1.4449)	(0.8130)
$\mathrm{D}\mathrm{u}\mathrm{a}\mathrm{l}$	0.0099	−0.0229	0.0081	0.0371	0.0006	0.0367
	(0.9208)	(−0.3768)	(1.1920)	(1.4022)	(0.6219)	(1.3860)
$\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{g}\mathrm{d}\mathrm{p}$	−0.0099	0.2204	−0.0128	0.2558 ***	−0.0011	0.2556 ***
	(−0.3585)	(1.4156)	(−0.6884)	(2.6664)	(−0.4145)	(2.6670)
$\mathrm{G}\mathrm{z}$	−0.7049	−0.1036	−0.3428	−1.3068	0.1051	−1.2984
	(−0.4877)	(−0.0128)	(−0.3442)	(−0.3202)	(0.7028)	(−0.3181)
Firm FE	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$
Time FE	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$
Kleibergen-Paap rk LM		17.827 [0.0000]		71.166 [0.0000]		144.049 [0.0000]
Kleibergen-Paap rk Wald F		20.968 {8.96}		631.048 {8.96}		14,000.00 {8.96}
$\mathrm{O}\mathrm{b}\mathrm{s}$	20,922	20,922	20,922	20,922	20,922	20,922
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.7887	−4.0249	0.9049	0.0137	0.9950	0.0139

Note: KP rk LM statistic is an identifiable test of instrumental variables, and the KP rk Wald F statistic is used to test for weak instruments. The value in [] corresponds to the p-value of the LM test, and the value in {} corresponds to the critical value of Stock-Yogo 15%.

presents the results from the instrumental variable estimations. The first-stage estimation results for all three instruments are significant at the 1% level, demonstrating a strong association between the instruments and digitalization. The second-stage results show that the model passes both the under-identification test and the weak instrument test. After accounting for endogeneity bias, the estimated effect of digitalization on the carbon performance of manufacturing firms remains significantly positive, confirming the reliability of the main findings of this research.

5.4. Mediation Mechanism Analysis

(1) This study tests the mediating role of technological absorptive capacity ($\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{h}$) in the relationship between digitalization and carbon performance. Wang et al. [56] use the ratio of R&D expenditure to total assets to represent a firm’s technological absorptive capacity. Higher R&D investment indicates stronger adaptability to new technologies, making it easier for firms to integrate such technologies into their production processes. However, this indicator may not fully capture the broader dynamics of internal learning and structural transformation. Therefore, this paper measures technological absorptive capacity from both the input and output dimensions, using the ratio of R&D expenditure to total assets ($\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{h}\text{\_}\mathrm{i}\mathrm{n}$) as a proxy for input, and the logarithm of the total number of green invention patent applications ($\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{h}\text{\_}\mathrm{o}\mathrm{u}\mathrm{t}$) as a proxy for output. As shown in Columns (1) and (2) of

Table 6. Results of mechanism test.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	$\mathbf{T}\mathbf{e}\mathbf{c}\mathbf{h}\text{\_}\mathbf{i}\mathbf{n}$	$\mathbf{T}\mathbf{e}\mathbf{c}\mathbf{h}\text{\_}\mathbf{o}\mathbf{u}\mathbf{t}$	$\mathbf{S}\mathbf{u}\mathbf{b}\mathbf{s}$	$\mathbf{T}\mathbf{F}\mathbf{P}\text{\_}\mathbf{G}\mathbf{M}\mathbf{M}$	$\mathbf{T}\mathbf{F}\mathbf{P}\text{\_}\mathbf{O}\mathbf{P}$	$\mathbf{T}\mathbf{F}\mathbf{P}\text{\_}\mathbf{L}\mathbf{P}$
$\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{l}$	0.4394 **	1.1254 **	1.6097 ***	0.0977 ***	0.0579 **	0.0884 ***
	(2.0229)	(2.1101)	(2.8264)	(3.9859)	(2.4230)	(3.6240)
$\mathrm{l}\mathrm{n}\mathrm{S}\mathrm{i}\mathrm{z}\mathrm{e}$	0.1375	2.5119 ***	−0.4530	0.0403 *	0.0745 ***	0.3645 ***
	(1.1472)	(3.5432)	(−1.4500)	(1.9598)	(3.3326)	(17.0389)
$\mathrm{l}\mathrm{n}\mathrm{A}\mathrm{g}\mathrm{e}$	−0.0579	−0.3263	1.0551 ***	0.0638 ***	0.1181 ***	0.1053 ***
	(−0.5864)	(−1.3764)	(3.8765)	(2.7647)	(4.9297)	(4.3661)
$\mathrm{L}\mathrm{e}\mathrm{v}$	−1.9739 ***	1.9691 **	1.7111 *	0.3962 ***	0.4573 ***	0.4013 ***
	(−3.2034)	(2.5060)	(1.9470)	(5.5448)	(6.3395)	(5.5942)
$\mathrm{R}\mathrm{o}\mathrm{e}$	−3.2238 ***	2.5563	−0.9642 **	0.7181 ***	0.7155 ***	0.7191 ***
	(−8.8781)	(1.5782)	(−2.0185)	(16.7920)	(16.9182)	(16.8352)
$\mathrm{l}\mathrm{n}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}$	−0.0695	−1.5196 *	−0.8722	0.0013	0.0108	0.0109
	(−0.3051)	(−1.8478)	(−1.5859)	(0.0402)	(0.3233)	(0.3344)
$\mathrm{l}\mathrm{n}\mathrm{B}\mathrm{o}\mathrm{a}\mathrm{r}\mathrm{d}$	−0.2486	0.0853	−0.0894	0.0584 *	0.0704 **	0.0590 *
	(−0.7975)	(0.0573)	(−0.1740)	(1.7774)	(2.0675)	(1.7140)
$\mathrm{D}\mathrm{u}\mathrm{a}\mathrm{l}$	−0.0105	1.2406	0.0353	0.0070	0.0059	0.0078
	(−0.1007)	(1.5170)	(0.1583)	(0.5095)	(0.4290)	(0.5601)
$\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{g}\mathrm{d}\mathrm{p}$	0.5907	1.0656 **	0.0721	0.0431	0.0451	0.0475
	(1.4938)	(2.0982)	(0.1075)	(1.1291)	(1.1309)	(1.2279)
$\mathrm{G}\mathrm{z}$	−52.9846 ***	−90.8511	−89.8283 **	1.5174	1.6441	1.4036
	(−3.1300)	(−0.9316)	(−2.5158)	(0.7904)	(0.8399)	(0.7199)
$\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}$	0.1710	−7.6004 *	12.1581	4.4633 ***	5.3222 ***	6.7711 ***
	(0.0411)	(−1.9261)	(1.4985)	(9.2160)	(10.5318)	(13.7788)
Firm FE	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$
Time FE	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$	$\mathrm{Y}\mathrm{E}\mathrm{S}$
$\mathrm{O}\mathrm{b}\mathrm{s}$	19,501	19,501	19,501	19,430	19,430	19,430
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.8601	0.8526	0.8112	0.8272	0.8452	0.8824

, the coefficient of digitalization is positive and statistically significant at the 5% level. This indicates that digitalization effectively stimulates firms to increase R&D investment and enhances their capacity for green technological innovation, thereby contributing to carbon emission reductions.

(2) This paper examines the mediating role of the factor substitution effect ($\mathrm{S}\mathrm{u}\mathrm{b}\mathrm{s}$) in the relationship between digitalization and carbon performance. The application of digital technologies reduces firms’ reliance on production-oriented labor while increasing the demand for R&D-oriented labor. This shift significantly enhances production efficiency and reduces the input of traditional production factors, thereby positively impacting firms’ carbon performance. Accordingly, this study uses the change in the proportion of R&D personnel to represent the factor substitution indicator. The test results, shown in Column (3) of Table 6, indicate that digitalization significantly increases the proportion of R&D personnel, thereby promoting the input of technical labor and reducing the input of production labor. This suggests that corporate digitalization achieves carbon reduction by substituting traditional labor with more technical labor.

(3) We test the resource allocation effect. The resource allocation effect mainly reflects the input proportion and utilization efficiency of various factors and resources in a firm’s production. Therefore, this study employs total factor productivity (TFP), estimated using the LP, OP, and GMM methods, as an indicator of firms’ resource allocation efficiency. The stronger the efficiency of integrating various factor inputs, the higher the resource allocation efficiency, meaning higher total factor productivity. Columns (4) to (6) of Table 6 present the estimation results. The results indicate that corporate digitalization not only increases the share of digital resources in factor inputs and optimizes a firm’s factor allocation structure but also alleviates factor allocation distortions, thereby reducing corporate carbon emissions.

5.5. Moderation Effect Analysis

To assess the influence of financing constraints on the carbon reduction effect of corporate digitalization, this study employs a partially linear functional-coefficient model for empirical validation. Figure 2 presents the estimation results of the functional coefficient.

Figure 2 illustrates that the marginal influence of financing constraints on the relationship between corporate digitalization and carbon performance exhibits an “inverted U-shaped” pattern. The possible reasons are as follows: When the financing environment is relatively relaxed, firms tend to have lower resource allocation efficiency and more discretionary investment choices, suppressing the carbon reduction effect of digitalization. However, as financing constraints increase, firms have greater difficulty obtaining funds, forcing them to optimize digitalization decisions under limited financial resources. In this context, firms are more inclined to use digitalization to improve production efficiency and market competitiveness, thereby enhancing both economic and environmental benefits. When financing constraints reach a relatively high level, firms experience increased cash flow pressure and restricted capital liquidity, leading to adjustments in investment behavior. According to liquidity constraint theory, higher financing constraints often make firms more short-sighted, focusing more on short-term cash flow and profitability rather than long-term strategic goals. Under such circumstances, firms tend to prioritize economic benefits and short-term investment returns in their digitalization efforts, rather than green innovation or sustainable development. This aligns with the EKC hypothesis, which suggests that, when funds are tight, firms prioritize economic growth while neglecting environmental objectives. Therefore, under high financing constraints, firms may cut back on green digital investments, hindering green technological innovation and creating a negative cycle of “financing constraints—investment short-sightedness—obstruction of green innovation.”

6. Conclusions and Implications

6.1. Conclusions

This paper uses A-share listed manufacturing firms in China from 2008 to 2023 as the sample and constructs a two-way fixed effects model, a mediation effect model, and a partially linear functional-coefficient model to explore the impact mechanism and boundary conditions of digitalization on corporate carbon performance. Based on the empirical results, several important conclusions are drawn.

(1) The research results show that digitalization significantly improves corporate carbon performance. This indicates that the continuous accumulation of digital elements, such as data and intelligent algorithms brought about by enterprise digitalization, has generated a positive feedback effect. For enterprises, the ongoing accumulation of data resources not only enhances their technological absorptive capacity but also improves their efficiency in managing carbon emissions. The accumulation of intelligent algorithms creates synergistic effects among production equipment, optimizing logistics, warehousing, and production planning, thereby improving resource allocation efficiency and reducing energy consumption and carbon emissions. This finding is consistent with the conclusions of the existing literature [57,58]. For example, Lyu et al. [14] pointed out that digitalization in manufacturing significantly promotes the industry’s low-carbon transformation. Yang et al. [7] argued that digital transformation can facilitate corporate carbon reduction. Our study further supports this view and confirms that digitalization can enhance corporate carbon performance.

(2) The mechanism tests reveal that digitalization enhances corporate carbon performance by strengthening firms’ technological absorptive capacity, promoting the substitution of traditional production factors, and optimizing resource allocation. According to the classical environmental impact assessment model, technology, structure, and scale are the main factors influencing carbon emissions. Digitalization helps bridge the “information gap”, thereby enhancing a firm’s ability to absorb technology and promoting carbon reduction [59]. In addition, digital elements exert a factor substitution effect by replacing traditional production inputs and improving resource allocation efficiency, which in turn improves corporate carbon performance [59,60].

(3) Further analysis shows that the carbon reduction effect of digitalization is influenced by corporate financing constraints, presenting an “inverted U-shaped” relationship. This finding extends previous research [61]. When the financing environment is relatively loose, firms tend to have lower resource allocation efficiency, making the positive impact of digitalization on carbon performance less significant. As financing constraints increase, firms improve their capital utilization efficiency, allowing digitalization to gradually deliver carbon reduction benefits. However, when financing constraints become excessively tight, firms face severe liquidity pressures, leading to cuts in green digital investment, which ultimately hinders carbon reduction efforts.

6.2. Policy Implications

Based on the findings of this study, the following policy recommendations are proposed to promote corporate digitalization and low-carbon transformation.

(1) Promote corporate digital transformation through digital investment incentive policies. The government should refine green digital investment incentives—such as tax reductions, fiscal subsidies, and green credit—to guide enterprises in increasing investments in digital technologies, with particular focus on applications in intelligent manufacturing and green production. Simultaneously, efforts should be made to establish a linkage mechanism between carbon markets and digital investment, effectively linking corporate digital transformation to carbon reduction outcomes.

(2) The government should optimize the corporate financing environment to stimulate the green efficiency of digital transformation. It should encourage state-owned financial institutions to develop green finance and specialized financing instruments for the digital industry, thereby providing more diversified financing channels for enterprises undergoing digital transformation. At the same time, dynamic monitoring of the corporate financing environment should be strengthened to ensure that financing policies strike a balance between promoting digital investment and advancing low-carbon development.

(3) Enterprises should actively build a data-driven green manufacturing system. Digitalization is an inevitable trend for sustainable development, and enterprises should embrace new technologies and actively promote digital transformation. At the same time, technologies such as big data, artificial intelligence, and the Internet of Things should be applied to energy conservation and emission reduction, thereby enhancing the competitiveness of enterprises in sustainable development.

6.3. Limitations and Future Directions

Due to factors such as data availability, the limitations of this study are mainly reflected in the following aspects: Firstly, this study primarily focuses on manufacturing companies listed on China’s A-share market, which inevitably excludes non-listed companies from the sample. Future research could broaden the scope by collecting data on non-listed companies through surveys or other methods, which would help provide a more comprehensive understanding of the relationship between digitalization and corporate carbon performance. Secondly, the sample used in this study only extends up to 2023, which limits the ability to uncover more novel conclusions. Future research can enrich the findings by incorporating the latest data through updated surveys or data collection. Finally, the relationship between digitalization and carbon emissions is multifaceted. Although this study considers technological absorption, factor substitution, and resource allocation as potential mechanisms, there may be other unidentified influencing mechanisms. These additional factors warrant further in-depth exploration in future research.