How Will the Relationship between Technological Innovation and Green Total Factor Productivity Change under the Influence of Service-Oriented Upgrading of Industrial Structure?

Abstract

Technological innovation is closely related to the green total factor productivity (GTFP), which has played an important role in China’s sustainable development goals. However, the relationship between technological innovation and GTFP may change due to the influence of economic factors such as the service-oriented upgrading of industrial structure. This study used a panel dual-threshold regression model to perform an empirical analysis in order to explore this change. We introduced dummy variables to divide the samples into three categories according to the threshold value for group regression. The results show that technological innovation will still promote GTFP under the influence of the service-oriented upgrading of industrial structure. However, this positive influence has a double threshold effect; that is, it led to a nonlinear nexus. The role of technological innovation in promoting GTFP will decrease when the service-oriented upgrading of industrial structures crosses the first and the second thresholds. Additionally, the promotion effect of technological innovation on GTFP in provinces with high levels of service-oriented upgrades of industrial structures is smaller than that in provinces with a low degree of service-oriented upgrading of industrial structures, and even tends to be 0. When the government guides technological innovation to promote the improvement of GTFP, it needs to reasonably consider the composition of industrial structure and coordinate with effective industrial policies.

1. Introduction

Global economic development almost always occurs at the expense of natural resources and results in environmental pollution. Conversely, environmental problems also restrict economic development. Today, development is complicated by conflicts between the environment and economy, and sustainable development is the key method to escaping this cycle. Therefore, the OECD proposed green growth, which aims to focus on natural resources to promote economic development while continuing to provide environmental services and essential assets for human welfare. GTFP emphasizes green growth. It integrates resource input and environmental pollution into the traditional total factor productivity (TFP) index and reflects both the quality and quantity of economic development. This improvement can help to resolve the contradiction between economic development and environmental problems and can contribute to the promotion of sustainable development. Consequently, GTFP has become a crucial indicator for measuring economic development quality, resource utilization efficiency and the degree of environmental protection in recent years [1,2,3].

China has formed an extensive economic development model since the reform and opening up policy. Although this model promotes the rapid growth of the economy, it has high energy and resource costs; it has led to the wasting of resources and to environmental pollution. The Chinese government has paid increasing attention to technological innovation and industrial restructuring in recent years; this can help to transform the model of economic development and promote green economy development. In 2020, the added value of China’s service industry increased by 2.1 percent, accounting for 54.5 percent of the GDP, which shows that the industrial structure is becoming more service oriented. This has led to improvements in GTFP and eased the conflict between economic development and environmental welfare to a certain extent. Therefore, the study of China’s GTFP in this paper is useful. It provides an important reference for developing countries to realize green development.

Schumpeter’s innovation theory purports that technological innovation can increase resource utilization to promote economic growth. The nexus between technological innovation and GTFP has become a hot topic, as GTFP measures the quality of economic development. Most scholars believe that technological innovation can lead to green technological progress and improve production efficiency, which is conducive to improving GTFP and achieving sustainable economic development [4,5,6]. In addition, technological innovation facilitates the formation of new markets, which can improve energy distribution efficiency and reduce pollutant emissions, making it an important driver of the green economy [7,8,9]. Nevertheless, some scholars have found that technological innovation has little or even a negligible impact on the growth of GTFP [10]. The reason may be that the initial cost of green technology innovation is relatively high, meaning it is restricted by economic development. Industry structures also affect GTFP. The adjustment of industrial structures can bring about a reasonable distribution of resources among industries, so that resources can be fully utilized to reduce environmental pollution and promote green economic growth [11,12,13]. Furthermore, the industrial structure is also closely related to technological innovation. The main driving force for upgrading the industrial structure, according to some scholars, is technological innovation [14]. However, some studies have found that the upgrading of industrial structures can stimulate technological innovation [15]. It is also believed that the nexus between technological innovation and industrial structure upgrading may be a two-way interaction [16].

There is no unified analytical system to explain the nexus between technological innovation, the upgrading of industrial structures and GTFP. Moreover, China’s industrial structure has been transformed from the secondary industry to the service industry, and the industrial structure has been upgraded to be service-oriented. Therefore, this paper focuses on the relationship between China’s technological innovation, the service-oriented upgrading of industrial structures and GTFP. What changes may occur in the impact of technological innovation on GTFP under the service-oriented upgrading of industrial structures? How does the effect of technological innovation on GTFP differ in provinces with a different degree of service-oriented upgrading of industrial structures?

To answer the above questions, we selected panel data from 30 provinces in China from 2010 to 2020. The data include autonomous regions and municipalities directly under the Central Government. Tibet, Macao, Hong Kong, Taiwan and other regions with incomplete data were not included. There are three research purposes and main contributions are as follows: (1) incorporate the service-oriented upgrading of industrial structures into the analytical framework between technological innovation and GTFP; (2) using the service-oriented upgrading of industrial structures as a threshold variable, a certain nonlinear nexus between technological innovation and GTFP was examined with a panel dual-threshold regression model; (3) introduce dummy variables to divide China’s 30 provinces into three categories according to the threshold value for group regression, and examine the influence of technological innovation on GTFP within different threshold value ranges, so as to provide more accurate results.

The structure of this article is as follows. Section 1 reveals the theme of this article by placing this research in the broader context of addressing the contradiction between the economy and environment. Section 2 provides a comprehensive review of the relevant research results and puts forward the innovation points of this paper. Section 3 introduces the model and method for measuring GTFP and extends the single-threshold model to models with dual-threshold and multi-thresholds. Section 4 uses the service-oriented upgrading of industrial structures as a threshold variable and identifies two threshold values by verifying the threshold effect. We empirically analyze the nonlinear nexus between technological innovation and GTFP under the service-oriented upgrading of industrial structures by using a dual-threshold regression model. We also introduce dummy variables to divide China’s 30 provinces into three categories according to the threshold value for group regression. We compared the differences in the impact of technological innovation on GTFP among groups within different threshold ranges. Finally, two hypotheses were verified. Section 5 summarizes and interprets the findings, stating the limitations of the article and directions for future research. Section 6 distills the findings of this study and provides corresponding policy recommendations accordingly.

2. Literature Review and Research Hypothesis

Our research topic involves three related areas in the literature. The first is about the nexus between technological innovation and GTFP. Porter et al. [17] believe that enterprises can realize energy conservation and emission reduction through technological innovation, as well as promote resource utilization efficiency. Many scholars then confirmed this hypothesis [18,19,20,21]. Wang et al. [22] found that China’s technological progress can significantly improve the development of GTFP through current environmental regulations by using the spatial Durbin model. Chen et al. [23] found that independent R&D has a more significant driving effect on GTFP than technology introduction by using a dynamic panel data model. Grossman et al. [24] found that technological innovation is an important catalyst for improving environmental quality, because more advanced technology tends to make economic development ‘greener’. Chiou et al. [25] discuss the impact of innovation on green development by using the spatial error model and note that green innovation can significantly promote environmental performance. Cui et al. [26] found that technological innovation had no significant effect on GTFP by using the generalized method of moments.

The second is the nexus between industrial structure upgrading and GTFP, as well as technological innovation. Zhu et al. [27] found that GTFP increased despite significant differences in the green industry structure of the five sub-sectors of the mining industry. Feng et al. [28] concluded that the structural dividend of China’s industrial restructuring must rely on GTFP by using the dynamic panel model. It can strengthen the resource utilization efficiency of industrial sectors and promote the coordinated operation of various economic sectors through upgrading industrial structures. Thus, it reduces all kinds of conflicts in economic operation and promotes high-quality economic development [29]. The upgrading of industrial structures will enhance innovation cooperation within and among industries and promote the formation of spatial spillover effects of technological innovation efficiency. Meanwhile, the method can be used to effectively promote the technological innovation capabilities of upstream and downstream industries [30]. Ngai et al. [31] state that technological innovation accelerates the process of the upgrading of industrial structures.

The third is the service-oriented upgrading of industrial structures and the rationalization of industrial structures. Vandermerwe et al. [32] first proposed the concept of “servitization”, finding that the traditional manufacturing industry can achieve greater value-added results by providing customer-centered services. Additionally, “service” has become the main component in the whole process, and this trend is referred to as “business servitization”. White et al. [33] proposed “servicizing” and defined it from the perspective of the changing roles and functions of manufacturing enterprises, arguing that more service offerings are characteristic of servicizing. This concept expands the meaning of servicizing and incorporates all enterprise products into the servicizing process. Szalavetz [34] divides tertiarization into two levels. First, the efficiency and quality of internal services are increasingly important to the competitiveness of enterprises, such as enterprise technology, human resources quality and operation efficiency. Second, external services related to goods are increasingly important to customers. Thus, the initial research on the tertiarization of the industrial structure mainly refers to the tertiarization of the manufacturing industry in a narrow sense. The definition of industrial structure servitization has evolved from enterprise business servitization to enterprise role servitization, and finally to industrial structure servitization, which is commonly mentioned in today’s research. In this article, the service-oriented upgrading of industrial structures is defined as a service-oriented trend in which the tertiary industry (or service industry) occupies an increasingly high proportion in the national economic structure and gradually exceeds the secondary industry in the process of industrial structure evolution. Gu et al. [35] found that the rationalization and upgrading of industrial structures in the Beijing–Tianjin–Hebei urban agglomeration significantly inhibits carbon emissions. Compared with the rationalization of the industrial structure, the upgrading of industrial structures in the Beijing–Tianjin–Hebei urban agglomeration has a better effect on carbon emission reduction. Zhu et al. [36] found that optimizing the rationalization of the industrial structure in Shanxi Province can improve air quality; in the study, they found that the improvement effect dropped first, and then began to rise. Zhang et al. [37] found that the industrial structure can reduce haze pollution through rationalization; however, the role of industrial structure upgrading has not yet been established.

In reviewing the relevant literature, it is found that there are correlations between technological innovation and GTFP, between the upgrading of industrial structures and GTFP, as well as between technological innovation and industrial structure upgrading. These studies focus on the impact on GTFP from the following two perspectives: technological innovation and industrial structure upgrading. Most of them use single-factor and linear models in their analysis. However, they do not comprehensively consider the relationship between the three factors. Considering the fact that China’s current industrial structure is increasingly service-oriented, this article focuses on the changes of the impact of technological innovation on GTFP under the service-oriented upgrading of industrial structures. The article does not consider the rationalization of industrial structures temporarily. The purpose is to identify how this influence will change and whether it will be positive or negative, and linear or nonlinear. Therefore, we propose two theoretical hypotheses presented in Figure 1.

Hypotheses 1 (H1).

Technological innovation will have a positive and nonlinear effect on GTFP under the influence of service-oriented upgrading of industrial structures.

Hypotheses 2 (H2).

The promotional effect of technological innovation on GTFP tends to be 0 in regions where the service-oriented upgrading of industrial structures has reached a certain degree.

3. Materials and Methods

3.1. The Calculation Model of GTFP

GTFP has become a crucial means to better balance priorities related to the economy and environment, and the widely used calculation method for GTFP is data envelopment analysis (DEA). This approach does not necessarily require a specific functional form, and can consider multiple inputs and outputs [38,39]. DEA cannot consider slack variables in the evaluation of input efficiency and output efficiency, although efficiency is measured according to the radial and angle. Chung et al. [40] first recommended the directional distance function (DDF) of polluting emissions as the unexpected output and proposed the Malmquist–Luenberger (ML) index to measure environmental productivity. It was widely used to measure GTFP. Based on the measures of slack variables, Tone [41] put forward a non-radial and non-angular model named the slack-based model (SBM). It solves the problem that the traditional DDF productivity index ignores the shortcomings of production. Oh [42] proposed an index that combines the GML (Global Malmquist–Luenberger) index recommended by Pastor et al. [43] and the DDF model.

This article uses the SBM–GML model to calculate GTFP. The first step is to build a set of production possibilities that contains the expected outputs and unexpected outputs. Assume that each decision-making unit $l(l=1,\dots ,L)$ of each period $t(t=1,\dots ,T)$ will use M inputs $x=(x_1,\dots ,x_M)\in R_M^{+}$ at the same time and will produce N expected outputs $y=(y_1,\dots ,y_N)\in R_N^{+}$, as well as I unexpected outputs $b=(b_1,\dots ,b_i)\in R_I^{+}$. For each input vector x, the production possibility set can produce an output combination (y, b) containing expected outputs and unexpected outputs. The current period of the production possibility set is transformed into Equation (1) by using the DEA method,

$$\begin{array}{c}{P}^t(x^t)=\{(y^t,b^t):{\displaystyle \sum _{l=1}^L{z}_l^t{y}_{ln}^t\ge y_{ln}^t,n=1,\dots ,N;{\displaystyle \sum _{l=1}^L{z}_l^t{b}_{li}^t=b_{li}^t,i=1,\dots ,I;}}\\ {\displaystyle \sum _{l=1}^L{z}_l^t{x}_{lm}^t\le x_{lm}^t,m=1,\dots ,M;{\displaystyle \sum _{l=1}^L{z}_l^t=1,z_l^t\ge 0,l=1,\dots ,L}}\}\end{array}$$

(1)

where $P^t(x^t)$ is the production possibilities set, and $z_l^t$ is the weight of cross-sectional observation. $z_l^t\ge 0$ represents constant returns to scale (CRS), which is the most popular form of returns to scale in measuring GTFP [44]. Therefore, this paper also constructs a production possibility set based on CRS.

Then, we transformed the current set of production possibility into a set of global production possibility (Equation (2)).

$$\begin{array}{c}{P}^t(x^t)=\{(y^t,b^t):{\displaystyle \sum _{t=1}^T{\displaystyle \sum _{l=1}^L{z}_l^t{y}_{ln}^t\ge y_{ln}^t,n=1,\dots N;}}{\displaystyle \sum _{t=1}^T{\displaystyle \sum _{l=1}^L{z}_l^t{b}_{li}^t=b_{li}^t,i=1,\dots I;}}\\ {\displaystyle \sum _{t=1}^T{\displaystyle \sum _{l=1}^L{z}_l^t{x}_{lm}^t\le x_{lm}^t,m=1,\dots ,M;}}{\displaystyle \sum _{l=1}^L{z}_l^t=1,z_l^t\ge 0,l=1,\dots ,L}\}\end{array}$$

(2)

According to the method of Fukuyama et al. [45], Equation (3) is obtained, which is the global SBM directional distance function considering pollution emissions,

$$\begin{array}{c}{S}_V^G(x^{t,l^{\prime }},y^{t,l^{\prime }},b^{t,l^{\prime }},g^x,g^y,g^b)=\underset{s^x,s^y,s^b}{\max }\frac{\frac{1}{M}{\displaystyle \sum _{m=1}^M\frac{s_m^x}{g_m^x}+\frac{1}{N+I}({\displaystyle \sum _{n=1}^N\frac{s_n^y}{g_n^y}}+{\displaystyle \sum _{i=1}^I\frac{s_i^b}{g_i^b}})}}{2}\\ \mathrm{s}.\mathrm{t}.{\displaystyle \sum _{t=1}^T{\displaystyle \sum _{l=1}^L{z}_l^t}}{x}_{lm}^t+s_m^x=x_{l^{\prime }m}^t,\forall m;{\displaystyle \sum _{t=1}^T{\displaystyle \sum _{l=1}^L{z}_l^t}}{y}_{ln}^t-s_n^y=y_{l^{\prime }n}^t,\forall n;{\displaystyle \sum _{t=1}^T{\displaystyle \sum _{l=1}^L{z}_l^t}}{b}_{li}^t+s_i^b=b_{l^{\prime }i}^t,\forall i;\\ {\displaystyle \sum _{l=1}^L{z}_l^t}=1,z_l^t\ge 0,\forall l;s_m^x\ge 0,\forall m;s_n^y\ge 0,\forall n;s_i^b\ge 0,\forall i\end{array}$$

(3)

where $(x^{t,l^{\prime }},y^{t,l^{\prime }},b^{t,l^{\prime }})$ represents the input vector and the vector of the output of l (decision-making unit) in t (period); $(g^x,g^y,g^b)$ represents the direction vector of the input decrease, expected output increase and unexpected output decrease. $(s_m^x,s_n^y,s_i^b)$ is the slack vector of input and output. When they are both positive, the real input and the unexpected output are larger than the boundaries of input and output, and the real output are smaller than the boundaries of the output. So, $s_m^x,s_n^y,s_i^b$ represents the vectors of excessive input, insufficient expected output and excessive unexpected output.

The GML index is constructed by the directional distance function of SBM, and Equation (4) is the calculation model of GML index from t to t + 1:

$$GM{L}_t^{t+1}=\frac{1+S_V^G(x^t,y^t,b^t;g)}{1+S_V^G(x^{t+1},y^{t+1},b^{t+1};g)}$$

(4)

The GML index of China’s 30 provinces was calculated by MaxDEA8.0 software, which is the GTFP growth rate. Therefore, assuming that the GTFP in 2010 is used as the base period, the GTFP of the corresponding year can be calculated by multiplying the GML index in turn.

3.2. The Panel Threshold Regression Model

In accordance with Hansen’s research [46], we construct the single threshold model after considering relevant control variables (Equation (5)):

$$\begin{array}{cc}\hfill GTF{P}_{it}=& \{\begin{array}{l}{\mu }_i+{\alpha }_{1}T{I}_{it}+{\beta }^{\prime }{x}_{it}+{\epsilon }_{it} if{\mathrm{m}}_{it}\le \gamma \\ {\mu }_i+{\alpha }_{2}T{I}_{it}+{\beta }^{\prime }{x}_{it}+{\epsilon }_{it} if{\mathrm{m}}_{it}>\gamma \end{array}\hfill \\ \hfill {\beta }^{\prime }=& {({\beta }_1,{\beta }_2,{\beta }_3,{\beta }_4,{\beta }_5)}^{\prime }\hspace{1em}{x}_{it}=(m_{it},q_{it},s_{it},v_{it},z_{it})\hfill \\ & \left\{GTF{P}_{it},T{I}_{it},x_{it}:1\le i\le n,1\le t\le T\right\}\hfill \end{array}$$

(5)

Equation (5) can also be written in the form of Equation (6):

$$GTF{P}_{it}={\mu }_i+{\alpha }_{1}T{I}_{it}\cdot I(m_{it}\le \gamma )+{\alpha }_{2}T{I}_{it}\cdot I(m_{it}>\gamma )+{\beta }^{\prime }{x}_{it}+{\epsilon }_{it}$$

(6)

where $GTF{P}_{it}$ is GTFP, $T{I}_{it}$ is technological innovation, and m_it is the service-oriented upgrading of industrial structure. Taking m_it as the threshold variable, this article studies how the influence of technological innovation on GTFP changes under the service-oriented upgrading of the industrial structure. α₁, α₂ represent the estimated coefficients of threshold variables with different thresholds; γ represents the estimated threshold; x_it is a 5 × 1 vector containing the control variables m_it, q_it, s_it, v_it and z_it, which represent the service-oriented upgrading of industrial structure, labor input, capital input, industry scale and degree of openness, respectively; β′ represents the estimated coefficients of the corresponding control variables. i represents provinces, t represents years, μ_i is the constant term, ${\epsilon }_{it}$ represents the error term and I (·) represents the indicative function.

Since there may be multiple thresholds in practical applications, the single-threshold regression model is extended to multi-threshold regression models (Equation (7)).

$$GTF{P}_{it}=\{\begin{array}{l}{\mu }_i+{\alpha }_{1}T{I}_{it}+{\beta }_i^{\prime }{x}_{it}+{\epsilon }_{it}\hspace{1em}if{m}_{it}\le {\gamma }_1\\ {\mu }_i+{\alpha }_{2}T{I}_{it}+{\beta }_i^{\prime }{x}_{it}+{\epsilon }_{it}\hspace{1em}if{\gamma }_1<m_{it}\le {\gamma }_2\\ \dots \\ {\mu }_i+{\alpha }_{n+1}T{I}_{it}+{\beta }_i^{\prime }{x}_{it}+{\epsilon }_{it}\hspace{1em}if{m}_{it}>{\gamma }_n\end{array}$$

(7)

Equation (7) can be simplified as follows:

$$GTF{P}_{it}={\mu }_i+{\alpha }_{1}T{I}_{it}\cdot I(m_{it}\le {\gamma }_1)+{\alpha }_{2}T{I}_{it}\cdot I({\gamma }_1<m_{it}\le {\gamma }_2)+\cdot \cdot \cdot +{\alpha }_{n+1}T{I}_{it}\cdot I(m_{it}>{\gamma }_n)+{\beta }_i^{\prime }{x}_{it}+{\epsilon }_{it}$$

(8)

where ${\gamma }_1<{\gamma }_2<\dots <{\gamma }_n$.

Therefore, the simplified form of the dual-threshold regression model is:

$$GTF{P}_{it}={\mu }_i+{\alpha }_{1}T{I}_{it}\cdot I(m_{it}\le {\gamma }_1)+{\alpha }_{2}T{I}_{it}\cdot I({\gamma }_1<m_{it}\le {\gamma }_2)+{\alpha }_{3}T{I}_{it}\cdot I(m_{it}>{\gamma }_2)+{\beta }_i^{\prime }{x}_{it}+{\epsilon }_{it}$$

(9)

where ${\gamma }_1<{\gamma }_2$.

The panel threshold model was estimated using stata15.1 software.

3.3. Variable Selection and Data Sources

The explained variable of this article is the GTFP, which is calculated using the SBM-GML model shown in Section 3.1. We selected capital, labor and energy as input indicators. Specifically, the full-time equivalent of R&D personnel and the total assets of industrial enterprises above a designated scale in 30 provinces are used to measure labor and capital input, respectively. Additionally, the energy input is electricity consumption. We selected two expected outputs, namely, the revenue from new product sales and the number of new product-development projects of industrial enterprises above a designated scale. There are three unexpected outputs, which are total wastewater discharges, smoke (dust) emissions and sulfur dioxide emissions.

The explanatory variable is technological innovation. There are many measures of technological innovation; for example, Kleinknecht et al. [47] use R&D expenditure, and Borghesi et al. [48] use the number of R&D institutions. Some scholars have measured technological innovation by using patent applications [49,50,51]. This study selects patent applications of industrial enterprises above a designated scale divided by the number of involved enterprises to represent technological innovation; that is, we look at the annual average patent applications of industrial enterprises above a designated scale.

The threshold variable is set as the service-oriented upgrading of industrial structures, which is expressed by the ratio of the added value of tertiary industry to the GDP. The higher the proportion of tertiary industry, the better developed education and technology services become. It has a considerable influence on the recruitment of scientific and technological talent, improving national innovation consciousness and strengthening enterprise innovation.

Other control variables include labor input, R&D investment, industry size and degree of openness, all of which use relevant data on industrial enterprises above a designated scale. The annual average employees; R&D expenditure; the ratio of its main business income to the number of units; and the ratio of its foreign capital to assets are analyzed.

Taking into account the availability of data, the article selected China’s panel data of 30 provinces from 2010 to 2020 for empirical analysis. The data include autonomous regions and municipalities directly under the Central Government. Tibet, Macao, Hong Kong, Taiwan and other regions with incomplete data were not included. We obtained data from the “China Science and Technology Statistical Yearbook”, “China Environment Statistical Yearbook”, the website of China’s National Bureau of Statistics and the websites of the statistical bureaus of each province. Missing data were supplemented by the statistical yearbooks and statistical bulletins of corresponding years in the 30 provinces. The descriptive statistics of each variable are shown in

Table 1. Descriptive statistics of variables.

Variable	Obs	Mean	Std. Dev.	Min	Max
GTFPit	330	1.3923	0.7030	0.1899	5.8590
TIit	330	1.2608	1.0543	0.1026	6.4607
mit	330	0.4435	0.0956	0.2830	0.8309
qit	330	307.6394	331.4487	9.7000	1568.0000
sit	330	267.7477	370.1199	0.8768	2107.2030
vit	330	2.8432	1.1551	0.6649	6.7049
zit	330	0.0201	0.0208	0.0016	0.1100

Note: The author obtained data through collation.

.

4. Results

4.1. Results of the Threshold Effect Test

In accordance with the study by Hansen [46], the bootstrap method was used to test the threshold effect by sampling 100 times. The model was tested with single, double and triple threshold effects, respectively, and the corresponding F value, P value and confidence interval were obtained.

Table 2. The tests of threshold effects.

Threshold Variable	Threshold Number	F-Statistics	p-Value	10%	5%	1%	Threshold Value	95% Confidence Interval
mit	Single	40.00 ***	0.0000	14.9499	21.5643	28.0033	0.4445	[0.4429, 0.4450]
mit	Double	27.22 ***	0.0100	13.7407	17.2065	24.9818	0.5103	[0.5086, 0.5128]
mit	Triple	13.20	0.7000	39.6897	46.5024	59.7104

Note: (a) *** represents a significant level of 1%. (b) All variables were logarithmically processed. Source: the author used Stata software for processing.

shows the significance test results, threshold estimates, and 95% confidence intervals of the model with the service-oriented upgrading of the industrial structure as the threshold variable. The LR graph corresponding to the threshold value was drawn by Stata (Figure 2).

According to Table 2 and Figure 2, 0.4445 and 0.5103 are the two thresholds of the model. This indicates that under the influence of service-oriented upgrading of industrial structures, technological innovation has a dual-threshold effect on GTFP.

4.2. Results of the Panel Dual-Threshold Model

By conducting an empirical study of the panel dual-threshold model (Equation (9)), this article illustrated the nonlinear nexus between technological innovation and GTFP under the service-oriented upgrading of industrial structures.

Table 3. The coefficients of variables estimated with the panel dual-threshold model.

Variables	Coefficient Estimated Values
TIit	mit < 0.4445	0.4445 ≤ mit < 0.5103	mit ≥ 0.5103
	0.8305 *** (0.1007)	0.5059 *** (0.0800)	0.2223 *** (0.0672)
mit	6.4345 *** (0.6695)
qit	0.0016 ** (0.0007)
sit	0.0002 (0.0002)
vit	−0.2163 *** (0.0448)
zit	7.1477 (4.5668)
F test	32.26 ***
Witn-R2	0.4691
Obs	330

Note: (a) *** and ** represent significance levels of 1% and 5%, respectively. (b) () indicates standard errors. (c) All variables were logarithmically processed. Source: the author used Stata software for processing.

shows the details.

Table 3 shows that the coefficient value of technological innovation is always positive no matter which threshold interval it is in. It shows that under the influence of the service-oriented upgrading of industrial structures, technological innovation has a positive influence on GTFP.

However, the coefficient of technological innovation is 0.8305 when the service-oriented upgrading of industrial structures is less than the first threshold of 0.4445. At this time, the impetus of technological innovation to GTFP is large. When it is between the threshold of 0.4445 and 0.5103, the technological innovation coefficient is 0.5059. The promoting effect is weakened. Additionally, the coefficient is 0.2223 when the service-oriented upgrading of industrial structures crosses the second threshold value of 0.51037, indicating that the promotion of technological innovation to GTFP is very small.

One possible reason is that the proportion of tertiary industry in the industrial structure will gradually increase, which will lead to an increasingly unreasonable industrial structure under the simple service-oriented upgrading of industrial structure. Finally, this positive impact will gradually slow down, and a similar “structural deceleration” phenomenon will occur. These results indicate that the promotion of technological innovation on GTFP under the influence of the service-oriented upgrading of industrial structures is nonlinear.

Consequently, H1 is validated.

After observing the coefficients of control variables, we find that both labor input and capital input have a significant positive influence on GTFP because labor and capital support enterprises in carrying out various production activities. The negative impact of industry size on GTFP is significant. Excessive industry scale leads to problems such as poor management and excessive transaction costs, which will be detrimental to improving GTFP. The degree of openness did not significantly promote GTFP because of the expansion of openness, advanced technology, talents and management experience have been introduced, and many foreign high-polluting enterprises or industries have also been undertaken. It also had a highly adverse impact on the environment while promoting economic development.

4.3. Robustness Test

It can be seen from the descriptive statistics in Table 1 that there are large differences in some variables and there may be outliers. Therefore, this article adopts the winsorization method to carry out the robustness test to ensure the stability of data and scientific regression results. All variables were winsorized at the 1% quantile to eliminate the influence of outliers on the estimated results. The first and last 1% outliers were removed to make the data more stable. Then, the estimation test was re-performed with the variable treated with winsorization. The results are shown in

Table 4. Winsorized robustness test results.

Variables	Coefficient Estimated Values
TIit	mit < 0.4445	0.4445 ≤ mit < 0.5103	mit ≥ 0.5103
	0.9010 *** (0.2584)	0.6178 *** (0.0532)	0.2417 *** (0.0485)
mit	1.0712 *** (0.4084)
qit	0.0158 ** (0.0073)
sit	0.0092 (0.0060)
vit	−0.1032 *** (0.0390)
zit	0.9720 (0.3301)
F test	19.27 ***
Witn-R2	0.5461
Obs	330

Note: (a) *** and ** represent significance of 1% and 5%, respectively. (b) () indicates standard errors. (c) All variables were logarithmically processed. Source: the author used Stata software for processing.

.

Table 4 shows that when the threshold variable is the service-oriented industrial structure, technological innovation presents a double-threshold promotion on GTFP, and the regression coefficients of technological innovation decrease after crossing the first and second threshold values and remain significantly positive. The threshold characteristics of the robustness test are basically consistent with the empirical results, and the significance and coefficient sign of variables are not changed after being winsorized, which shows that the model and regression results in the article are robust and scientifically sound.

4.4. Results of the Endogeneity Analysis

Due to the complex factors affecting GTFP, the model in this paper may omit variables, which will lead to endogeneity problems. In reality, the impact of technological innovation on GTFP lags behind to a certain extent, and the technological innovation of the previous period will have a certain impact on the green total factor productivity of the current period. Additionally, the GTFP of the current period will not affect the technological innovation of the previous period. Therefore, the one-stage lag of technological innovation is selected as the instrumental variable (IV) of technological innovation to solve the potential endogeneity problem. The results are shown in

Table 5. Endogeneity analysis: one stage lag.

Variables	Coefficient Estimated Values
L.TIit	mit < 0.4445	0.4445 ≤ mit < 0.5103	mit ≥ 0.5103
	0.8841 ** (0.4112)	0.5846 *** (0.2190)	0.2837 *** (0.0982)
mit	0.9815 ** (0.4421)
qit	0.0251 ** (0.0115)
sit	0.0041 * (0.0024)
vit	−0.0380 * (0.0212)
zit	2.8291 (1.7671)
F test	24.53 ***
Witn-R2	0.4712
Obs	300

Note: (a) ***, ** and * represent significance of 1%, 5% and 10%, respectively. (b) () indicates standard errors. (c) All variables were logarithmically processed. Source: the author used Stata software for processing.

.

The results show that technological innovation with the one-stage lag has a double-threshold promotion effect on GTFP under the service-oriented industrial structure. The regression coefficients of technological innovation with the one-stage lag decrease after crossing the first and second threshold values and remain significantly positive. This indicates that the threshold regression results mentioned above are highly reliable.

4.5. Results of Group Regression

To further explore whether the effect of technological innovation on GTFP is heterogeneous in provinces with different threshold ranges under the influence of service-oriented upgrading of industrial structures, the 30 provinces were grouped according to two thresholds using the dummy variable D_i. The value is 1, if the conditions are met; otherwise, the value is 0 (Equation (10)).

$$D_1=\{\begin{array}{l}1if{m}_{it}<0.4445\\ 0otherwise\end{array}\hspace{1em}{D}_2=\{\begin{array}{l}1if0.4445\le m_{it}<0.5103\\ 0otherwise\end{array}\hspace{1em}{D}_3=\{\begin{array}{l}1if{m}_{it}\ge 0.5103\\ 0otherwise\end{array}$$

(10)

D₁, D₂ and D₃ represent the provinces where the service-oriented upgrading of industrial structures is lower than 0.4445, between 0.4445 and 0.5103, and higher than 0.5103, respectively. Table 5 shows the results of group regression.

Table 6. The group regression results.

Variables	Coefficient Estimated Values
	D1	D2	D3
TIit	0.6901 *** (0.0725)	0.2461 ** (0.1172)	0.0742 ** (0.0373)
Z	control	control	control
F test	20.31 ***	12.81 **	18.35 **
Obs	192	84	54
Witn-R2	0.5371	0.4681	0.6012

Note: (a) *** and ** represent significance of 1% and 5%, respectively. (b) () indicates standard errors. (c) All variables were logarithmically processed. Source: the author used Stata software for processing.

shows that the technological innovation coefficient of group D₁, group D₂ and group D₃ are 0.6901, 0.2461 and 0.0742, respectively, which are all significant. It shows that the role of technological innovation in accelerating GTFP in provinces with a high degree of service-oriented upgrading of industrial structures is smaller than that of provinces with a low degree, and even tends to be 0.

This finding is statistically significant but seems counterintuitive. Generally speaking, the effect of the promotion of technological innovation on GTFP should be greater when tertiary industry occupies a large proportion of the industrial structure. A possible reason is that the industrial structure will gradually deviate from rationalization if a region only pays attention to the highly service-oriented upgrading of industrial structures. Therefore, the promotion of technological innovation on GTFP becomes smaller and less significant, and finally becomes 0.

Accordingly, H2 can be confirmed.

After the above verification, we believe that the service-oriented upgrading of industrial structures is an important economic factor that can affect the nexus between technological innovation and GTFP. It results in the promotion of GTFP by technological innovation becoming nonlinear, and eventually trending towards 0 (Figure 3).

5. Discussion

This article focuses on the relationship between China’s technological innovation, the service-oriented upgrading of industrial structures and GTFP. What changes may occur in the impact of technological innovation on GTFP under the service-oriented upgrading of industrial structures? How does the effect of technological innovation on GTFP differ in provinces with different degrees of service-oriented upgrading of industrial structures?

The relevant literature mainly examines the effect of technological innovation on GTFP, or industrial structure upgrading on it, and most studies use single-factor and linear models in the analysis. Du et al. [4], and Wang et al. [22] state that technological innovation could enhance production efficiency, thereby promoting the growth of GTFP. Jin et al. [10], and Cui et al. [26] note that technological innovation did not promote GTFP. Cortuk et al. [13], and Zhang [29] found that industrial structure adjustment can promote green economic growth. Some scholars have found that there may also be a linear nexus between technological innovation and industrial structure upgrading. Peneder [14], and Ngai et al. [31] found that technological innovation is a crucial way to promote industrial structure upgrading. Azadegan et al. [15] believe that industrial structure upgrading promotes the technological innovation of enterprises. Antonelli [30] believes that the nexus between technological innovation and industrial structure upgrading is a two-way interaction. There is no unified analytical framework for the study on the nexus between technological innovation, industrial structure upgrading and GTFP. In fact, the nexus between technological innovation and GTFP is influenced by some economic factors such as industrial structure upgrading. This may cause their relationship to be nonlinear, which depends on a threshold variable.

This article selected panel data from China’s 30 provinces from 2010 to 2020. It includes autonomous regions and municipalities directly under the Central Government. Tibet, Macao, Hong Kong, Taiwan and other regions with incomplete data were not included. We studied the nonlinear nexus between technological innovation and GTFP under the service-oriented upgrading of industrial structures by adopting the panel dual-threshold regression model. The results show that when the threshold variable is the service-oriented upgrading of industrial structures, technological innovation will play a dual-threshold promotion role for GTFP, showing a nonlinear nexus. Considering the heterogeneity of the service-oriented upgrading level of industrial structures in the 30 provinces of China, dummy variables were introduced to divide them into three categories according to the threshold value, and group regression was performed. The study found that the effect of the promotion of technological innovation on GTFP in provinces with a high degree of service-oriented upgrading of industrial structures is smaller than that of provinces with a low degree of service-oriented upgrading of industrial structures, and even tends to be 0. These results have practical significance and could help to guide the government to improve GTFP by formulating relevant technological innovation policies and industrial policies.

Our research found that the industrial structure will become increasingly unreasonable when the ratio of tertiary industry in the industrial structure gradually increases. Although technological innovation has promoted GTFP under the simple service-oriented upgrading of industrial structures, eventually this positive impact will slow down, and a similar “structural deceleration” phenomenon will occur. The conclusions of this paper have many similarities with the phenomenon of “structural deceleration”, which means that the economic growth rate does not match the advanced process of industrial structures. This study found that the effect of the promotion of technological innovation on GTFP does not match the service-oriented upgrading process of industrial structures, which manifests as the “structural deceleration” of the green economy. Only focusing on the highly service-oriented upgrading of industrial structures will lead to an increasingly unreasonable industrial structure and a lack of coordination between industries. This unreasonable industrial structure will hinder the promotion of technological innovation to GTFP.

This article provides important practical significance for government departments to study technological innovation policies and industrial policies to improve GTFP. The applied statistical threshold regression model and group regression method provide more general results for government departments to formulate relevant policies in various regions with different levels of service-oriented upgrades of industrial structures.

6. Conclusions

The purpose of this study was to explore how the nexus between technological innovation and GTFP would change under the service-oriented upgrading of industrial structures. The SBM-GML model was used to calculate GTFP, and China’s relevant input–output data from 30 provinces were collected for the period of 2010–2020. Then, the threshold model was constructed with the service-oriented upgrading of industrial structures as a threshold variable. Finally, dummy variables were introduced to perform group regression. The results verified two hypotheses. Technological innovation has a nonlinear positive effect on GTFP under the influence of the service-oriented upgrading of industrial structures; this positive effect tends to be 0 in areas where the service-oriented upgrading of industrial structures reaches a certain level.

The article makes corresponding policy recommendations based on these statistical results. Firstly, the government needs to rationally formulate differentiated incentive policies for technological innovation by studying the actual situation of the industrial structure in each region because the promoting effect of technological innovation on GTFP will be affected and constrained by the service-oriented upgrading of industrial structures. Secondly, the government should not only pursue the service-oriented upgrading of industrial structures, but also comprehensively consider whether the industrial structure is reasonable, so that technological innovation can improve GTFP stably and effectively according to reasonable industrial policies. In addition, the industry scale should be controlled. If the scale of the industry is too large, there will be problems such as poor management and excessive transaction costs, which will negatively affect the improvement of GTFP. When expanding the degree of openness, it is necessary to reasonably undertake high-polluting enterprises or industries while introducing advanced technology, talent and management expertise. Finally, continuously and effectively increasing labor input and capital input will support the promotion of GTFP.

There are some limitations to this study. The service-oriented upgrading of industrial structures is not necessarily a complete industrial structure upgrading, but only a form of it. In future research, we will consider the combination of industrial structure upgrading and rationalization to construct a threshold model and compare the characteristics of and differences in the influence of technological innovation on GTFP in these two cases.