Industrial Structure Upgrading and Carbon Emission Intensity: The Mediating Roles of Green Total Factor Productivity and Labor Misallocation

Abstract

Industrial structure upgrading serves as an important driving force for the sustained and healthy development of the economy, and it has a positive effect on reducing carbon emission intensity. This study uses provincial panel data from China from 2004 to 2019, starting from the dual perspectives of green total factor productivity and labor misallocation, and employs a four-stage mediation regression model to estimate the mechanism of industrial structure upgrading on carbon emission intensity. The research findings show that: for every 1% increase in industrial structure upgrading, carbon emission intensity will decrease by 0.296%; the central region shows the most significant effect, followed by the western region, while the eastern region shows no significant effect. From the view of the influencing mechanism, industrial structure upgrading will promote green total factor productivity and labor misallocation. When each of the two mediating variables increase by 1%, carbon emission intensity will decrease by 0.12% and 0.054%, respectively. Under the influence of industrial structure upgrading, the inhibitory effects of green total factor productivity and labor misallocation on carbon emission intensity have weakened, and the two factors have made it difficult to form a mediating superposition effect within the sample period. The research conclusion provides the policy implications for China to continuously adhere to industrial structure upgrading, pay attention to improving green total factor productivity, and enhance the low-carbon technical level of workers to achieve the “dual carbon” goals.

1. Introduction

Global climate change constitutes one of the most substantial challenges facing the world today. In this context, promoting the transition to a low-carbon economy has increasingly emerged as a central concern for the international community. The United Nations Framework Convention on Climate Change and the Paris Agreement both require each country to set appropriate emission reduction targets based on their own development level and technological capabilities. As a major economic power, China proposed the “dual carbon” goals to the world in September 2020. While jointly addressing global climate change, China aims to achieve its own sustainable development. In recent years, China has achieved substantial progress in transitioning to a more sustainable and low-carbon economy. The National Bureau of Statistics has provided the following information: from 2013 to 2023, China’s energy consumption intensity decreased by 26.1% cumulatively, and the carbon emission intensity decreased by more than 35% compared to 2012. Advancing a holistic green transformation of the economy and society is crucial for China’s economy to attain high-quality development in the new era and it is a crucial link in building a harmonious relationship between humans and nature.

Currently, China is undergoing a pivotal phase of industrial structural development; therefore, advancing industrial structure upgrading ($ISU$) is crucial for attaining low-carbon development. The “Action Plan for Carbon Peak by 2030” asserts the necessity to optimize the industrial structure, expedite the phasing out of obsolete production capacity, robustly advance strategic emerging sectors, and hasten the green and low-carbon transformation of conventional industries. At the end of 2024, the Central Economic Work Conference further emphasized the need to comprehensively promote carbon reduction, pollution reduction, green development and growth, accelerate the integration of economic and social advancement with the overall green transformation, and create a healthy ecosystem for the advancement of green and ecologically friendly industries. Therefore, promoting $ISU$ is not only the inevitable path for China’s economy to shift from fast growth to high-quality development, but also a necessary decision for promoting sustainable economic development in the new era. During the process of industrial transformation, China still faces many challenges. On one hand, the proposal of China’s “dual carbon” goals requires that economic growth must be deeply integrated with carbon reduction and resource utilization efficiency, and enhancing green total factor productivity ($GTFP$) is a crucial driving force for China’s green and sustainable development. On the other hand, China has long been plagued by the problem of labor misallocation ($Lamis)$. Moreover, in recent years, the aging population has accelerated, and the population dividend has declined rapidly. In conjunction with the demands of China’s ecological transformation and development, the labor misallocation (Lamis) will cause the labor force to remain in high-energy-consuming and low-output industries, unable to flow to green and efficient sectors. This will hinder the development of the green economy and thus become a key bottleneck hindering the growth of the green economy. However, current domestic and international research mostly focuses on the economic effects brought about by industrial structure upgrading, while paying relatively less attention to the specific mechanisms and paths of emission reduction. The existing research on emission reduction paths also only starts from a single path, and there is relatively little research on the simultaneous role of $GTFP$ and $Lamis$ as transmission paths. In particular, the role of labor misallocation as a key bottleneck hindering green transformation has not received sufficient attention in international research, especially in the context of $ISU$ and its impact on emissions reduction. Against the backdrop of the global search for low-carbon transformation, a deeper understanding of how $ISU$ affects $CEI$ through the dual paths of “improving green total factor productivity” and “improving labor misallocation “is not only crucial for China to achieve its “dual carbon” goals but also provides a valuable reference for other developing economies facing similar challenges. Therefore, based on the panel data of 30 provinces in China from 2004 to 2019, this study adopts a four-stage mediation regression model to analyze the impact of $ISU$ on $CEI$. Also, $GTFP$ and $Lamis$ are incorporated into the research framework to deeply explore the specific mechanism path by which $ISU$ affects $CEI$.

The marginal contributions of this study may include the following: Firstly, from the perspective of research, there are many ways to reduce carbon emissions. However, this study, based on the long-term difficult issues faced by China’s economic development, takes $ISU$ as the entry point, which has a more profound significance. At the same time, although there are many studies on industrial transformation, they focus on the purpose of the transformation varies; However, this study combines its transformation purpose with the current global focus on carbon emissions, not only making the significance of industrial transformation more significant but also providing more important inspirations for developing countries. Secondly, from the perspective of research methods, in the mediation effect model, this study not only incorporates the fact that the mediating variable becomes a post-treatment variable due to the influence of the core explanatory variable, which leads to the “bad control” problem in the model, but also extends this phenomenon in a reversed manner to enhance the attention paid to the mediating variable, thereby uncovering more economic phenomena that the mediating variable may bring about. Meanwhile, by estimating the coefficient values of the core explanatory variables on the mediating variables, and the mediating variables on the dependent variable, the indirect effect can be obtained. By subtracting the indirect effect from the estimated total effect, the direct effect can be effectively avoided, thereby effectively avoiding the problems caused by the “bad control” phenomenon. This provides important references for such research. Thirdly, from a practical perspective, on one hand, this study measures the mediating effect of $GTFP$ on $CEI$, providing a scientific basis for the formulation of industrial policies under the “dual carbon” goals and contributing to the green transformation of the Chinese economy. On the other hand, by analyzing the role of labor misallocation in the relationship between $ISU$ and $CEI$, this study provides new policy ideas for reducing the degree of the structural contradictions in China’s labor market and improving the efficiency of factor allocation. It helps to better unleash the potential for emission reduction and promote the low-carbon transformation of the Chinese economy; it also offers Chinese wisdom and solutions for global climate governance.

The organizational framework of the subsequent sections of this article is as follows: Section 2 is the literature review and theoretical hypotheses. It thoroughly examines the relevant studies conducted by international and Chinese scholars on $ISU$ and $CEI$, as well as the mechanism and path through which $ISU$ affects $CEI$, and proposes three research hypotheses. Section 3 is the research design; a benchmark regression model and a four-stage mediation model are constructed. The sample data is introduced and processed. Variables are selected and calculated. Section 4 is the analysis of empirical results. Diagnostic tests are conducted for multicollinearity, panel cross-sectional correlation, and serial correlation, a benchmark regression is performed, robustness and endogeneity tests are carried out, regional heterogeneity is analyzed, and the influence mechanism based on $GTFP$ and labor misallocation is mainly examined. Section 5 is the conclusion and policy implications. The research findings yield corresponding policy recommendations, while also highlighting the limitations of the current study and suggest future directions.

2. Literature Review and Theoretical Hypotheses

2.1. Industrial Structure Upgrading and Carbon Emission Intensity

The industrial structure illustrates the makeup of diverse industries and the interconnections and proportional relationships among them. The modification and evolution of the industrial structure signify the progression from a lower to a higher level, thereby more accurately reflecting the economic, industrial, and technological advancement and competitiveness of a nation or region [1]. Initial research on industrial structure upgrading $\left(ISU\right)$ primarily concentrated on the economic and social impacts resulting from this upgrading. Gan et al. [2] delineate the process of upgrading industrial structures into two separate tiers: the rationalization of industrial structure and the enhancement of industrial structure. Their research revealed that the rationalization and enhancement processes of industrial structure exhibit distinct phased characteristics concerning economic growth, and the rationalization of industrial structure significantly inhibits swings in economic growth. Zhou and Li [3] first analyzed the impact of industrial structure upgrading on the income gap between urban and rural areas from the national macro perspective, and then conducted in-depth analysis in the eastern, central and western regions. They reached the conclusion that different regions exhibited different phenomena.

As the economy continues to grow, it puts enormous strain on the ecological environment. Some scholars have gradually shifted their research focus to the impact of industrial structure upgrading on the environment. In prior studies, Grossman and Krugger [4] employed the Environmental Kuznets Curve (EKC) to clarify the relationship between economic development and environmental contamination. They posited a “U-shaped” relationship between the two, indicating that as economic development improves, environmental contamination initially rises and then declines. During the initial phase of economic development, growth predominantly depended on resource-intensive industries. These businesses typically exhibited traits of elevated energy use and significant emissions. The vast economic growth resulted in inefficient resource utilization and a substantial rise in carbon emissions. However, once reaching a particular level of economic development, the upgrading of industrial structure, technological advancement, and regulatory intervention collectively facilitated the mitigation of pollution. Moreover, the theory of industrial structure posits that upgrading entails a shift from low-value-added industries to high-value-added industries. Throughout this process, the share of the primary and secondary sectors, characterized by significant pollution and high energy consumption, diminishes, but the share of the tertiary industry, which exhibits comparatively lower carbon emission intensity, progressively rises. The enhancement of inter-industry practices diminishes the overall carbon emission intensity. Hu et al. [5] established the nonlinear autoregressive distributes lag and pooled mean group estimator (NARDL-PMG) model and found that, in the long run, when the industrial structure becomes more reasonable, carbon emissions will also decrease. Han and Cao [6] utilized the urban data of the urban agglomeration in the Yangtze River Delta region, improved the spatial lag model, and their research found that industrial structure adjustment is related to the local environment. Zhou et al. [7] utilized provincial panel data from China from 1995 to 2009 to explore the relationship between the structural transformation of industries and the emissions of carbon. Their research indicated that enhancing and optimizing the industrial structure can result in a reduction in carbon emissions. Cheng et al. [8] examined the influence of advancements in industrial structure on $CEI$ by utilizing a dynamic geographic panel model. The findings indicated that the enhancement and optimization of industrial structure facilitate a reduction in $CEI$. While technological advancement cannot directly diminish carbon emission intensity, it can indirectly do this through the enhancement of industrial structure. Cheng et al. [9] studied provincial-level data from China spanning 2011 to 2020 to evaluate the effects of green technical innovation and industrial structure upgrading on carbon emission intensity via the lens of spatial econometrics. The research revealed that both parameters significantly reduce $CEI$ and demonstrate a pronounced regional spillover impact. Zhang et al. [10] performed a quantitative analysis regarding the influence of industrial structure on $CEI$. The research findings demonstrated that the growth of the tertiary sector significantly contributed to the reduction in carbon emission intensity. Mi et al. [11] utilized the input-output model to assess the potential impact of industrial structure on carbon dioxide emissions. The study demonstrated that modifications in industrial structure have considerable potential for energy conservation and emission reduction. Hao et al. [12] believe that there is a correlation between urbanization, industrial structure, and environmental pollution. When the proportion of the secondary industry increases, urbanization will promote environmental pollution, whereas when the proportion of the tertiary industry increases, this promoting effect will decrease. Therefore, this paper proposes the following hypothesis:

Hypothesis 1.

Industrial structure upgrading can reduce carbon emission intensity.

2.2. Research on the Mechanism and Pathway of $ISU on CEI$

Under the dual backdrop of the “dual carbon” goals and high-quality economic development, the impact mechanism of industrial structure upgrading on carbon emission intensity has become the focus of academic attention. Existing studies mainly examine it from two aspects: First, the exploration methods of the mechanism path between the two, most studies adopt the traditional three-stage mediation test model for the test of the mediating effect [13,14], while the application of the four-stage mediation test model is relatively rare. Second, regarding the specific mechanisms, the academic community mainly focuses on perspectives such as energy structure, technological innovation, and green taxation. Fan et al. [15] utilized the Stochastic Impacts by Regression on Population, Affluence, and Technology (STIRPAT) model with energy structure optimization as a mediating variable, and discovered that upgrading the industrial structure can directly mitigate carbon emissions and also do so indirectly by fostering energy structure optimization. Sun et al. [16] established a theoretical framework to investigate the link between technological innovation, carbon emissions, and the upgrading of industrial structures. The research indicated that renovating industrial structures can substantially reduce carbon emissions by improving technical innovation levels. Chen et al. [17] examined the impact of industrial structure improvement on $CEI$ in the context of green taxes. They contended that upgrading industrial structures can markedly diminish carbon emission intensity, and that regional and temporal heterogeneity exists under the regulatory influence of green taxes. Liu and Xin [18] performed an empirical analysis utilizing the mediation effect model, contending that the industrial structure directly influences the reduction in carbon emissions and also indirectly affects this reduction through technological innovation. Among numerous studies, most of the literature adopts a single path for mechanism testing, while studies using multiple paths for testing are relatively rare. As a result, it is difficult to fully understand the key roles of more factors in the process of $ISU$ affecting $CEI.$

2.2.1. The $ISU$, $GTFP$, and $CEI$

Total factor productivity ($TFP$) is used to measure the “input-output” efficiency. The Solow residual concept, through the growth accounting model, decomposes the factors contributing to economic growth into capital, labor, and total factor productivity, thereby establishing its fundamental position in economic growth analysis. Clark [19] believes that production factors will flow from low-productivity sectors to higher ones. The “structural dividend hypothesis” posited by Timmer and Szirmai [20] asserts that the transfer of resources from lower-productivity sectors to higher-productivity sectors will augment total economic factor productivity, thereby promoting the application of $TFP$ in the research on industrial structure transformation. As environmental issues become increasingly severe, traditional $TFP$ fails to consider the limitations imposed by resources and the environment; therefore, some scholars have gradually incorporated environmental factors into total factor productivity to examine their impact on economic growth [21,22]. Yang et al. [23] found that the $ISU$ could significantly suppress carbon emissions by enhancing $GTFP$; however, it would have a negative spatial spillover effect on adjacent regions. Therefore, based on the above research, this study posits the subsequent hypothesis:

Hypothesis 2.

Industrial structure upgrading can reduce carbon emission intensity by enhancing green total factor productivity.

2.2.2. The $ISU$, $Lamis$, and $CEI$

Resource misallocation denotes a scenario in which the allocation of resources does not attain “Pareto optimality” and diverges from the ideal condition of resource distribution [24]. Enhancing resource allocation and ensuring an equitable distribution of resources across diverse industries is essential for the vigorous advancement of the economy [25]. The optimization of labor factors, a crucial resource for economic development, particularly in minimizing the misallocation of labor resources, is highly significant for economic progress. For a long time, the inherent urban-rural dual structure in Chinese society has severely hindered the mobility of rural and urban labor forces. Moreover, the uneven development of various regions and industries has also affected the mobility of labor. However, the institutional constraints in the labor market (such as restrictions on household registration) have also impacted the mobility of labor. Therefore, the phenomenon of labor misallocation in Chinese society is quite serious, which restricts economic development. This is also the important reason why this study selects labor misallocation as the mediating variable. Regarding the theories of industrial structure and $Lamis$, British economist William Petty pointed out in “Papers on Arithmetic” that the differences in relative income among different industries would prompt the labor force to flow to the higher-income sectors. Meanwhile, Clark in 1940 in “Conditions for Economic Progress” summarized the regularity of the structural changes of the labor force in the three industries and the increase in per capita national income through the collation and comparison of the data on labor input and output in the three industries in 40 countries and regions at different times. It has been noted that as economic development occurs and per capita national income rises, the labor force initially transitions from the primary sector to the secondary sector, and subsequently to the tertiary sector. As the industrial structure continues to optimize and upgrade, it drives the labor force to gradually shift from high-pollution and high-energy-consuming industries to low-carbon industries. At the same time, the demand for high-skilled labor in the low-carbon industries promotes the improvement of the degree of labor misallocation, leading to an increase in the technical efficiency of industries, and ultimately promoting the reduction in $CEI$ and facilitating the low-carbon transformation of the economy. Therefore, the following hypothesis is proposed:

Hypothesis 3.

Industrial structure upgrading can reduce carbon emission intensity by improving labor factor mismatch.

3. Research Design

3.1. Model Construction

3.1.1. Benchmark Regression Model

Based on the analysis in the previous text, this work develops an econometric model to investigate the association between $ISU$ and $CEI$:

$$\ln {CEI}_{it}={\alpha }_0+{{\beta }_0\ln ISU}_{it}+{\rho }_0{Z}_{kit}+{\mu }_i+{\phi }_t+{\epsilon }_{it}$$

(1)

Here, $i$ denotes the province, and $t$ denotes the year. ${CEI}_{it}$ represents the carbon emission intensity of region $i$ in year $t$; ${ISU}_{it}$ represents the industrial structure upgrading of region $i$ in year $t$; the control variables are represented by $Z_{kit}$; and ${\mu }_i$ and ${\phi }_t$ represent individual fixed effects and year fixed effects, respectively; ${\epsilon }_{it}$ represents the random error term.

3.1.2. Four-Stage Mediation Effect Test Model

This study utilizes the research of Niu et al. [26], and adopts the four-stage mediation effect test approach. Based on Equation (1), the following mediation model is constructed:

$$\ln M_{it}={\alpha }_1+{{\beta }_1\ln ISU}_{it}+{\rho }_1{Z}_{kit}+{\mu }_i+{\phi }_t+{\epsilon }_{it}$$

(2)

$$\ln {CEI}_{it}={\alpha }_2+{{\gamma }_0\ln M}_{it}+{\rho }_2{Z}_{kit}+{\mu }_i+{\phi }_t+{\epsilon }_{it}$$

(3)

$$\ln {CEI}_{it}={\alpha }_3+{{\beta }_2\ln ISU}_{it}+{\gamma }_1\ln M_{it}+{\rho }_3{Z}_{kit}+{\mu }_i+{\phi }_t+{\epsilon }_{it}$$

(4)

Among them, $M_{it}$ serves as the mediating variable, including ${GTFP}_{it}$ and ${Lamis}_{it}$. The definitions of the other variables are equivalent to those in Equation (1).

Jiang [27] pointed out that sometimes it might be beneficial to examine the regression of the explained variable on the mediating variable. Based on the basic factual characteristics of China’s economic development, the two mediating variables selected in this study, namely ${GTFP}_{it}$ and the concern regarding ${Lamis}_{it}$ preceding carbon emissions intensity, indicate that the causal relationship between the explained variable and the mediating variable is relatively clear. Therefore, unlike the stepwise method (Causal Steps Approach) proposed by Baron and Kenny [28], this study adds Equation (3), thereby obtaining the following conceptual framework diagram of the mediating structure:

Equation (2) reflects the influence of $ISU$ on the mediating variable $M$. Then, in Equation (4), a problem arises: the mediating variable $M$ has become somewhat like a control variable, but it is a post-treatment variable that is influenced by the core explanatory variable “$ISU$” and is a typical “bad control” [29], which will lead to the inability to obtain a consistent estimate [27]. This is also the reason, as shown in Figure 1, it is difficult to reflect the influence of $ISU$ on CEI. On the contrary, if M is taken as the object of focus (that is, the new core explanatory variable), and ISU is regarded as the new control variable, then ISU will be a good treatment variable. In this way, observing the change in the coefficient (${\gamma }_1$) of M would be an intentional attempt.

Consequently, this paper proposes the subsequent estimation strategy: First, the stepwise method (Causal Steps Approach) proposed by Baron and Kenny [28] is still used to estimate the above (1), (2), and (4) equations, but only the regression result of Equation (4) is used as additional evidence [30]. The examination of whether the absolute value of the coefficient estimate of the core explanatory variable “$ISU$” decreases after adding the mediating variable $M$ in Equation (4); if it decreases, it indicates that there may be a mediating effect, that is, the $ISU$ may have reduced the $CEI$ through $GTFP$ and improvement of labor misallocation, and this can be regarded as tentative evidence. Second, employing the method suggested by Niu et al. [26], the above Equations (1)–(3) are estimated. The total effect is obtained from the coefficient estimate of $\mathit{ln}{ISU}_{it}$ in Equation (1), the indirect effect (${\widehat{\beta }}_1{\widehat{\gamma }}_0$) is obtained by multiplying the coefficient estimates of $\mathit{ln}{ISU}_{it}$ in Equation (2) and $M$ in Equation (3), and the direct effect is obtained by the formula ${{\widehat{\beta }}_0-\widehat{\beta }}_1{\widehat{\gamma }}_0$. Consequently, the empirical method diagram illustrating the mediating effect in this paper has been derived (Figure 2):

3.2. Data Source and Processing

In 2004, due to the first national economic census, the core economic data were systematically revised to improve their accuracy, and the data classification for related aspects became more stable. Consequently, this article selects 2004 as the initial point for the sample. In 2019, the outbreak of the COVID-19 pandemic significantly affected the global economy and China’s industrial structure upgrades were also severely affected. Moreover, in 2020, China proposed the “dual carbon” target to the world, which was a powerful policy shock that might have a significant impact on the empirical findings. Meanwhile, due to the significant lack of official authoritative data after 2019, in order to ensure data consistency and estimation accuracy, this study utilizes panel data from 30 Chinese provinces, including autonomous regions and municipalities, covering the period from 2004 to 2019, to ensure data consistency in empirical research. This dataset omits Tibet, Hong Kong, Macao, and Taiwan. Additionally, all price-based indices were adjusted to the base year of 2000. When collecting the “Urbanization Level” data, the statistical yearbook data for the permanent urban residents in Anhui, Guangdong, and Sichuan in 2004 were missing; therefore, based on the assumption of time series continuity, the linear interpolation method was employed to populate the data by utilizing the data from adjacent years (2003 and 2005). To reduce the impact of heteroscedasticity and to more clearly quantify the degree of response of CEI to the changes in ISU, while also addressing the issue of some control variables being excessively large, this study adheres to the methodologies of Liu et al. [31] and Zhou et al. [14], and takes the logarithms of the main variables. The information in this study primarily derives from the official website of the National Bureau of Statistics of China, the statistical yearbooks of various provinces and regions, the “China Industrial Economic Statistical Yearbook,” the “China Energy Statistical Yearbook,” the “China Environmental Statistical Yearbook,” and Zhongjing Network.

3.3. Selection and Calculation of Variables

3.3.1. Selection and Calculation of Carbon Emission Intensity ($CEI$)

$CEI$ is an important indicator for assessing whether a country or region can achieve low-carbon development. The computation of carbon emissions adheres to the methodologies employed by Du [32] and Zhu et al. [33]. Following the methods of IPCC [34] and the National Climate Change Response Coordination Group Office and the National Development and Reform Commission Energy Research Institute [35], seven fossil energy sources—coal, coke, gasoline, kerosene, diesel, fuel oil, and natural gas—along with cement manufacturing data, were used to compute the estimated carbon dioxide emissions resulting from fossil fuel burning and emissions generated during the cement production process. The precise formula for the computation is as presented below:

$${CO}_2={\sum }_{n=1}^7{EC}_n\times {CF}_n\times {CCE}_n\times {EOF}_n\times {\displaystyle \frac{44}{12}}+K_0\times Q$$

(5)

Among them, ${EC}_n$ denotes the energy consumption volume; ${CF}_n$ denotes the calorific value of the respective energy; ${CCE}_n$ signifies the carbon content of the energy; ${EOF}_n$ indicates the oxidation factor of the energy; 44/12 denotes the ratio of the relative molecular weights of ${CO}_2$ and $C$; $Q$ denotes the volume of cement output; and $K_0$ signifies the carbon emission coefficient associated with cement manufacturing. The carbon dioxide emission coefficients (${CF}_n\times {CCE}_n\times {EOF}_n\times {\displaystyle \frac{44}{12}}$) of seven fossil fuels such as coke, coal, kerosene, diesel, gasoline, fuel oil, and natural gas, and the carbon dioxide emission coefficient ($K_0$) during the cement production process are 2.8481, 1.6470, 3.1742, 3.0451, 3.0642, 21.6704, and, 0.5271, respectively.

Additional computation of carbon emission intensity:

$${CEI}_{it}={\displaystyle \frac{({{CO}_2)}_{it}}{{GDP}_{it}}}$$

(6)

Among them, ${CO}_2$ denotes the volume of carbon dioxide emissions in region $i$ during year $t$; ${GDP}_{it}$ denotes the aggregate $GDP$ in region $i$ during year $t$; the greater the value of carbon emission intensity (${CEI}_{it}$), the more carbon dioxide emissions are generated per unit of economic output.

In the above calculation process, this study adheres to the classification methodology of economic regions by the National Bureau of Statistics of China, the research sample is divided into three major regions: the east, the central region, and the west. The east region includes 11 provinces such as Beijing, Tianjin, Hebei, Liaoning, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong, and Hainan; the central region includes 8 provinces such as Shanxi, Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hubei, and Hunan; the western region includes 11 provinces such as Inner Mongolia, Guangxi, Chongqing, Sichuan, Guizhou, Yunnan, Shaanxi, Gansu, Qinghai, Ningxia, and Xinjiang. It explores the changing trends of $CEI$ in these three regions within the sample range, to present the changing trend of $CEI$ more intuitively, as illustrated in Figure 3a,b. By comparing (a) and (b) in Figure 3, it is found that from 2004 to 2019, the $CEI$ of most provinces in the eastern region remained at a relatively low level; however, the absolute value of $CEI$ in Hebei was higher than that of other provinces in the eastern region, and there was no sign of relief. The $CEI$ in Liaoning also gradually increased. These two provinces were held back by traditional high-energy-consuming industries such as steel, facing great pressure for emission reduction and having a heavy reliance on heavy industries. The task of green transformation of their industries is arduous. Nevertheless, the $CEI$ of Hebei and Liaoning was in the third level. The $CEI$ in most provinces in the central region had a relatively stable changing trend, but the $CEI$ in Shanxi Province was significantly higher than that in other provinces. The main reason is that Shanxi is dominated by coal resource-based industries, and its energy production and consumption are highly carbonized, making it difficult to reduce emissions. Heilongjiang’s $CEI$ rose from the second level to the third level. The $CEI$ in the western region shows a significant overall differentiation. Owing to the substantial dependence on fossil fuels, including coal, and energy-intensive sectors such as power generation and steel production, the $CEI$ in provinces such as Inner Mongolia, Ningxia, Xinjiang, Qinghai and Shaanxi have been on the rise. Inner Mongolia and Ningxia have both moved up from the fourth to the fifth level, with Inner Mongolia having the highest $CEI$ in China and Ningxia following closely behind. Xinjiang has jumped from the second to the fourth level, while Qinghai and Shaanxi have moved up from the second to the third level. Gansu has remained in the third level. Guizhou’s $CEI$ decreased (from the fourth to the third level). Therefore, on the whole, the $CEI$ in the western region was the highest, followed by the central region, and the eastern region had the lowest. The $CEI$ of provinces relying on high-pollution and high-energy-consuming industries shows an upward trend. This is essentially the result of the combined effect of industrial structure and energy path, which also lays the foundation for further research on the relationship between $ISU$ and $CEI$.

3.3.2. Industrial Structure Upgrading ($ISU$)

$ISU$ denotes to the process or trend where industries gradually transform lower-tier configuration to a higher-tier configuration. Therefore, this study adopts the methodology of Huang et al. [1] to develop an index for industrial structure upgrading, aimed at assessing the modification of the industrial structure.

$${ISU}_{it}=\sum _{j=1}^j{S}_{ijt}\times P_{ijt}$$

(7)

In this context, $i$, $j$, and $t$ denote region, industry, and time, respectively. $S_{ijt}$ represents an industrial productivity indicator, measured by the proportion of the added value of industry j in region i at time t to the total added value of all industries; $P_{ijt}$ is labor productivity, measured by the added value of industry j in region i at time t divided by the number of employed people. An elevated index of industrial structure upgrading is associated with an enhanced level of industrial structure upgrading.

3.3.3. Mediating Variable

• Green Total Factor Productivity ($GTFP$)

This study utilizes the super-efficiency SBM model introduced by Andersen [36] for a solution. In comparison to the general radial DEA model, the super-efficiency SBM model introduces non-desired outputs, which can comprehensively consider the influences of inputs, desired outputs, and non-desired outputs. Moreover, by introducing slack variables, it is more conducive to reducing the deviations caused by radial and angular measurements. Based on this, by further drawing on the Malmquist index proposed by Malmquist [37] and the Malmquist–Luenberger index (referred to as ML index) proposed by Chambers et al. [38], the super-efficiency SBM-GML index for non-desired outputs can be calculated. This can yield a more thorough evaluation and comparison of efficiency.

The input-output variables and data processing are outlined as follows: (1) Selection of input variables; following the approaches of Long et al. [39] and Tian and Feng [40], capital stock, labor, and energy consumption are chosen as the input variables. Regarding the estimation of capital stock, the prevalent methodology in the academic community today is the perpetual inventory method (PIM), introduced by Goldsmith in 1951. The total stock of fixed assets in the entire society can reflect the scale of material capital accumulation in different regions and directly affect the economic output capacity. Therefore, this study adopts the methodology of Zhang et al. [41] and employs the total stock of fixed assets as a measure of capital investment. The calculation of the depreciation rate of fixed assets is 9.6% [14,42]. Regarding labor input, Chen and Golley [43] hold that labor is the fundamental input variable in the production function; therefore, the annual employment figures for each region serve as the metric for labor input assessment. Energy consumption is the main source of carbon emissions and industrial pollution. Ignoring energy input would lead to an overestimation of production efficiency; consequently, this study follows the method of Chen et al. [44] and uses the total energy consumption as the measurement basis for energy input. (2) Selection of the expected output variable. As the core output indicator of the national economic accounting system, Gross Domestic Product ($GDP$) is currently the most commonly used measure of expected output in the academic community [45]. So, this study also follows this practice and uses the real $GDP$ to measure the expected output. (3) Selection of non-desired output variables. Both industrial wastewater discharge and industrial sulfur dioxide emissions are mandatory emission reduction targets stipulated in China’s “11th Five-Year Plan” to “14th Five-Year Plan”. So, this study follows the approach of Zhou et al. [14] and uses industrial wastewater discharge and industrial sulfur dioxide emissions to measure non-desired outputs.

Table 1. Green total factor productivity index framework.

Category	Primary Indicators	Secondary Indicators
Input	Capital Input	The total stock of fixed assets in the entire society
	Labor Input	Annual employment
	Energy Input	Total energy consumption
Output	Desirable Output	Real GDP
	Undesirable Output	Industrial sulfur dioxide emissions
		Industrial wastewater discharge volume

specifies the particulars of each indication.

2.

Labor Misallocation ($Lamis$)

This research utilizes the resource misallocation theoretical framework developed by Hsieh and Klenow [46] and incorporates the methodology of Bai and Liu [24]. The computation formula is as follows:

$${Lamis}_i={\displaystyle \frac{1}{{\gamma }_{Li}}}-1$$

(8)

A higher ${Lamis}_i$ value signifies a greater degree of labor misallocation and a less rational allocation of labor factors; conversely, a lower value implies reduced misallocation and optimized allocation of labor elements. In this equation, ${\gamma }_{Li}$ denotes the distortion coefficient of labor factor pricing, and the computation formula is as follows:

$${\gamma }_{Li}=\left({\displaystyle \frac{L_i}{L}}\right)/\left({\displaystyle \frac{s_i{\beta }_{L_i}}{{\beta }_L}}\right)$$

(9)

Among them, ${\displaystyle \frac{L_i}{L}}$ denotes the ratio of the employment population in region $i$ to the total employment population; $s_i$ signifies the ratio of the output in region $i$ to the total output; and ${\displaystyle \frac{s_i{\beta }_{L_i}}{{\beta }_L}}$ indicates the ratio of labor factors employed in region $i$ when allocated efficiently. Based on the idea of the Solow residual, this study posits that the Cobb–Douglas production function exhibits constant returns to scale concerning labor output elasticity, ${\beta }_{L_i}$. This argument is supported by the material presented here.

The model is configured as follows:

$$Y_{it}=A{K}_{it}^{\beta K_{it}}{L}_{it}^{1-\beta K_{it}}$$

(10)

In this context, $Y_{it}$ denotes the real $GDP$ of each region, whereas $K_{it}$ signifies the total fixed asset stock of the entire society within each region. By drawing on the method of Bai and Liu [24], who took the natural logarithm of both sides of Equation (10) and then rearranged it, Equation (11) was obtained. Moreover, the provincial panel data utilized in this study reveals disparities in economic development across provinces, leading to differing capital and labor output elasticities in each region. Therefore, when using the least squares dummy variable method (LSDV) to regress Equation (11), individual dummy variables and the interaction term between the dummy variables and ($\mathit{ln}{K}_{it}/L_{it}$) are introduced for variable coefficient regression, which makes the estimation of output elasticity more accurate.

$$\mathit{ln}({\displaystyle \frac{Y_{it}}{L_{it}}})=\mathit{ln}A+{\beta }_{K_{it}}\mathit{ln}({\displaystyle \frac{K_{it}}{L_{it}}})+{\epsilon }_{it}$$

(11)

From the above process, the capital-output elasticity (${\beta }_{K_{it}}$) can be calculated. Predicated on the premise of constant returns to scale [46], the labor-output elasticity (${\beta }_{L_i}$) can be determined. The assumption of constant returns to scale can stabilize the separation of the marginal contributions of capital and labor and can avoid the distortion of the misallocation degree estimation caused by the deviation in elasticity setting. The $Lamis$ for 30 provinces in China from 2004 to 2019 can be computed based on the aforementioned calculation process. Labor misallocation refers to the situation where the allocation of labor factors is unreasonable. If the $Lamis$ exceeds 0, it signifies an inadequate allocation of labor factors; if it is less than 0, it indicates that the allocation of labor factors is excessive. To guarantee the coherence of the economic relevance of the coefficients in the ensuing regression analysis, following the approaches of Bai and Liu [24] and Cui et al. [47], the labor misallocation is processed by taking its absolute value.

3.3.4. Control Variables

This study employs the subsequent control variables derived from relevant studies: (1) Energy consumption structure. The fraction of coal consumption within total energy consumption significantly affects carbon emissions. Enhancing the energy consumption framework can facilitate carbon reduction [48]. (2) Urbanization level. It is quantified by the ratio of permanent urban inhabitants to the total permanent residents. Zhang et al. [49] utilized the STIRPAT theoretical framework and cross-national panel data to demonstrate that urbanization levels significantly influence carbon emissions. (3) Degree of openness. It is quantified by the total amount of imports and exports from domestic destinations and supply sources as a percentage of the regional gross domestic product ($GDP$). Ding et al. [50] discovered that a suitable degree of external engagement can diminish carbon emissions. (4) Transportation level. Assessed by the yearly freight volume of each region. Schipper et al. [51] examined the mechanism behind the link between transportation activities and carbon emissions over a long-term perspective, concluding that transportation activities constitute a significant source of carbon emissions in the United States. (5) Human Capital. Research indicates that individuals significantly influence carbon emissions. The impact of human capital on carbon emissions initially rises and subsequently declines, exhibiting a “U-shaped” trajectory [52]. Consequently, this paper quantifies human capital by the ratio of students enrolled in conventional higher education institutions relative to the total population. A greater proportion indicates an elevated level of human capital within society. (6) Proportion of the primary industry and proportion of the secondary industry. Han et al. [53] assert that the economic structure is the primary determinant influencing the environment. Consequently, this study employs the proportions of the primary and secondary industries as control variables for analysis.

Table 2. Descriptive statistics of each variable.

Index	Meaning	N	Mean	Std.	Min	Max
$\ln CEI$	Carbon Emission Intensity	480	1.998	0.622	0.335	3.836
$\ln ISU$	Industrial Structure Upgrading	480	10.280	0.333	9.598	11.33
$\ln GTFP$	Green Total Factor Productivity	480	−0.465	0.396	−1.584	0.511
$\ln Lamis$	Labor Misallocation	480	−1.460	1.165	−6.064	1.214
lnUrban	Urbanization Level	480	−0.657	0.267	−1.974	−0.110
Energy	Energy Consumption Structure	480	0.000	1.000	−2.319	3.787
Opening	Degree of Openness	480	0.311	0.347	0.012	1.664
Prind	The Proportion of Primary Industry	480	0.111	0.060	0.003	0.339
Seind	The Proportion of Secondary Industry	480	0.431	0.082	0.160	0.620
Edu	Human Capital	480	0.0175	0.00631	0.00461	0.0389
$\ln trans$	Transportation Level	480	11.330	0.872	8.734	12.980

displays the descriptive statistics for each variable.

4. Empirical Results Analysis

4.1. Multiple Collinear Analysis

To calculate whether the independent variables exhibit a high degree of correlation, we conducted a multicollinearity test on the established model.

Table 3. Test for multiple collinearity.

Variable	VIF	Variable	VIF
$\ln ISU$	2.87	Opening	2.70
$\ln GTFP$	1.48	Prind	3.57
$\ln Lamis$	1.95	Seind	2.14
lnUrban	4.34	Edu	2.02
Energy	1.82	lntrans	1.57

illustrates the outcomes of the examination. The variance inflation factor (VIF) of each variable is less than five, as evidenced by the multicollinearity test results. Overall, the model has successfully passed the multicollinearity test and is less influenced by multicollinearity, resulting in a high level of credibility for the results.

4.2. Panel Diagnosis

Before constructing the model, this study conducted cross-sectional correlation and serial correlation tests to ensure the validity of the estimation results.

Table 4. Panel diagnostic results.

Type of Test	Variable	Title 2	Title 3
Pasaran CD-Test	lnCEI	43.95	0.000 ***
	lnISU	12.31	0.000 ***
	lnGTFP	65.69	0.000 ***
	lnLamis	22.91	0.000 ***
	Urban	76.67	0.000 ***
	Energy	10.56	0.000 ***
	Opening	29.70	0.000 ***
	Prind	64.89	0.000 ***
	Seind	49.25	0.000 ***
	Edu	46.80	0.000 ***
	lntrans	58.01	0.000 ***
Wooldridge-Test		184.897	0.000 ***

The symbols *** indicate significance at the 1% confidence levels.

illustrates the findings. As evidenced by Table 4, all the results rejected the null hypothesis of “no correlation”, indicating the existence of cross-sectional correlation and serial correlation. In response to this situation, this paper adopts the Driscoll–Kraay (DK) [54] standard error for correction. Under this framework, in a finite sample with a sufficiently large time dimension (T), regardless of the size of N (even if N is larger than T), the estimation is still valid; in the Monte Carlo simulation, when T is 10 and N can be any value, the coverage rate of the DK standard error still reaches approximately 80%, and the obtained results remain reliable. The value of T selected for the sample in this article is 16, which is greater than 10. This indicates that, regardless of the value of N, the estimates obtained using the DK standard error are reliable. The subsequent regression analysis will adopt this standard error.

4.3. Benchmark Regression Results

The findings presented in

Table 5. Benchmark regression results.

	(1) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$	(2) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$	(3) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$	(4) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$
$\mathit{ln}ISU$	−0.547 ***	−0.461 ***	−0.190 ***	−0.296 ***
	(0.117)	(0.146)	(0.045)	(0.041)
Prind		−0.456		−0.503
		(0.524)		(0.417)
Seind		−0.246		0.400
		(0.384)		(0.339)
Edu		−13.707 ***		−0.567
		(4.516)		(3.932)
Energy		0.517 ***		0.308 ***
		(0.014)		(0.011)
lnUrban		0.441 ***		0.050
		(0.121)		(0.084)
$\mathit{ln}trans$		−0.115 ***		0.101 ***
		(0.033)		(0.027)
Constant	7.618 ***	8.727 ***	3.636 ***	3.589 ***
	(1.253)	(1.786)	(0.461)	(0.223)
Individual Fixed effect	NO	NO	YES	YES
Year Fixed effect	NO	NO	YES	YES
N	480.000	480.000	480.000	480.000
$\mathrm{Adjusted}\text{\_}{R}^2$	0.086	0.683	0.4942	0.7865

The symbols *** indicate significance at the 1% confidence levels.

are derived using the benchmark regression model (Equation (1)). The results of Table 5 are presented in columns (1) and (2) without the inclusion of fixed effects, while columns (3) and (4) display the results after the integration of individual and year fixed effects. The regression results show that the regression coefficient of industrial structure upgrading ($ISU$) is significantly negative in all models; from column (4), it can be seen that for every 1% increase in $ISU$, the $CEI$ will decrease by 0.296%. This indicates that the $ISU$ facilitates the transition of regionally high-energy-consuming and high-pollution industries into low-energy-consuming and high-value-added industries. This transition can markedly diminish regional $CEI$ and foster regional sustainable development. Comparing the regression findings with and without the adjustment for individual and year fixed effects revealed a significant decrease in the regression coefficient of $ISU$ after accounting for the fixed factors. This suggests that neglecting to compensate for fixed variables will lead to an overestimation of the influence of $ISU$ on $CEI$. The results demonstrate that $ISU$ is a significant factor in decreasing carbon emission intensity; consequently, hypothesis 1 is confirmed. Consequently, by advocating for $ISU$, we can diminish our dependence on energy-intensive and polluting industries, substantially lowering carbon emission intensity, and establish a vital pathway for attaining low-carbon objectives.

4.4. Robustness Test

This paper conducts a series of robustness experiments to evaluate the impact of $ISU$ on $CEI$, as demonstrated in

Table 6. Robustness test results.

	Variable Substitution	Winsorization	Model Transformation
	(1) $\mathit{ln}\mathit{C}$	(2) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$	(3) Fixed Effect $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$	(4) Random Effect $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$	(5) Mixed OLS $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$
$\mathit{ln}ISU$	−0.192 ***	−0.302 ***	−0.296 ***	−0.266 **	−0.461 ***
	(0.052)	(0.046)	(0.041)	(0.047)	(0.082)
Control Variables	Control	Control	Control	Control	Control
Constant	9.837 ***	3.707 ***	3.589 ***	3.006 ***	8.727 ***
	(0.497)	(0.208)	(0.223)	(0.548)	(1.093)
Individual Fixed effect	Fixed	Fixed	Fixed
Year Fixed effect	Fixed	Fixed	Fixed
N	480.000	480.000	480.000	480.000	480.000
$\mathrm{Adjusted}\text{\_}{R}^2$	0.9121	0.789	0.787	0.730	0.683

The symbols *** and ** indicate significance at the 1% and 5% confidence levels, respectively.

.

Firstly, substitute the specified variable. Utilize carbon emission volume ($\mathit{ln}C$) in lieu of carbon emission intensity ($\mathit{ln}CEI$) to analyze the absolute quantity of carbon emissions. Column (1) of Table 6 demonstrates that, following the substitution of the explained variable, the regression coefficient for industrial structure upgrading ($ISU$) persistently remains considerably negative at the 1% significance level. This signifies that, whether assessed from an absolute or relative quantity standpoint, the detrimental impact of the $ISU$ on $CEI$ remains robust, thereby enhancing the reliability of the benchmark regression model results.

Secondly, the variables undergo truncation processing. In panel data analysis, outliers can significantly influence regression outcomes, leading to distortions in parameter estimates. Consequently, in this study, the variables are truncated at 1% and above 99%. Column (2) of Table 6 indicates that, following the tail-cutting operation, a substantial negative connection persists between $ISU$ and $CEI$. This further illustrates the model’s robustness against extreme values, suggesting that the research conclusions are less impacted by individual outliers, hence affirming the trustworthiness and accuracy of the regression results.

Thirdly, transformation of models. This paper employs fixed effect, random effect, and mixed OLS models to rigorously evaluate the influence of $ISU$ on $CEI$. Columns (3) to (5) of Table 6 indicate that the regression findings remained largely unchanged across different models. The inverse link between the two remained, indicating that the claim that $ISU$ reduces $CEI$ is significantly robust. Moreover, the Hausman test rejected the null hypothesis of random effects, corroborating the appropriateness of employing the fixed effects model in the benchmark regression analysis.

4.5. Endogeneity Test

This study employs the proportion of heavy industry in each province in 1985 and the number of students enrolled in regular universities in 1985 as the instrumental variable to reduce the effects of endogeneity issues. The reasons for selection are as follows: Firstly, for the instrumental variable “the proportion of heavy industry in each province in 1985”, heavy industry often has the characteristic of path dependence and can have an impact on the industrial structure; for the instrumental variable “the number of students enrolled in ordinary universities in 1985”, it can reflect the reserve of human capital in the early stage of reform and opening up, and can provide human resources for the subsequent development of industries, so both of them meet the requirement of correlation. Secondly, the foundation of the two variables is 19 years away from the sample period of this study. The time interval is relatively long, and future carbon emission policies cannot be predicted. The direct impact of the $CEI$ in the sample period is relatively small, which meets the requirement of exogeneity, considering that the two variables were selected as cross-sectional variables and could not directly participate in the panel regression. Therefore, adhering to the methodology proposed by Goldsmith et al. [55] we multiply “the proportion of heavy industry in each province in 1985” and “the number of students enrolled in regular universities in 1985” with the time variable to construct the instrumental variables, which are represented by $iv1\_trend$ and $iv2\_trend$, respectively. The test outcomes are presented in

Table 7. Results of endogeneity test.

	A Two-Stage Least Squares Regression
	(1) First	(2) Second
	$\mathit{ln}\mathit{I}\mathit{S}\mathit{U}$	$\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$
$iv1\_trend$	0.044 ***
	(3.68)
$iv2\_trend$	0.003 ***
	(8.66)
$\mathit{ln}ISU$		−0.873 ***
		(−7.29)
Control Variables	Control	Control
Observations	480	480
Individual Fixed Effect	YES	YES
Year Fixed Effect	YES	YES
Kleibergen–Paap rk LM		75.681 p-value: 0.000
Kleibergen–Paap rk Wald F		53.156 Critical value: 19.930
Hansen J		1.042 p-value: 0.307

The symbols *** indicate significance at the 1% confidence levels.

.

Column (1) of Table 7 reveals that in the initial stage, the regression coefficients of the two instrumental variable ($iv1\_trend$ and $iv2\_trend$) are significantly positive at the 1% level, indicating a correlation between the selected instrumental variable ($iv1\_trend$ and $iv2\_trend$) and the original explanatory variable ($ISU$). Moreover, as evidenced in column (2), the second stage reveals a substantial negative association between $\mathit{ln}ISU$ and $\mathit{ln}CEI$, suggesting that $ISU$ can markedly reduce $CEI$. Additionally, the p-value of the Kleibergen–Paap rk LM test is 0.000, thereby rejecting the null hypothesis of inadequate identification of the instrumental variables. In the Kleibergen–Paap rk Wald F test, the statistic was 53.156, exceeding the critical value of 19.930, which signifies the absence of weak instrumental variable issues. Meanwhile, the p-value of the Hansen J test was 0.307, which did not reject the null hypothesis that “all the instrumental variables are exogenous”, and thus passed the overidentification test. In conclusion, the endogeneity test demonstrates that $ISU$ has a consistent and dependable impact on reducing $CEI$.

4.6. Analysis of Regional Heterogeneity

To conduct a more thorough examination of the regional heterogeneity impact of $ISU$ on $CEI$, this study categorizes the 30 provinces in the sample into three regions—the eastern, central and western regions for regional heterogeneity analysis.

Table 8. Results of the regional heterogeneity test.

	(1)	(2)	(3)
	East	Central	West
$\mathit{ln}ISU$	−0.1000	−0.361 ***	−0.192 **
	(−1.58)	(−4.10)	(−2.54)
Constant	2.782 **	4.043 ***	2.374 **
	(2.71)	(4.10)	(2.89)
Control Variables	Control	Control	Control
Observations	450	450	450
Individual Fixed Effect	YES	YES	YES
Year Fixed Effect	YES	YES	YES
N	176	128	176
$\mathrm{Adjusted}\text{\_}{R}^2$	0.9124	0.8126	0.8973

The symbols *** and ** indicate significance at the 1% and 5% confidence levels, respectively.

illustrates the regression findings. The impact of $ISU$ in the eastern region on $CEI$ is not significant, as evidenced by column (1) of Table 8. The following are potential explanations: Firstly, from the historical viewpoint of China’s reforms, the eastern region has long been the region with the fastest development speed and has entered the post-industrial stage. Its industrial structure has undergone multiple adjustments and upgrades, and the economy dominated by the tertiary industry has led to the marginal effect of emission reduction from industrial structure upgrading approaching saturation. Secondly, from the viewpoint of energy consumption, the eastern region belongs to the areas with relatively scarce resources in China. The demand for energy mainly comes from the west. This has resulted in better institutional arrangements in the eastern region, emphasizing the efficiency of energy use and thus, the inhibitory effect of $ISU$ on $CEI$ is not significant.

It is evident from column (2) of Table 8 that the impact of the $ISU$ in the central region on $CEI$ is significantly negative at the 1% level, and the absolute value of the coefficient is as high as 0.361, indicating the strongest emission reduction effect. First of all, for a long time, the central region has been the area with the second-fastest development speed in China, and it is also an important traditional industrial base in China. Furthermore, the proportion of traditional energy and chemical industries, which are high-energy-consuming and high-carbon-emission industries, is relatively large. Therefore, as the industrial structure of the central region continues to optimize and upgrade, “high-carbon” industries gradually transform and upgrade to technology-intensive and other low-carbon industries, and the effect of reducing $CEI$ is relatively obvious. Furthermore, the execution of the Central Region Rise Strategy has offered policy assistance for industrial upgrading in the central region, promoting continuous transformation and upgrading of the industrial structure and its improvement, gradually reversing the traditional extensive growth model, enhancing energy efficiency, reducing relative carbon emissions per unit, and thereby curbing the carbon emission intensity.

From column (3), it is evident that the impact of the $ISU$ in the western region on $CEI$ is significantly negative at the 5% level, indicating that the $ISU$ in the western region has a certain inhibitory effect on $CEI$; however, it is relatively smaller compared to the negative effect of the $ISU$ in the central region on $CEI$. The reasons might be as follows: Firstly, the western region is the area with the richest resources in China. It can leverage its energy advantages to develop clean energy industries and promote the reduction in $CEI$; however, the western region has long been the area with the most extensive economic growth and still suffers from “resource curse” and “resource lock-in” phenomena. The industries there are difficult to be largely replaced in the short term. Secondly, the western region has relatively insufficient innovation capabilities, relatively slow technological development, and lower resource allocation efficiency. This has led to a significantly slower process of industrial structure upgrading compared to the central region, resulting in a relatively smaller impact of $ISU$ on $CEI$ compared to the central region.

In conclusion, this regional disparity essentially reflects the differences in economic development stages, industrial foundations, and resource endowments among different regions in China. The impact of $ISU$ in the eastern region on $CEI$ is not significant. This is not because industrial structure upgrading has failed to affect $CEI$, but because the eastern region has undergone multiple adjustments in its industrial structure, resulting in saturation of its emission reduction effects. This also highlights the need for the eastern region to explore more refined carbon emission reduction paths, such as by promoting green and low-carbon high-tech, green and low-carbon lifestyles and consumption patterns, to further enhance the emission reduction effect. For the central and western regions, the $ISU$ has a significant impact on the $CEI$, precisely demonstrating the crucial role of $ISU$ in reducing $CEI$. Principally, the impact of the $ISU$ in the western region on the $CEI$ is smaller than that in the central region. This also indicates that if the western region wants to enhance the effect of $ISU$ on the $CEI$, it needs to increase policy support, improve regional innovation capabilities, promote technological progress, further break through the “resource curse” phenomenon, and thereby release the emission reduction potential of industrial structure upgrading.

4.7. Mechanism Verification

The prior literature indicates that $ISU$ exerts a suppressive influence on $CEI$. This research examines the internal mechanisms and pathways by which $ISU$ influences $CEI$, utilizing $GTFP$ and $Lamis$ as mediating variables for mechanism testing.

4.7.1. Mechanism Verification Based on GTFP

Column (1) in

Table 9. Results of the mechanism test based on $GTFP$.

	(1) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$	(2) $\mathit{ln}\mathit{G}\mathit{T}\mathit{F}\mathit{P}$	(3) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$	(4) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$
$\mathit{ln}ISU$	−0.300 ***	0.322 ***		−0.275 ***
	(0.042)	(0.064)		(0.041)
$\mathit{ln}GTFP$			−0.120 ***	−0.080 ***
			(0.027)	(0.021)
Control Variables	Control	Control	Control	Control
Constant	4.581 ***	−0.119	1.723 ***	4.577 ***
	(0.247)	(1.602)	(0.299)	(0.267)
Individual Fixed Effect	YES	YES	YES	YES
Year Fixed Effect	YES	YES	YES	YES
N	480.000	480.000	480.000	480.000
$\mathrm{Adjusted}\text{\_}{R}^2$	0.7787	0.6647	0.7632	0.7822

The symbols *** indicate significance at the 1% confidence levels.

represents the benchmark regression result, and there are slight differences in the coefficients compared to the result in column (4) of Table 5. The reasons are as follows: from China’s experience of over 50 years of reform and openness, considering degree of external openness ($Opening$) can significantly influence green total factor productivity (GTFP) by introducing advanced technologies, etc., while the Transportation Level ($lntrans$) has a smaller impact on $GTFP$. Therefore, here “$Opening$” has been added and “$lntrans$” has been removed. Column (2) indicates that $Industry$ positively influences $GTFP$ at the 1% significant level, and for every 1% increase in $ISU$, it will result in a 0.322% increase in $GTFP$. This signifies that during the process of industrial structure upgrading, through means such as technological progress, scale effects, and specialized division of labor, the allocation of resources is optimized, facilitating the transition of production factors to green and low-carbon, high-value-added industries, and thereby enhancing $GTFP$. Column (3) shows that at the 1% significance level, the $GTFP$ has a negative effect on carbon emission intensity (CEI), and when $GTFP$ increases by 1%, it will cause $CEI$ to decrease by 0.12%. This indicates that as $GTFP$ increases, more output is obtained from the production process with less resource input and a reduction in environmental costs, thereby further reducing carbon emission intensity. Column (4) shows the combined impact of $ISU$ and $GTFP$ on $CEI$. The results indicate that both factors exert a considerable adverse impact on CEI. By comparing column (4) with column (1), it is found that after the inclusion of the mediating variable $GTFP$, the absolute value of the coefficient estimate for $ISU$ has decreased from the original 0.3 to 0.275 (a reduction of 8.33%), indicating that $GTFP$ may be a channel through which the $ISU$ reduces $CEI$. By comparing column (3) and column (4), it is found that regardless of whether $ISU$ is considered, $GTFP$ significantly reduces $CEI$. However, in the context of the $ISU$, the negative effect of $GTFP$ on $CEI$ has weakened to some extent; this indicates that during the sample period, China’s industrial structure upgrading has not yet achieved the goal of strengthening the negative effect of $GTFP$ on $CEI$, reflecting that there remains potential for additional enhancement in the upgrading of China’s industrial structure.

In Table 9, the overall effect of $ISU$ in reducing $CEI$ is 0.3. Among them, the indirect effect of $GTFP$ is only 0.0386 (0.322 in column (2) multiplied by 0.12 in column (3)), accounting for 12.88%, while the direct effect of $ISU$ is as high as 0.2614, accounting for 87.12%. This suggests that $GTFP$ partially mediates the relationship between $ISU$ and $CEI$. It suggests that the $ISU$ is not only directly reduce $CEI$ but also indirectly achieves this through $GTFP$, so substantiating hypothesis 2. Consequently, $GTFP$ serves a vital mediating and transmitting function in the correlation between $ISU$ and $CEI$. By augmenting $GTFP$, $CEI$ can be significantly diminished, representing an effective strategy for advancing the transition and development of a low-carbon economy in various regions. This conclusion offers valuable references for the government in developing green policy.

4.7.2. Mechanism Verification Based on the $Lamis$

The coefficients in column (1) of

Table 10. Results of mechanism testing based on labor misallocation.

	(1) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$	(2) $\mathit{ln}\mathit{L}\mathit{a}\mathit{m}\mathit{i}\mathit{s}$	(3) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$	(4) $\mathit{ln}\mathit{C}\mathit{E}\mathit{I}$
$\mathit{ln}ISU$	−0.317 ***	0.537 ***		−0.292 ***
	(0.073)	(0.145)		(0.071)
$\mathit{ln}GTFP$			−0.054 ***	−0.047 ***
			(0.009)	(0.009)
Control Variables	Control	Control	Control	Control
Constant	2.371 ***	−2.651	−0.613	2.247 ***
	(0.651)	(1.594)	(0.737)	(0.636)
Individual Fixed Effect	YES	YES	YES	YES
Year Fixed Effect	YES	YES	YES	YES
N	480.000	480.000	480.000	480.000
$\mathrm{Adjusted}\text{\_}{R}^2$	0.5774	0.2544	0.5678	0.5901

The symbols *** indicate significance at the 1% confidence levels.

and column (4) of Table 5 have a slight difference in their benchmark regression results. The reason is that the impact of the energy structure (Energy) on the labor misallocation (Lamis) is relatively small. Therefore, when conducting the mediation effect test of Lamis, Energy was removed to better improve the estimation efficiency of the model. Column (2) indicates that at the 1% significance level, the $ISU$ exerts a positive effect on $Lamis$. It is indicated that during the sample period, the $ISU$ made $Lamis$ even more severe. The possible reason is that $ISU$ forces workers to acquire corresponding green and low-carbon technologies, but the serious shortage of low-carbon technologies among workers precisely exposes the lagging development of China’s labor market. Column (3) shows that $Lamis$ has a negative effect on $CEI$. That is, although there was a misallocation of labor factors in China during the sample period, it still had a certain inhibitory effect on $CEI$; this inhibitory effect was relatively limited. Column (4) shows the impact of $ISU$ and $Lamis$ on $CEI$, simultaneously. The results show that at the 1% significance level, both have a negative effect. By comparing column (4) with column (1), it is found that after adding the mediating variable $Lamis$, the absolute value of the coefficient estimate of ISU decreased from the original 0.317 to 0.292 (a reduction of 7.89%). Of course, the degree of this reduction is weaker than that in Table 9 for the case of $GTFP$; however, it also indicates that labor misallocation may be another channel for the $ISU$ to reduce the $CEI$. By comparing column (4) with column (3), it can be observed that, regardless of whether the industrial structure upgrading is taken into account or not, labor misallocation significantly increased the $CEI$. Nonetheless, when $ISU$ is taken into account, the negative effect of $Lamis$ on $CEI$ is substantially reduced. Just as the above analysis indicates, throughout the sample period, the upgrading of China’s industrial structure revealed the lagging problem in the development of the labor market and failed to effectively enhance the suppression of carbon emission intensity.

In Table 10, the overall effect of $ISU$ in reducing $CEI$ is 0.317. Among them, the indirect effect of labor misallocation is only 0.029 (0.537 in column (2) multiplied by 0.054 in column (3)), accounting for 9.15%. The direct effect resulting from $ISU$ is as high as 0.288, accounting for 90.85%. Compared with Table 9, the indirect effect caused by $Lamis$ is lower than that of the $GTFP$ (3.73% lower). It indicates that the $Lamis$ also played a certain mediating role between $ISU$ and $CEI$, partially verifying hypothesis 3.

4.7.3. An Exploration of the Relationship Between $GTFP$ and $Lamis$

It is worth noting that, there is a corresponding relationship between $GTFP$ and $Lamis$, which leads to a cumulative effect of the aforementioned intermediary mechanism. Therefore, the trend charts of the two are drawn, as shown in Figure 4 and Figure 5. Comparing Figure 4a,b, although the $GTFP$ of eastern provinces such as Beijing, Tianjin, Shandong, Jiangsu, Shanghai, Zhejiang, and Guangdong has relatively developed steadily or even slightly increased, the GTFP of the remaining eastern provinces including Hebei, Liaoning, Fujian, and Hainan, as well as all central and western provinces, has shown a declining trend. This signifies that the $ISU$ of each province in China did not enable the $GTFP$ to escape the downward trend (although the mechanism test in Table 9 shows that the $ISU$ has a promoting effect on GTFP). The increase in $GTFP$ is a long-term process, and China still needs to better promote the green transformation of industries through energy structure transformation and green technological innovation. Figure 5 illustrates that, from the overall trend, most provinces in the sample period showed an improving trend for $Lamis$ (except for a few regions such as Chongqing and Hainan); the reason is that from the historical perspective of China’s labor market development, the reform of the household registration system and the improvement of the urban-rural dual structure have enhanced the mobility of labor force among different regions and alleviated the problem of labor misallocation in each province to a certain extent.

By combining Figure 4 and Figure 5, it can be observed that during the sample period, on the one hand, the decline in $GTFP$ did not exert the impelling effect on the improvement of workers’ skills in the labor market through factors such as green technology demand; on the other hand, the improvement of $Lamis$ mainly reflects the “quantitative” flow level of workers, but does not truly reflect the “qualitative” dimension such as the improvement of workers’ technical levels, thus it is difficult to have a promoting effect on $GTFP$, that is, the two did not form a positive interaction relationship.

5. Conclusions and Policy Implications

5.1. Research Conclusions

This study employs a four-stage mediation regression model to empirically examine the effect of industrial structure upgrading ($ISU$) on carbon emission intensity ($CEI)$, utilizing panel data from 30 provinces in China from 2004 to 2019, along with the precise mechanisms involved. The subsequent research findings are derived:

Firstly, $ISU$ can significantly reduce the $CEI$. For every 1% increase in the $ISU$, the $CEI$ will decrease by 0.296%. Moreover, through methods such as replacing the dependent variable, truncation handling, model transformation, and endogeneity testing, this conclusion remains robust. When viewed from different regions, this effect is most pronounced in the central region, with a 0.361% reduction in $CEI$ while the western region follows, with a 0.192% reduction; but the eastern region shows no significant effect.

Secondly, the $ISU$ can reduce the $CEI$ of a region by enhancing the green total factor productivity ($GTFP$). For every 1% increase $ISU$, the $GTFP$ increases by 0.322%, and for every 1% increase in $GTFP$, $CEI$ will decrease by 0.12%. The $ISU$ and $GTFP$ can have a negative impact on $CEI$ simultaneously. The former has a direct effect of reducing carbon emissions with a proportion as high as 87.12%; however, under the influence of $ISU$, the inhibitory effect of $GTFP$ on $CEI$ has weakened, and the indirect effect of reducing carbon emissions is 12.88%. Therefore, upgrading the industrial structure is the most direct and effective measure for China to reduce its carbon emission intensity; however, it may still be in its initial stage and has not yet contributed to the reduction in $CEI$ by the $GTFP$.

Thirdly, the $ISU$ can reduce the $CEI$ by influencing the degree of labor misallocation ($Lamis)$. During the sample period, the $ISU$ made the $Lamis$ more severe, revealing the lagging problem in the development of China’s labor market; however, the $Lamis$ had a limited inhibitory effect on the $CEI$. When the $ISU$ is combined with the $Lamis$, they both have a negative impact on $CEI$. The former has a direct effect on reducing carbon emissions, which accounts for as high as 90.85%. Under the influence of $ISU$, the inhibitory effect of $Lamis$ on $CEI$ has weakened somewhat, and the indirect effect of reducing carbon emissions accounts for 9.15%. As China gradually lifted the restrictions on the household registration system, the problem of labor misallocation has gradually been alleviated, and its corresponding effects have begun to be released. However, the low-carbon skills of the workforce still fail to adapt to the upgrading of the industrial structure, and they also do not form an intermediate superimposition effect with the green total factor productivity.

5.2. Policy Implications

(1) Continuously promote industrial structure upgrading to facilitate carbon reduction. The central and western regions should continuously optimize and upgrade their industrial structures, focusing on the transformation of traditional high-energy-consuming and high-pollution industries and the development of low-carbon industries, to achieve a smooth transition of new and old energy sources. The eastern regions need to break through the saturation problem of existing industrial structure upgrading and carbon reduction through low-carbon technologies. At the same time, all regions need to deeply integrate the concept of green development into the formulation and implementation of industrial policies, focus on developing green and intensive industries, and systematically promote regional carbon reduction, thereby driving regional high-quality and sustainable development.

(2) Enhance green technological innovation and improve green total factor productivity. The government can adopt measures such as implementing tax incentives and subsidies to strengthen support for green technology research and development, especially for technologies related to industrial energy efficiency and renewable energy integration that are closely linked to green total factor productivity. By drawing on international experience, establish a “green technology bank” to better facilitate the transformation and application of green technology achievements. Seize the opportunities presented by the new round of technological and industrial revolutions, fully leverage the role of artificial intelligence technology in enhancing green total factor productivity, and promote energy conservation and emission reduction.

(3) Comprehensively deepen labor market reforms and enhance the low-carbon technical proficiency of workers. The government should further deepen reforms in areas such as household registration, reducing the institutional barriers to labor mobility across regions. They can do this by establishing regional employment collaboration platforms, promoting the matching of labor factors in low-carbon industries, conducting vocational skills training for laborers in low-carbon industries, continuously improving the low-carbon technical proficiency of workers, and creating a synergistic effect with green total factor productivity. They should draw on international experience to establish a sound labor transition guarantee mechanism, provide targeted subsidies for workers leaving high-carbon industries and those employed in low-carbon industries, and support the orderly transfer of labor to low-carbon industries.

This study possesses specific limitations that need to be clarified: Firstly, there is a limitation regarding the time range. Due to the data availability constraints, there are many missing official and authoritative data after 2019. To guarantee the rigor and comparability of the research results, this study sets the sample period as 2004–2019, excluding the latest data from 2019 onwards. Secondly, there is a limitation in the research mechanism. This study only separately examines the mediating role of $GTFP$ and labor misallocation. It is difficult to fully capture the interaction and superimposition effects between the two within the sample period. These limitations also leave room for future research. After the official data is updated after 2019, we will further expand the related research.