The Impact of Digital Economy Agglomeration on Regional Green Total Factor Productivity Disparity: Evidence from 285 Cities in China

Abstract

Purpose—The unbalanced layout of digital economy agglomeration has a significant impact on regional high-quality development. This study aims to explore the impact of digital economy agglomeration on regional green total factor productivity (GTFP) disparity from two aspects, including theoretical mechanism and empirical effect. Design/methodology/approach—Based on the empirical data of 285 cities above the prefecture level in China from 2003 to 2018, super-efficiency undesired SBM model, spatial Dubin model, and intermediary effect model are utilized to analyze how digital economy agglomeration affects regional GTFP disparity. Findings—The results show that the GTFP of China is on the rise as a whole, but the gap among cities is gradually expanding. Digital economy agglomeration has significant positive direct effects and positive spillover effects on GTFP, but digital economy agglomeration also aggravates the regional GTFP disparity due to disequilibrium industrial upgrading mechanism. Originality/value—The paper confirms the relationship between digital economy agglomeration and regional GTFP disparity for the first time. Different from previous studies, the industrial upgrading mechanism in this paper includes industrial structure upgrading and industrial spatial evolution. The study calls for the industrial bottleneck of “low-end locking” in underdeveloped cities to be noticed.

1. Introduction

The top priority of the economic growth in China is changing from high-speed to high-quality. The goal of regional coordinated development also changes from balancing the growth rate to balancing the growth quality [1]. Along with these changes, new imbalanced economic situations among regions are emerging. Empirical data show that the regional total factor productivity (TFP) disparity in China is gradually expanding in the 21st century [2,3,4,5]. The key to reversing productivity differentiation lies in filling the technological gap among regions [6,7,8]. With the transformation of growth momentum, digital technology becomes the prime power of technology progress. Digital technology not only unleashes the potential of economic development, but also accelerates the flow of production factors, which creates a historic opportunity to reshape the regional high-quality development pattern. In order to take a leadership position in the field of digital technology, Chinese cities have set off a boom in digital construction, which enables the digital technology element to flow and accumulate at an extraordinary scale and speed across the county. The level of digital economy agglomeration varies from city to city, and the economic effects of digital economy agglomeration are also not the same. What impact does digital economy agglomeration have on regional TFP disparity under the new background? In this paper, the impact is analyzed from two aspects, including theoretical mechanism and empirical effect.

With the intensification of ecological environment problems, the necessity of incorporating environmental pollution factors for the TFP analysis has drawn the attention of many researchers. Commonly used models to measure green total factor productivity (GTFP) include DEA model [9], SBM model [10], ML Index [11], etc. Based on the economic statistics of China, Chen et al. [12], Liu et al. [13], Cheng and Li [14] found that the regional GTFP disparity in China showed a widening trend, and the main factors that affect the regional GTFP disparity include resource endowment [15], technological level [16], industrial structure [17], institutional constraints [18], environmental regulation [19], etc. Different from traditional classical economics, the new economic geography emphasizes the influence of spatial agglomeration factors on regional productivity disparity [20,21]. The theory holds that the spatial agglomeration of economic activities has an impact on regional productivity disparity due to the joint effects of agglomeration and dispersion [22]. Some studies pointed out that the aggregation of production factors with knowledge attributes promotes the growth of production efficiency due to sharing effects, allocation effects and knowledge effects, and it can also exacerbate regional productivity disparity due to cumulative circular effects and self-proliferation effects [23,24]. However, some studies indicated that the diffusion effects, competition effects and crowding effects caused by the agglomeration of production factors in high agglomeration areas can indirectly compensate for the efficiency loss in low agglomeration areas, which is conducive to the balanced growth of regional productivity [25,26]. Other studies emphasized that the role of agglomeration effect and dispersion effect need to be matched with the local development situation, especially the impact of adjustment factors such as economic development level which should be considered [27,28]. The empirical tests are conducted based on the perspectives of human capital agglomeration [29], innovation agglomeration [30], technology industry agglomeration [31], financial agglomeration [32], etc.

In the existing research on digital economy, scholars focus on its green growth effect. Traditional quantitative methods ignore the correlation effect of economic activities among regions. With the development of the new economic geography theory, spatial concepts such as distance are introduced into social science analysis. The mainstream spatial measurement methods include Spatial Autoregressive Model (SAR), Spatial Error Model (SEM) and Spatial Durbin Model (SDM). Hu and Guo [33] used the SDM model and found that digital economy has a significant positive impact and a trickle-down effect on GTFP. Based on the data from Italian business sector companies, Giannini et al. [34] applied the SEM model to confirm a positive direct relationship and a positive spatial spillovers relationship between ICT facilities and TFP. Some scholars pay attention to the impact of digital economy on regional productivity differences, but the conclusions of these studies are uncertain. Liu et al. [35] found that the higher the level of regional development, the greater the promotion effect of the digital economy on GTFP, and industrial structures upgrading is an intermediary transmission mechanism for the digital economy to promote GTFP. But Ma and Zhu [36] held that it has positive nonlinear effects on high-quality green development, and the marginal effects are clearly diminishing. The study by Iammarino and Jona-Lasinio [37] suggested a complementary relationship between ICT production and diffusion in explaining inter-regional differences in productivity performances. Based on the spatial perspective of digital economy, Ren et al. [38] discussed the impact of digital economy agglomeration on GTFP for the first time. Their research pointed out that digital economy agglomeration has a positive effect on inclusive green growth, while it inhibits the inclusive green growth of neighboring cities. To sum up, there are no specific direct empirical contributions on the effects and the theoretical mechanism of digital economy agglomeration on regional GTFP disparity. That is the novelty of this contribution.

The primary goals of the study are to assess the effect of digital economy agglomeration on regional GTFP disparity and identify the transmission mechanism of this effect. The specific contributions of this research are as follows: firstly, the regional heterogeneity and spatial correlation effects of digital economy agglomeration on GTFP is identified by using the static panel model and SDM model, which is helpful to improve the research that ignores the factors of digital economy agglomeration; secondly, the relationship between digital economy agglomeration and regional GTFP disparity is confirmed by using urban panel data, which fills the empirical gap of existing research; thirdly, the existing research on industrial upgrading only focuses on industrial structure optimization. In this paper, industrial upgrading involves both industrial structure optimization and industrial space evolution, which is the novelty of this research. On this basis, the mediating effect of industrial upgrading for digital economy agglomeration aggravates the regional GTFP disparity is confirmed by using the mediation effect model.

Section 2 conducts the theoretical analysis and proposes the hypotheses. Section 3 describes the data for the variables and discusses the empirical methodology. Section 4 points out the empirical results. Section 5 explains the influence mechanism test model and presents the influence mechanism test results. Section 6 discusses the findings. Section 7 concludes the paper and puts forward policy recommendations.

2. Theoretical Analysis and Hypothesis

2.1. Hypothesis of the Effect of Digital Economy Agglomeration on Regional GTFP Disparity

Digital economy agglomeration is defined as the clustering phenomenon of digital economy elements within a specific geographical scope, which involves digital users, digital products, digital services, digital industries, etc. It has powerful functions of information transmission, data processing, and network connectivity, which can enhance regional competitiveness by reducing transaction costs, optimizing resource allocation, improving production models, and stimulating technological innovation. It can continuously attract the influx of high-quality production factors and create favorable conditions for regional green development [39].

Since technological embedding should be matched with the growth stage of the economy, the moderating influence of economic development level on the promotion effect of digital economy agglomeration on GTFP needs to be considered. Combined with the new economic geography theory, the productivity effect of economic activity agglomeration is affected by agglomeration force and diffusion force. With its super absorptive capacity and leading technological advantages, the core regions can quickly tap the potential of digital integration to improve GTFP. Due to the weak comprehensive strength, it is difficult for peripheral regions to rely on digital technology to overcome the industrial bottleneck of low-end lock-in in the short term. The green efficiency effect of digital economy agglomeration in peripheral regions is limited. With the expansion of the scale of digital economy agglomeration, the comparative advantage and scale effect in core areas are becoming increasingly obvious, which inevitably leads to the siphon phenomenon that high-quality talents and high-end industries agglomerating are concentrated from peripheral areas to core areas. As a result, the Matthew effect of “the stronger is stronger, the weaker is weaker” is generated, which exacerbates the regional GTFP disparity [40]. The green efficiency effect of digital economy agglomeration in core areas may also be lower than in peripheral areas, because under the law of diminishing utility it is more difficult for core regions to improve economic quality by relying on digital scale dividends. The phenomena of market congestion, homogenization and vicious competition in core areas are emerging due to excessive agglomeration, which inhibits the growth of productivity. In contrast, the peripheral areas are in the early stages of improving economic quality. The technological productivity generated by digital economy agglomeration can circumvent the restrictions of market segmentation and broaden the market space of leading industries, which is conducive for peripheral areas to completing the leap-forward development of high-quality economy [41]. Accordingly, it is hypothesized that:

Hypothesis 1a.

With the improvement of the level of economic development, the marginal improvement effect of digital economy agglomeration on GTFP is greater, which further aggravates the regional GTFP disparity.

Hypothesis 1b.

With the improvement of the level of economic development, the marginal improvement effect of digital economy agglomeration on GTFP is smaller, which further narrows the regional GTFP disparity.

2.2. Hypothesis of the Mediating Mechanism of Digital Economy Agglomeration Affecting Regional GTFP Disparity

There is a consensus on the driving effect of digital technology on industrial upgrading. Scholars pointed out that the integration of the digital economy and the real economy has changed the production methods, technical efficiency, and value creation methods of traditional industries, which is conducive to the transformation of the product value chain from low added value to high added value [42]. It can also promote the formation of new products, new formats, and new markets, and guide the industrial structure to be rationalized, advanced, and innovative.

According to the theory of open industrial upgrading, industrial digital integration promotes the flow of production factors to the high-skilled, high-efficiency, and high-value-added industries, and spontaneously eliminates the low-skilled, low-efficiency, and low-value-added industries. Under the effect of industrial upgrading caused by the accumulation of digital elements, the trend of industrial evolution is formed where the proportion of high-skilled industries is increased, and the proportion of low-skilled industries is decreased in terms of geographical distribution. Affected by the differences in regional industrial development, the promotion effect of digital economy agglomeration on industrial upgrading in the core area is different from that in the peripheral areas. There is uncertainty about the impact of digital economy agglomeration on regional GTFP disparity. On the one hand, there is a good development environment for enterprises with the relatively complete industrial supporting system in the core area, and digital economy agglomeration generates stronger scale effects, spillover effects and innovation effects. As the level of high-skilled industry agglomeration increases, it contributes to the promotion of regional GTFP. Due to the weak industrial foundation and lack of high-efficiency human resources, the basic driving force for industrial transformation and upgrading in peripheral areas is insufficient, which is not conducive to narrowing the disparity of economic quality. On the other hand, since the adjustment of industrial structure in core areas has reached a mature stage, the role of industrial optimization released by digital economy agglomeration is limited. In contrast, the industrial structure in peripheral areas can be optimized to a great extent. The flying-geese theory points out that regional industry development has the characteristic of gradient transfer. Industrial spatial transfer can drive the change in technological structure in the receiving area, which creates conditions for alleviating the imbalance of economic development among regions. Accordingly, it is hypothesized that:

Hypothesis 2a.

The regional GTFP disparity is aggravated by digital economy agglomeration due to differentiated industrial upgrading mechanisms, which is manifested in the fact that the marginal effect of digital economy agglomeration in core areas to promote the agglomeration of high-skilled industries and inhibit the agglomeration of low-skilled industries is greater than that in peripheral areas.

Hypothesis 2b.

The regional GTFP disparity is narrowed by digital economy agglomeration due to differentiated industrial upgrading mechanisms, which is manifested in the fact that the marginal effect of digital economy agglomeration in core areas to promote the agglomeration of high-skilled industries and inhibit the agglomeration of low-skilled industries is smaller than that in peripheral areas.

3. Empirical Research Design

3.1. Measurement and Evolution Characteristics of GTFP

The super-efficiency undesired SBM model proposed by Tone [43] can effectively solve the input–output relaxation improvement problem, the undesired output problem and the indistinguishable problem of the same effective DMU efficiency value. A super-efficiency non-expected SBM model with constant returns to scale is constructed to measure urban GTFP in this paper.

$$\begin{gathered} Min\hspace{1em}\rho =\left(1-\frac{1}{m}{\displaystyle \sum _{i=1}^m{s}_i^{-}/}{x}_{ik}\right)/\left(1+\frac{1}{q_1+q_2}({\displaystyle \sum _{r=1}^{q_1}{s}_r^{+}/}{y}_{rk}+{\displaystyle \sum _{g=1}^{q_2}{s}_g^{\mathrm{b}-}/}{b}_{gk})\right) \\ s.t.{\displaystyle \sum _{j=1,j\ne k}^n{x}_{ij}{\lambda }_j+s_i^{-}=x_{ik}};{\displaystyle \sum _{j=1,j\ne k}^n{y}_{ij}{\lambda }_j-s_r^{+}=y_{ik}};{\displaystyle \sum _{j=1,j\ne k}^n{b}_{ij}{\lambda }_j+s_g^{b^{-}}=b_{ik}}; \lambda ,s^{+},s^{-}\ge 0 \end{gathered}$$

(1)

where $\rho$ is the GTFP value; $m$, $q_1$, $q_2$ and $s^{-}$, $s^{+}$, $s^{{\mathrm{b}}^{-}}$ are the types and slack variables of input $x$, expected output $y$, and undesired output $b$; $n$ is the number of decision-making units; $\lambda$ is the weight of each decision-making unit.

GTFP is the main driving force for sustainable economic development under the constraints of resources and environment. Different from the traditional calculation method of TFP, GTFP is calculated on the basis of minimizing the input of production factors such as capital and labor and maximizing the economic output, while adding the undesired indicators such as resource consumption and environmental pollution. Referring to Lu et al. [44], the indicator system of GTFP is constructed. Input indicators include five elements: capital, labor, land, water resources, and energy, which are measured by the stock of fixed capital, the total number of employees, the land area of administrative regions, the total water supply and the total electricity consumption. Expected output indicators include the actual gross domestic product. Undesired output indicators include three elements: the industrial wastewater discharge, industrial sulfur dioxide discharge and industrial soot discharge. The capital stock is accounted for using the perpetual inventory method, and the formula is $K_{it}=(1-{\sigma }_{it})K_{i,t-1}+I_{it}$, where $K$ is the capital stock, i represents the city, t is the year, $\sigma$ is the depreciation rate, which is 9.6%, $I$ is the total investment in fixed assets at constant price, the initial capital stock is determined by dividing the actual total fixed asset investment in 2004 by the sum of the depreciation rate and the average annual growth rate of the actual total fixed asset investment from 2004 to 2008.

Figure 1 shows the evolution trend of the nuclear density curve of GTFP from 2003 to 2018. The peak of the nuclear density curve of GTFP gradually shifted to the right and continued to rise, and the double-peak feature of the curve was more significant. It shows that the overall level of GTFP in Chinese cities maintains an upward trend, but the GTFP disparity among cities has an evolutionary characteristic of expanding volatility.

3.2. Variables and Data

(1)

Explained variable: ① Green Total Factor Productivity (GTFP). GTFP is measured by the super-efficiency undesired SBM model; ② Regional GTFP Disparity (RGD). Referring to the methods of Zhong and Lin [45], regional GTFP disparity is calculated using the coefficient of variation (CV). The formula is $C{V}_{it}^2={(Y_{it}-\overline{Y_{\mathrm{t}}})}^2/{\overline{Y_t}}^2$, where $Y_{it}$ is the GTFP of city i in period t, $\overline{Y_{\mathrm{t}}}$ is the national average GTFP value in period t. The larger the coefficient of variation, the larger the regional GTFP disparity.

(2)

Explaining variable: Digital Economy Agglomeration (Dag). The systematic indicator system for digital economic accounting has not yet been formed. According to provincial or municipal statistical data, there are statistical problems of inconsistent statistical calibers and statistical discontinuity for key indicators, such as telecommunication business volume, express delivery business volume, number of Internet pages, and number of Internet URLs. It is difficult to construct a composite index evaluation system for the digital economy. Huang et al. [46], Zhao et al. [47], Deng and Zhang [48] mainly calculated the level of digital economy from two aspects, including digital user scale and digital industry development. In this paper, the geographical agglomeration characteristics of digital economy are measured also from two dimensions: digital user agglomeration (Dag^a) and digital industry agglomeration (Dag^b). The former is calculated by the agglomeration level of mobile phone users, and the latter is calculated by the agglomeration level of information software service industry. The degree of agglomeration is measured by the location entropy index, and the formula is $Da{g}_{it}=(d{q}_{it}/q_{it})/({\displaystyle \sum d{q}_t}/{\displaystyle \sum q_t})$, where $d{q}_{it}$ is the number of mobile phone users or information software service employees of city i in period t, $q_{it}$ is the total number of jobs of city i in period t, ${\displaystyle \sum d{q}_t}$ is the number of mobile phone users or information software service employees in the country in period t, ${\displaystyle \sum q_t}$ is the total employments in the country in period t. The larger the value of this indicator, the higher the degree of digital economy agglomeration.

(3)

Control variables: according to the existing literature, the influencing factors of regional GTFP mainly include labor skill factors, economic development factors, urban scale factors, infrastructure factors, institutional factors, environmental policy factors, etc. Referring to Li et al. [49], Zhang et al. [50], Guo and Chen [51], Human Capital (Hcp), Financial Development Level (Fin), Population Density (Pop), Transportation Level (Tra), Government Intervention (Gov), and Environmental Regulation Intensity (Reg) are used as control variables. Human capital is measured by the ratio of the number of students in colleges and universities to the total amount of employment. The level of financial development is measured by the logarithm of per capita loans of financial institutions at the end of the year. The population density is measured by the logarithm of the proportion of the total population at the end of the year to the land area of the administrative area. The transportation level is measured by the logarithm of the per capita occupied area of urban roads. The degree of government intervention is measured by the proportion of expenditure within the fiscal budget to the regional GDP. The intensity of environmental regulation is measured by the proportion of industrial sulfur dioxide removal to the total amount of emissions and removal. The nonlinear effect of environmental regulation intensity is considered by calculating the square term of environmental regulation intensity. The descriptions of the variables are summarized in

Table 1. The definition and description of the variables.

Variable Types	Variable Name	Variable Symbol	Description
Explained variable	Green Total Factor Productivity	GTFP	Measured by the super-efficiency undesired SBM model
	Regional GTFP Disparity	RGD	Calculated using the coefficient of variation
Explaining variable	Digital Economy Agglomeration	Daga	Digital user agglomeration (Daga) is calculated by the agglomeration level of mobile phone users
		Dagb	Digital industry agglomeration (Dagb) is calculated by the agglomeration level of information software service industry
Control variables	Human Capital	Hcp	The ratio of the number of students in colleges and universities to the total amount of employment
	Financial Development Level	Fin	The logarithm of per capita loans of financial institutions at the end of the year
	Population Density	Pop	The logarithm of the proportion of the total population at the end of the year to the land area of the administrative area
	Transportation Level	Tra	The logarithm of the per capita occupied area of urban roads
	Government Intervention	Gov	The proportion of expenditure within the fiscal budget to the regional GDP
	Environmental Regulation Intensity	Reg	The proportion of industrial sulfur dioxide removal to the total amount of emissions and removal
		Reg2	The nonlinear effect of environmental regulation intensity is considered by calculating the square term of Reg

.

(4)

Data description: according to the comprehensiveness and availability of empirical data, the panel data of 285 cities above the prefecture level in China from 2003 to 2018 is used for analysis. The data come from the China Statistical Yearbook, China Urban Statistical Yearbook, local (province, municipal, autonomous region) statistical yearbooks and the Statistical Bulletin of National Economic and Social Development. Some missing data were filled by interpolation method or exponential smoothing method. In order to eliminate the influence of inflation, all value form data are deflated by the GDP index, CPI index or fixed asset investment price index of the province where the city is located.

3.3. Empirical Model

A benchmark panel regression model is built to explore the impact of digital economy agglomeration on GTFP. The model is expressed as follows:

$$GTF{P}_{it}={\alpha }_0+\delta Da{g}_{it}+\gamma Contro{l}_{it}+{\epsilon }_{it}$$

(2)

Considering the effect of economic spatial correlation factors on GTFP, a spatial panel regression model is constructed. The model is expressed as follows:

$$\begin{array}{l}GTF{P}_{it}=\rho W GTF{P}_{it}+{\alpha }_0+\delta Da{g}_{it}+\gamma Contro{l}_{it}+\phi W Da{g}_{it}+\theta WContro{l}_{it}+{\mu }_{it}\\ {\mu }_{it}=\lambda W{\mu }_{it}+{\epsilon }_{it}\end{array}$$

(3)

According to the regional differences in the green effect of digital economy agglomeration, a static panel regression model is constructed to explore the impact of digital economy agglomeration on regional GTFP disparity. The model is expressed as follows:

$$RG{D}_{it}={\alpha }_0+\delta Da{g}_{it}+\gamma Contro{l}_{it}+{\epsilon }_{it}$$

(4)

where ${\alpha }_0$ is the constant term, ${\epsilon }_{it}$ is the random error term, $\delta$ is the regression coefficient of digital economy agglomeration, $\gamma$ is the control variable regression coefficient vector, $\rho$, $\phi$, $\theta$ represent the effects of GTFP, digital economy agglomeration and control variables of neighboring cities on GTFP of local cities, $\lambda$ is the interaction effect between the error terms of two regions. $W$ is the spatial weight matrix, both the geographic distance matrix ($W^G$) and the nested economic distance matrix ($W^E$) are used as the spatial weight matrix in this paper. The main diagonal elements of both matrices are 0, and the non-main diagonal elements are $W_{ij}^G=1/d_{ij}^2$ or $W_{ij}^E=W_{ij}^G\ast (1/\left|\overline{Pgd{p}_i}-\overline{Pgd{p}_j}\right|)$, where $d_{ij}^2$ is the square of the geographic straight-line distance between city i and city j, $\overline{Pgd{p}_i}$ is the average per capita real GDP of city i. Spatial econometric models mainly include Spatial Autoregressive Model (SAR), Spatial Error Model (SEM) and Spatial Durbin Model (SDM). When $\phi$, $\theta$, $\lambda$ are 0, Equation (2) is SAR; when $\rho$, $\phi$, $\theta$ are 0, Equation (2) is SEM; when $\lambda$ is 0, Equation (2) is SDM; when $\rho$, $\phi$, $\theta$, $\lambda$ are 0, Equation (2) degenerates into an ordinary panel regression model.

4. Empirical Results Analysis

4.1. The Impact of Digital Economy Agglomeration on GTFP

(1) Regression results based on full sample. A benchmark regression test is performed. After the F test, Hausman test, and time effect test, a two-way fixed-effect model with city dummy variables and year dummy variables was finally selected (Due to space limitations, the test results are not listed). All models use robust standard errors. As tabulated in

Table 2. Estimated results of the impact of digital economy agglomeration on GTFP with full sample.

Variable	Explaining Variable: Daga			Explaining Variable: Dagb
	FE	$\mathbf{SDM}: {\mathit{W}}^{\mathit{G}}$	$\mathbf{SDM}: {\mathit{W}}^{\mathit{E}}$	FE	$\mathbf{SDM}: {\mathit{W}}^{\mathit{G}}$	$\mathbf{SDM}: {\mathit{W}}^{\mathit{E}}$
	Model (1)	Model (2)	Model (3)	Model (4)	Model (5)	Model (6)
Dag	0.100 *** (9.52)	0.083 *** (11.31)	0.093 *** (12.36)	0.016 ** (2.15)	0.011 * (1.74)	0.016 ** (2.54)
Hcp	0.098 (0.78)	0.178 * (1.71)	0.135 (1.29)	0.618 *** (5.67)	0.605 *** (6.28)	0.609 *** (6.25)
Fin	0.045 *** (2.79)	0.038 *** (3.08)	0.038 *** (3.07)	0.043 *** (2.62)	0.037 *** (3.05)	0.037 *** (3.04)
Pop	−0.163 *** (−3.38)	−0.082 * (−1.69)	−0.135 *** (−2.66)	−0.213 *** (−3.98)	−0.115 ** (−2.26)	−0.177 *** (−3.53)
Tra	−0.023 ** (−2.33)	−0.025 *** (−3.06)	−0.024 *** (-2.86)	−0.016 (−1.59)	−0.019 ** (−2.19)	−0.018** (−1.98)
Gov	−0.421 *** (−4.30)	−0.541 *** (-8.03)	−0.485 *** (-6.86)	−0.241 *** (−2.59)	−0.425 *** (−6.31)	−0.406 *** (−5.47)
Reg	−0.062 * (−1.69)	−0.057 * (−1.66)	−0.056 * (−1.67)	−0.071 * (−1.88)	−0.078 ** (−2.12)	−0.077 ** (−2.21)
Reg2	0.122 *** (2.99)	0.110 *** (3.00)	0.112 *** (3.03)	0.132 *** (3.18)	0.134 *** (3.42)	0.136 *** (3.63)
W*Dag		0.091 *** (4.37)	0.039 ** (2.37)		0.091 *** (4.28)	0.025 * (1.68)
Constant/ρ	1.569 *** (3.97)	0.184 *** (6.62)	0.072 *** (3.27)	1.843 *** (4.29)	0.227 *** (8.34)	0.096 *** (4.38)
W *Contorl		Yes	Yes		Yes	Yes
city/year	Yes	Yes	Yes	Yes	Yes	Yes
Wald_spa_lag		49.872 ***	17.778**		77.106 ***	30.151 ***
LR_spa_lag		49.221 ***	17.739**		76.475 ***	30.048 ***
Wald_spa_err		58.175 ***	19.365**		76.759 ***	29.652 ***
LR_spa_err		57.973 ***	19.375**		76.744 ***	29.588 ***
Adjust R2	0.757	0.764	0.759	0.746	0.757	0.750
N	4560	4560	4560	4560	4560	4560

Note: *** p < 0.01, ** p < 0.05, * p < 0.1. The corresponding t-statistics are in parentheses. Source: Authors’ elaboration.

, the regression coefficients of the explanatory variables in Model (1) and Model (4) are both significantly positive at the 1% level, which indicates that digital economy agglomeration has a positive effect on GTFP.

A spatial regression test is performed. After the global Moran’s I statistic test, LR test, LM test, Wald statistic test, LR statistic test and Hausman test, the spatial Durbin model with two-way fixed effects was finally selected (Due to space limitations, the test results are not listed). Since the direct influence coefficient and spillover influence coefficient obtained by point estimation have a certain bias, the direct effect and indirect effect of explanatory variables are obtained by using partial differential decomposition. According to Model (2), Model (3), Model (5) and Model (6) in Table 2, the spatial autoregressive coefficients of GTFP are significantly positive at the 1% level, which indicates that there is a positive spillover effect of GTFP among cities. The regression coefficients of Dag^a and Dag^b and their spatial lag regression coefficients all passed the significance test, which means that digital economy agglomeration has a positive direct effect and a positive spillover effect on GTFP. The regression coefficient of digital user agglomeration (Dag^a) is greater than that of digital industry agglomeration (Dag^b). That is to say, the former has a greater effect on the promotion of urban GTFP, which indicates that the efficiency growth effect of digital technology is mainly based on the expansion of digital demand at this stage.

Among the control variables, the marginal contribution of human capital (Hcp) and financial development level (Fin) is positive. It shows that the improvement of human capital stock and financial development level helps to improve GTFP. The regression coefficients of both the population density (Pop) and the transportation level (Tra) are significantly negative. It indicates that the scale effect of population agglomeration is offset by the crowding effect, and GTFP is inhibited by the unscientific population distribution. The high-intensity transportation needs of a city cannot be met due to low-support and low-efficiency transportation facilities. The regression coefficient of government intervention degree (Gov) is significantly negative at the 1% level. The reason is that the centralized administrative management system usually restricts the functioning of economic entities, which is not conducive to the market-oriented adjustment of the economic system. The impact of environmental regulation intensity (Reg) on GTFP presents a “positive U-shaped” relationship, which means that only when environmental regulation intensity exceeds the threshold value can companies be stimulated to improve production processes.

(2) Regression results based on subsample. In order to examine the heterogeneity of the impact of digital economy agglomeration on regional GTFP, the samples are divided into eastern region and central and western region, core cities and peripheral cities for analysis. Core cities are cities whose per capita real GDP is greater than the national per capita real GDP during the sample period, while peripheral cities are cities whose per capita real GDP is smaller than the national per capita real GDP during the sample period. The results are shown in

Table 3. Estimated results of the impact of digital economy agglomeration on GTFP with sub-sample.

Variable	Explaining Variable:Daga				Explaining Variable:Dagb
	Eastern	Central and Western	Core Cities	Peripheral Cities	Eastern	Central and Western	Core Cities	Peripheral Cities
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Dag	0.136 *** (10.46)	0.057 *** (4.23)	0.137 *** (10.66)	0.077 *** (5.61)	0.070 *** (5.24)	−0.006 (−0.80)	0.035 *** (3.16)	0.017 (1.48)
Constant	0.754 (0.82)	0.570 (1.55)	1.932 *** (2.75)	0.973 ** (2.17)	2.038 ** (2.20)	0.527 (1.38)	2.629 *** (3.70)	0.854 * (1.76)
Contorl	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
city/year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Adjust R2	0.772	0.683	0.810	0.696	0.756	0.679	0.795	0.690
N	1616	2944	1408	3152	1616	2944	1408	3152

Note: *** p < 0.01, ** p < 0.05, * p < 0.1. The corresponding t-statistics are in parentheses. Source: Authors’ elaboration.

.

From the regression results of digital user agglomeration (Dag^a), the elasticity coefficients in Models (1–4) are significantly positive at the level of 1%. The promotion effect in the eastern region is significantly greater than that in the central and western region, and the promotion effect of core cities is significantly greater than that of peripheral cities. From the regression results of digital industry agglomeration (Dag^b), in Models (5–8), only the regression coefficients of the eastern region and core city are significantly positive, while the negative impact of the central and western region and the positive impact of peripheral cities have not passed the significance test, it indicates that digital industry agglomeration in underdeveloped areas has not effectively improved GTFP. There is regional heterogeneity in the promotion effect of digital economy agglomeration on GTFP. The higher the level of regional economic development, the more obvious the marginal promotion effect of digital economic agglomeration on GTFP. This suggests that regional GTFP disparity is aggravated by digital economy agglomeration, which verifies that the hypothesis H1a holds.

4.2. The Impact of Digital Economy Agglomeration on the Regional GTFP Disparity

(1) Benchmark regression results. According to

Table 4. Estimated results of the impact of digital economy agglomeration on Regional GTFP disparity with sub-sample.

Variable	Benchmark Regression: Daga		Benchmark Regression: Dagb
	Model (1)	Model (2)	Model (3)	Model (4)
observation	4560	4560	4560	4560
Dag	0.154 *** (6.44)	0.177 *** (6.79)	0.056 *** (2.94)	0.057 *** (3.03)
Constant	1.324 *** (8.14)	3.754 *** (3.83)	1.330 *** (5.67)	4.156 *** (3.98)
city/year	Yes	Yes	Yes	Yes
Adjust R2	0.618	0.623	0.611	0.615

Note: *** p < 0.01. The corresponding t-statistics are in parentheses. Source: Authors’ elaboration.

, it can be seen that the regression coefficients of Dag in Models (1–4) are significantly positive regardless of whether control variables are added, which indicates that regional GTFP disparity is exacerbated by digital economy agglomeration. It also means that the craze for ultra-conventional digital construction in China has not achieved the purpose of narrowing productivity differentiation. This has also further verified that the hypothesis H1a holds. Comparing the regression coefficients, it can be seen that the impact of digital user agglomeration on the regional GTFP disparity is greater than that of digital industry agglomeration.

(2) Robustness test. To ensure the reliability of the empirical results, the explanatory variables are replaced or the outliers are removed for robustness testing in this paper. As shown in

Table 5. Robustness test results.

Variable	Robustness Test: Daga		Robustness Test: Dagb
	Model (1)	Model (2)	Model (3)	Model (4)
Dag	0.221 *** (9.94)	0.162 *** (6.93)	0.036 ** (2.21)	0.051 *** (2.80)
Constant	3.767 *** (4.43)	3.039 *** (3.37)	4.376 *** (4.64)	3.410 *** (3.67)
city/year	Yes	Yes	Yes	Yes
Adjust R2	0.749	0.629	0.738	0.622
N	4560	4560	4560	4560

Note: *** p < 0.01, ** p < 0.05. The corresponding t-statistics are in parentheses. Source: Authors’ elaboration.

, the regional GTFP disparity in Model (1) and Model (3) is replaced by the ratio of the urban GTFP to the national average GTFP. The winsorize method is used to perform 1% tail reduction processing on both sides of the explanatory variable in Model (2) and Model (4). The results show that although the regression coefficient values of explanatory variable (Dag) have slight changes, the coefficient sign has not changed and still passes the test. It indicates that the research conclusion that regional GTFP disparity is exacerbated by digital economy agglomeration has good robustness.

5. Influence Mechanism Test

5.1. Influence Mechanism Test Model

Combined with theoretical analysis, it can be seen that regional GTFP disparity may be affected by digital economy agglomeration due to differentiated industrial upgrading mechanisms. Yuan [52] pointed out that industrial upgrading is not only a structural transformation from agriculture to industry or service industry, but also a qualitative transformation from low-skilled industries to high-skilled industries or from low-value-added to high-value-added industries. The level of industrial upgrading is measured by the agglomeration degree of heterogeneous skill industries in this paper. Firstly, Model (5) is constructed to examine the direction and regional differences in the impact of digital economy agglomeration on industrial upgrading, and then Model (6) is constructed to identify the mediating transmission mechanism. The model is expressed as follows:

$$HSLA{G}_{it} or LSLA{G}_{it}={\alpha }_0+\delta Da{g}_{it}+\gamma Contro{l}_{it}+{\epsilon }_{it}$$

(5)

$$\begin{gathered} RG{D}_{it}={\alpha }_0+{\delta }^{1}Da{g}_{it}+\gamma Contro{l}_{it}+{\epsilon }_{it} \\ M={\alpha }_0+\theta Da{g}_{it}+\gamma Contro{l}_{it}+{\epsilon }_{it} \\ RG{D}_{it}={\alpha }_0+\lambda M+{\delta }^{2}Da{g}_{it}+\gamma Contro{l}_{it}+{\epsilon }_{it} \end{gathered}$$

(6)

where HSLAG and LSLAG are the agglomeration degree of high-skilled industries and low-skilled industries. High-skilled industries and low-skilled industries are distinguished according to “whether the proportion of employees in the industry who have received college education or above is higher than 40%”. For details, please refer to Zhang and Zhang [53]. $M$ is the mediating variable (industrial upgrading), which is measured by “the proportion of high-skilled industry agglomeration and low-skilled industry agglomeration”. The mediation effect test needs to meet the following requirements: firstly, the regression coefficient ${\delta }^1$ of digital economy agglomeration (Dag) passed the significance test; secondly, when both $\theta$ and $\lambda$ are significant, if ${\delta }^2$ is significant, it means that $M$ has a partial mediating effect; if ${\delta }^2$ is not significant, it means that $M$ has a complete mediating effect, and when at least one of $\theta$ and $\lambda$ is not significant, it needs to be judged by the Sobel test.

5.2. Influence Mechanism Test Results

Table 6. Estimated results of digital economy agglomeration affecting heterogeneous industrial agglomeration.

Explained Variable	Variable	Explaining Variable: Daga				Explaining Variable: Dagb
		Eastern	Central and Western	Core Cities	Peripheral Cities	Eastern	Central and Western	Core Cities	Peripheral Cities
		(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
HSIAG	Dag	0.158 *** (9.23)	0.084 *** (6.56)	0.220 *** (11.16)	0.090 *** (7.16)	0.082 *** (5.34)	0.014 ** (1.96)	0.045 *** (3.71)	0.019 ** (2.09)
	Constant	1.323 ** (2.01)	2.567 *** (5.06)	0.200 (0.31)	4.325 *** (5.94)	2.814 *** (3.81)	2.487 *** (4.60)	1.336 * (1.88)	4.189 *** (5.19)
	Contorl	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
	city/year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
	Adjust R2	0.838	0.846	0.789	0.828	0.819	0.842	0.726	0.822
	N	1616	2944	1408	3152	1616	2944	1408	3152
LSIAG	Dag	−0.083 *** (−8.65)	−0.042 *** (−5.95)	−0.118 *** (−10.88)	−0.046 *** (−6.74)	−0.049***(−5.72)	−0.010** (−2.50)	−0.024*** (−3.70)	−0.013 *** (−2.57)
	Constant	0.840 ** (2.22)	0.012 (0.04)	1.390*** (4.02)	−0.941 ** (-2.15)	0.062 (0.15)	0.054 (0.17)	0.778 ** (2.06)	−0.860 * (−1.80)
	Contorl	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
	city/year	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
	Adjust R2	0.836	0.840	0.793	0.825	0.821	0.837	0.732	0.821
	N	1616	2944	1408	3152	1616	2944	1408	3152

Note: *** p < 0.01, ** p < 0.05, * p < 0.1. The corresponding t-statistics are in parentheses. Source: Authors’ elaboration.

shows the estimated results of the impact of digital economy agglomeration on heterogeneous industrial agglomeration. From the Models (1–8), the regression coefficients of digital economy agglomeration (Dag) to high-skilled industrial agglomeration are all significantly positive, while the regression coefficients of Dag to low-skilled industrial agglomeration are all significantly negative. It shows that digital economy agglomeration promotes the agglomeration of high-skilled industries but inhibits the agglomeration of low-skilled industries. Comparing the regression coefficients, it can be seen that whether it is the strengthening effect of high-skilled industry agglomeration or the inhibitory effect of low-skilled industry agglomeration, the marginal contribution of the eastern region is greater than that of the central and western regions, and the marginal contribution of core cities is greater than that of peripheral cities. That is to say, the industrial upgrading effect of digital economy agglomeration in developed regions is greater than that in less developed regions.

It can be seen from

Table 7. Mediation test results of regional GTFP disparity affected by digital economy agglomeration.

Variable	Explaining Variable: Daga			Explaining Variable: Dagb
	RGD	M	RGD	RGD	M	RGD
M			0.030 * (1.66)			0.055 *** (1.86)
Dag	0.177 *** (6.79)	0.211 *** (8.11)	0.171 *** (6.48)	0.057 *** (3.03)	0.062 *** (3.96)	0.054 *** (2.87)
Constant	3.754 *** (3.83)	4.364 *** (5.05)	3.622 *** (3.70)	4.156 *** (3.98)	4.863 *** (5.19)	3.888 *** (3.76)
Contorl	Yes	Yes	Yes	Yes	Yes	Yes
city/year	Yes	Yes	Yes	Yes	Yes	Yes
Adjust R2	0.623	0.822	0.623	0.615	0.816	0.616
N	4560	4560	4560	4560	4560	4560

Note: *** p < 0.01, * p < 0.1. The corresponding t-statistics are in parentheses. Source: Authors’ elaboration.

that the following research conclusions are robust: Firstly, regardless of the influence of mediator variables, the regression coefficients of the explanatory variable (Dag) to regional GTFP disparity are all significantly positive; Secondly, the regression coefficients of the explanatory variable (Dag) to the intermediary variables (M) are all significantly positive, which indicates that digital economy agglomeration has a positive effect on regional industrial upgrading; Thirdly, after adding the intermediary variable (M), the impact of industrial upgrading on the regional GTFP disparity is significantly positive; Fourthly, after adding the intermediary variable, the impact of industrial upgrading on regional GTFP disparity is significantly positive, and the impact of the explanatory variable (Dag) on the regional GTFP disparity is also significantly positive, and the absolute value of the regression coefficient of the explanatory variable (Dag) becomes smaller. It shows that industrial upgrading has a partial mediating effect on regional GTFP disparity affected by digital economy agglomeration, thereby it is verified that the hypothesis H2a holds.

6. Discussion

According to Section 4.1, digital economy agglomeration has a significant positive direct effect and a positive spillover effect on GTFP during the sample period. With the improvement of the level of economic development, the marginal improvement effect of digital economy agglomeration on GTFP is greater. This result is consistent with the recent works of Hu and Guo [33] and Liu et al. [35]. According to Section 4.2, digital economy agglomeration significantly exacerbates the regional GTFP disparity. This finding is a useful complement and further extension to the study of Ren et al. [38]. Although digital economy agglomeration can promote the balanced growth of regional GTFP due to sharing effects, allocation effects and diffusion effects, it can also exacerbate regional productivity disparity due to cumulative circular effects, self-proliferation effects and Matthew effects. More importantly, the polarization effect of the latter is greater than the trickle-down effect of the former, which leads to the widening of regional GTFP disparity. The possible explanation is that the regional economic development disparity is significant in China. In contrast, less developed cities lag behind developed cities in terms of labor skills upgrading, physical capital accumulation, infrastructure conditions, and institutional design. This leads to inefficient allocation of resources in less developed regions, as well as limiting the green efficacy of digital economy agglomeration. Table 4 indicates that the impact of digital user agglomeration on the regional GTFP disparity is greater than that of digital industry agglomeration, reconfirming the importance of digital user scale on GTFP. It means that the direction of digital policy support should be more focused on the expansion of digital user scale.

In the influence mechanism analysis, it can be seen that disequilibrium industrial upgrading played a significant mediating role in the relationship between digital economy agglomeration and regional GTFP disparity. The research outcome is similar to the conclusion of Liu et al. [35]. With the accumulation of digital elements, the social resource allocation mode and the production relations of the industrial system are optimized, which is conducive to the improvement of GTFP. However, affected by the differences in regional industrial development, the promotion effect of digital economy agglomeration on industrial upgrading in the core area is greater than that in peripheral areas. The existing studies on industrial upgrading only pay attention to structural adjustment, which is improved in this paper. This study finds that digital economy agglomeration not only has the effect of industrial structure adjustment, but also produces the effect of industrial spatial evolution. Table 6 indicates that the marginal effects of digital economy agglomeration to promote the agglomeration of high-skilled industries and restrain the agglomeration of low-skilled industries in developed cities are significantly greater than those in underdeveloped cities, which further confirms the theoretical views of Li [54]. The flying-geese model of technology space can drive the flying-geese model among industries, but affected by the dependence of technology path, the industrial structure migration cannot effectively narrow the regional economic efficiency disparity [55]. It is necessary to be alert to the phenomenon of “low-end industry lock” in underdeveloped cities.

7. Conclusions and Policy Recommendations

In this paper, the impact of digital economy agglomeration on the regional GTFP disparity is explored from two aspects, including theoretical mechanism and empirical effect. The results show that: Firstly, the GTFP of Chinese cities maintains an upward trend, but the gap among cities is gradually widening. Secondly, digital economy agglomeration has a significant positive direct effect and a positive spillover effect on GTFP and there is heterogeneity in different regions. The higher the level of regional economic development, the more obvious the marginal promotion effect of digital economy agglomeration on GTFP. Thirdly, digital economy agglomeration significantly exacerbates the regional GTFP disparity, and the impact of digital user agglomeration is greater than that of digital industry agglomeration. It means that the craze for ultra-conventional digital construction in China has not achieved the purpose of narrowing productivity differentiation, and digital user expansion has a much greater impact on regional GTFP disparity than digital industry accumulation. Fourthly, the regional GTFP disparity is aggravated by digital economy agglomeration due to disequilibrium industrial upgrading mechanism. Concretely, the marginal effects of digital economy agglomeration to promote the agglomeration of high-skilled industries and inhibit the agglomeration of low-skilled industries in developed cities are significantly greater than those in underdeveloped cities. It means that there is an industrial bottleneck of “low-end lock-in” in underdeveloped areas under the background of digital integration.

Based on the conclusions, the following policy implementation recommendations are given. Firstly, digital economy agglomeration plays a key role in promoting the high-quality development of urban economy, and this promoting effect on economic growth is consistent with the concept of “green development” at the present stage in China. It is necessary to continuously improve the degree of digital resource agglomeration, create an external environment conducive to the sharing of scientific and technological achievements, and strengthen the spillover effect of digital economy agglomeration in improving inter-regional economic quality.

Secondly, the extraordinary boom of digital popularization has not shaken the ranking of inter-city technology level, and the digital gap among cities is still obvious. The coverage density of digital resources in underdeveloped cities needs to be continuously increased, but the direction of policy support should focus on the expansion of digital users rather than the cultivation of digital industries. Based on the strategy of regional coordinated development, the construction of regional network space should be accelerated, which can effectively improve the radiation effect and trickle-down effect of the core region on the periphery region.

Thirdly, considering the spatial heterogeneity of the impact of digital economy agglomeration on industrial structural upgrading and industrial spatial upgrading, it is necessary to implement differentiated industrial optimization strategies according to regional resource conditions and development stages. For underdeveloped cities, the key to industrial upgrading is to realize the high-tech transformation of leading advantage industries in the process of undertaking industrial transfer. With the improvement of technological base and industrial strength, less developed cities can gradually solve the industrial development dilemma of low-end locking. The government should promote the cooperation of digital technology innovation among regions, in order to achieve a balanced regional welfare in which the digital economy drives industrial upgrading and high-quality economic development.

This paper presents a preliminary discussion about the impact of digital economy agglomeration on regional GTFP disparity, but much remains to be done. Firstly, this paper only considers the influence path of industrial upgrading, which can be further expanded from the aspects of technological innovation and international trade in the future. Secondly, this paper only carries out the analysis from the macro perspective. With the improvement of the database, it is necessary to conduct more adequate empirical analysis from the perspective of micro enterprises in terms of research scope.