Identifying the influencing factors of green development is considered the primary factor in improving the efficiency of green development, and numerous scholars have analyzed the influencing factors of green development efficiency from different perspectives and dimensions. Since GTFP is developed on the basis of TFP, theories related to TFP can be transplanted into GTFP.

Emerging technologies, such as big data, artificial intelligence, and the Internet of Things, are driving changes in the mode of economic operation, guiding economic entities to adopt more advanced technologies and modern applications, and forming a different scope, scale, and level of production. Under the current technological and economic paradigm of the digital economy, digital and information technologies have rapidly realized industrialization and marketization and accelerated their penetration into the whole range of economic activities, changing the original mode of production, organization, and management (Du and Zhang, 2021). Therefore, through technological innovation in the ICT sector, the digital economy spreads technology to different production sectors, optimizes production allocation efficiency, and finally, realizes the improvement of macro TFP. Specifically, the mechanism of the digital economy to improve productivity includes the following two aspects.

On the one hand, the digital economy leads to the transformation of production factors and production functions. The popularization of the Internet can improve the real per capita GDP and change the industrial structure (Liu and Chen, 2017). When “data” production factors are added to the production process, production efficiency can be greatly improved through the channel effect of data development and application and data dissemination and sharing (Li and Wang, 2021). On the other hand, the characteristic that the value of “data” increases with the increase in the amount of data can generate the increasing return to scale effect, which expands the production possibility curve and greatly improves output efficiency (Shi et al., 2019; Ding, 2020).

On the other hand, the digital economy promotes technological efficiency and progress. It has been found that the digital economy has greatly contributed to social productivity through high-tech innovations and applications (Nambisan, 2017). The association between the digital economy index and provincial TFP in China was demonstrated by Pan et al. (2022), demonstrating the role of the digital economy as an innovation engine for the broad and sustained development of TFP (Pan et al., 2022). Zhang et al. (2022) found that the digital economy can effectively promote technological progress and efficiency improvement and promote green total factor productivity growth under the coupling effect (Zhang et al., 2022a). The application of ICT can also improve the digitalization of enterprises and governments, thereby increasing productivity and governance efficiency. According to Sadik-Zada et al. (2022), the adoption of electronic government in the delivery of public sector services has been the central factor that has contributed to the reduction of almost all corruption in developing and transition economies. E-government presents one of the greatest opportunities for socio-economic development and offers solutions for improving the efficiency and effectiveness of public administration (Sadik-Zada et al., 2022).

However, considering the environmental constraints, the impact of the digital economy on GTFP is more complex and difficult to judge. It is difficult to judge whether the digital economy promotes or inhibits GTFP. Modern information technology promotes the innovation of urban development modes, which produce technological effects, allocation effects, and structural effects through innovation drives and then reduces urban environmental pollution through the above three effects. However, the ICT industry itself is a high energy consumption industry (Shi et al., 2018). In the early stages of the digital economy, limited resources are devoted to the development of infrastructure, leading to few opportunities for industrial structure optimization. Through increased expenditure on digital devices and infrastructure, as well as the digitization of existing commercial enterprises, the digital economy increases pollution emissions and energy consumption (Wang, 2022). The degree of penetration of the digital economy in different industries was obviously unbalanced. According to Guo and Liang (2021), the siphon effect of talent and capital brought on by digital industrialization hampered technical advancement (Guo and Liang, 2021). According to Zhou et al. (2021), based on the panel data of Chinese cities from 2011 to 2019, the digital economy significantly increased the GTFP of central cities, but the “siphon effect” hindered the green total factor productivity improvement of peripheral cities (Zhou et al., 2021). Xu and Liu. (2023) found that the digital economy and the green economy in central and western China have not yet had a positive interaction (Xu and Liu, 2023). There is a vertical deepening process from the Internet economy to the digital economy, and the dimensions of the digital economy “enabling” economic transformation and green development are more extensive and the mechanism of action is more complex (Guo and Liang, 2021).

While technology embedding in economic transformation based on digital technology development takes a lot of time, with the transformation of the digital economy and industrial structure, future development of the digital economy can improve GTFP through innovative technology, better capacity management, increased worker productivity, resource management, more rational environmental regulations, higher energy efficiency, reduced energy consumption, and lower pollution (Lange et al., 2020; Huang and Lei, 2021). Therefore, the promotion effect of the digital economy on GTFP may have a time-lag effect.

On the basis of these elements, we propose the following theory.

Hypothesis 1:

The digital economy has a nonlinear impact on GTFP. At the same time, the promotion effect of the digital economy on GTFP has a time lag.

Factors that are considered for GTFP include energy input and output, meaning that efficiency of energy usage is crucial for GFTP. Pao andd Fu. (2013) revealed the positive relationship between clean energy and green growth, thereby explaining the importance of promoting clean energy applications (Pao and Fu, 2013). Taskin et al. (2020) also described the positive effect of renewable resources on sustainable development, which is consistent with Pao and Fu. (2013) (Taskin et al., 2020).

Considering that the process of digital industrialization can affect the energy structure, the role of energy transformation in studies of the digital economy’s impact on GTFP cannot be ignored. The digital economy encourages the digital transformation of established sectors and the development of new business models, particularly those with low energy consumption and emissions. The digital economy can also have a positive impact on energy transformation by enhancing government governance, improving the efficiency of traditional energy use, and promoting the generation and consumption of renewable energy (Shahbaz et al., 2022). Fan et al. (2022) demonstrated that the development of new digital infrastructure has a positive effect on China’s energy restructuring (Fan et al., 2022). Wang et al. (2023) found that RETI (renewable energy technological innovation) significantly improves GTFP in China (Wang et al., 2023). According to some data, the digital economy may improve energy efficiency, lower CO2 emissions, and alter the composition of energy consumption (Rodríguez Casal et al., 2005).

It can be seen that the digital economy may affect GTFP through energy transition; however, few studies have discussed the mediating role of energy transition in the past.

On the basis of these factors, we put forth the following theory.

Hypothesis 2:

Energy transition plays a mediating role in the digital economy–GTFP nexus.

As the aforementioned study demonstrates, the digital economy can contribute to energy transition, and we consider two aspects of energy transition: the structure of the use of renewable energy and the structure of the production of renewable energy.

The development of the digital economy not only accelerates the spread of information and reduces the cost of information flow but also creates new resources and promotes knowledge sharing. The digital economy can break the traditional time and space constraints, making closer ties between regions and countries, thus improving labor efficiency, productivity, and management locally, and even crossing borders, presenting spatial spillover effects from digitization to GFTP. Furthermore, according to the theory of technology diffusion and the new economic geography, geographical proximity may be an important factor influencing the effect of the digital economy (Zhao et al., 2022a; Shahbaz et al., 2022).

At the same time, the Internet may exhibit different effects on socio-economic and green development in regions with different levels of economic development and different locational conditions. Countries differ greatly in terms of available resources and capabilities for designing e-government strategies and measures. A country’s e-government development plans may not necessarily benefit from government spending or economic growth due to internal political, social, and bureaucratic issues (Niftiyev, 2022b). Due to historical, cultural, and religious reasons in countries along the Belt and Road, there may also be spatial heterogeneity in the impact of the digital economy on GTFP in these countries. Niftiyev (2022) analyzed three countries in the South Caucasus (Armenia, Azerbaijan, and Georgia), which are also Belt and Road Initiative (BRI) countries, and found that the three countries have different growth speeds in the ICT sector and digitalization levels, resulting in differences in manufacturing labor productivity. This has also led to differences in China’s foreign economic interests in these countries (Niftiyev, 2022a). However, this point is typically ignored by academics.

We propose Hypothesis 3 according to the above study.

Hypothesis 3:

The spatial spillover effect of digital economy on GTFP is statistically significant.

In general, our research focuses on three mechanisms. (Figure 1).

The major dependent variable in this study is green total factor productivity (GTFP), whereas the key independent variable is the digital economy (DE), with the aim of examining the link between these two variables. This study introduces a series of control variables to control for the impact of macroeconomic factors. Due to the fact that panel data is selected for empirical analysis, we construct a multivariate framework as Eq. 1: G T F P i , t = α 0 + α 1 D E i , t + ∑ 3 7 α k C O N i , t + μ i + γ t + ε i , t (1) where subscripts i represent countries and t represent years; The explanatory variable GTFP is green total factor productivity, DE is the digital economy indicator, and CON stands for a vector that contains the control variables; μ _i denotes individual fixed effects of countries i that do not vary over time; γ _t refers to time fixed effects; ε denotes random disturbance terms; and α ₀ is an intercept term, α ₁ and α _k are the coefficients for DE and CON, respectively.

Since many studies have shown that there may be a non-linear relationship between the digital economy and green development, in order to verify Hypothesis 1, we run the regression in two ways.

First, a quadratic term for the level of development of the digital economy DE ² is added to the previous linear model and, thus, can be transformed into a regression model, given as Eq. 2. G T F P i , t = α 0 + α 1 D E i , t + α 2 D E i , t 2 + ∑ 3 7 α k C O N i , t + μ i + γ t + ε i , t (2)

In addition, according to previous studies, digital technology needs time to spread, which means the digital economy may have a time lag; therefore, the lag period of DE is added to Eq. 2.

In the previous discussion of the transmission mechanism, we discussed how the digital economy may affect GTFP through energy transition. The indirect effect of the independent variable on the dependent variable through the intermediate variable is called the mediating effect (Mackinnon et al., 2000). To test whether the above factors can play the role of mediating variables, this paper adopts a standardized intermediary effect model to carry out further empirical investigation. Specifically, to verify Hypothesis 2, the indirect influence of the explanatory variable (X) on the explained variable (Y) through the intermediate variable (M), the following Eqs 3–5 are used: Y = α X + ε 1 (3) M = β X + ε 2 (4) Y = α ′ X + γ M + ε 3 (5)

This paper regards GTFP as the explanatory variable Y. Renewable energy generation (REG) and renewable energy consumption (REC) are regarded as intermediary variables M to be tested, and the DE is regarded as an explanatory variable X to construct the intermediary effect model.

Moreover, spatial panel econometric models are usually used in empirical studies of regions because they can take into account the inherent properties of the regions themselves and their spatial linkages. To verify Hypothesis 3, the underlying panel regression is extended to a spatial panel Durbin model (Eq. 6). G T F P i , t = α 0 + ρ W G T F P i , t + ϕ 1 W D E i , t + α 1 D E i , t + ϕ 2 W D E i , t 2 + α 2 D E i , t 2 + ϕ 3 W C O N i , t + α 3 C O N i , t + μ i + δ t + ε i , t (6) where: ρ represents the spatial autoregressive coefficient; W is the spatial weight matrix; ϕ 1 , ϕ 2 , and ϕ 3 are the elasticity coefficients of the primary and secondary terms of the digital economic development level and the spatial interaction terms of the control variables.

In model estimation, the variables are standardized from 0 to 1 to avoid the effect of different magnitudes on the results. In order to accurately estimate the interrelationship between the digital economy and green development, it is necessary to select the most appropriate model among different types of spatial panel econometric models for parameter estimation, i.e., combining LM, Robust LM, Wald, and other statistics with the Hausman test for judgment and selection (Jiang, 2016).

Where, if ϕ 1,2,3 = 0 , the spatial Durbin model can be reduced to a spatial lag model (7); if ϕ 1,2,3 + ρ α 1,2,3 = 0 , the spatial Durbin model can be reduced to a spatial error model (8). G T F P i , t = γ W G T F P i , t + α 0 + α 1 D E i , t + α 2 D E i , t 2 + α 3 C O N i , t + μ i + δ t + ε i , t (7) G T F P i , t = λ W G T F P i , t + α 0 + α 1 D E i , t + α 2 D E i , t 2 + α 3 C O N i , t + μ i + δ t + ε i , t (8)

Where: γ denotes the degree of influence of the GTFP of neighboring countries in the previous period on the GTFP of the region; λ denotes the spatial dependence effect of the GTFP of countries.

Table 4 demonstrates the correlation coefficients between the variables, showing that variables are low correlated, thus having a lower probability of multicollinearity.

While the data have been analyzed using descriptive statistics, there is a need for a slope heterogeneity analysis. The research factors in panel data analysis may be impacted by a variety of information, including social, economic, or technological information. Therefore, before starting the estimation procedure, slope heterogeneity and cross-sectional correlation tests should be used. The results for slope heterogeneity are shown in Table 5. The slope coefficient test’s statistical value is significant at the 1% level, rejecting the null hypothesis of homogeneity. The findings indicate that the variables are heterogeneous, necessitating a cross-sectional dependence test.

When analyzing the relationship between all variables in panel data models, cross-sectional correlation is a key issue that needs to be considered; ignorance of cross-section dependence may cause substantial estimation bias and size distortions (Pesaran, 2007). Thus, before assessing the stationary nature of the variables, this analysis first examines the presence of any potential cross-sectional dependence in the panel.

The Pesaran test, the Friedman test, and the Frees test are applied, and the results are presented in Table 6. As shown in the table, the statistics for the Pesaran test and Frees test significantly reject the null hypothesis at the 1% significance level, and the statistics for the Friedman test significantly reject the null hypothesis at the 5% significance level, which provides strong evidence for the cross-sectional dependence among B&R countries. Therefore, in further analysis using this panel sample, we use estimation techniques that allow for cross-sectional dependence.

To prevent erroneous regression, panel unit root tests must be carried out prior to parameter estimation in the panel data model. The first-generation conventional panel unit root tests, including the Levin-Lin Chu (LLC), Im, Pesaran, Shin (IPS), augmented Dickey-Fuller (ADF), and Phillips-Perron (PP) tests, are not appropriate due to the presence of cross-sectional dependency in the panel data (Dou et al., 2021). As a result, the Pesaran cross-sectionally augmented IPS (CIPS) test and the cross-sectionally ADF (CADF) test, two second-generation panel unit root tests that take into account cross-sectional dependence, are more applicable in this investigation (Pesaran et al., 2007). The stationarity and level of integration of the variables are therefore examined in this study using the CADF test. Table 7 displays the results of the stationarity test.

It can be seen from Table 7 that not all variables I (0) are stationary, but the null hypothesis of unit root is significantly rejected in all first-order differences. Thus, our choice of variables is order-one stationary, providing the conditions for us to perform the cointegration test. The Westerlund ECM Cointegration Test is employed in the study to achieve this. The test results in Table 8 show the rejection of the null hypothesis. The significant p-values indicate the presence of long-term associations between variables indicating DE, and the control variables are associated with GTFP in B&R countries, illustrating long-term equilibrium relationships among the selected variables. Therefore, the following estimate of the DE-GTFP nexus is reliable and valid.

After the discussion of data stationarity and cointegration, this paper conducts an empirical analysis of the DE-GTFP nexus by estimating Eq. 1. Table 9 shows the baseline regression results. For the OLS method, the Hausman test results suggest that the fixed effect model should be used. If the OLS estimation is still used when there are unit roots and cointegration relationships among the variables, although the OLS super-consistent estimator will be obtained, the asymptotic distribution of the OLS estimator is non-standard and is affected by noise parameters, which cause the commonly used test procedures to fail. In order to avoid possible endogeneity problems among variables, Phillips and Hansen (1990) suggested using non-parametric methods to modify OLS estimators and proposed the fully modified least squares method (FMOLS) for time series (Hansen and Phillips, 1990). On this basis, Pedroni (2001) proposed FMOLS estimation for panel data, including within-dimension FMOLS estimation and between-dimension panel estimation (Pedroni, 2001). The two methods are also compared, and it is found that the inter-group panel FMOLS has better small-sample properties and flexible condition setting than the intra-group FMOLS, so we use the inter-group panel FMOLS for estimation. In order to correct the cross-sectional dependence problem, we also adopt the PCSE estimation method. Regressions in columns (1), (3), and (5) show the estimated coefficients of the digital economy (i.e., DE) based on the OLS, FMOLS, and PCSE methods are consistently negative, indicating a monotonically decreasing relationship between the digital economy and green total factor productivity (GTFP).

The findings also support the validity and reliability of the theoretical derivation, which is supported by our theoretical expectation and Hypothesis 1, given in Section 3. The estimated coefficients for DE differ because two cross sections are dropped from the FMOLS estimation because of the missing data. Specifically, in columns (1), (2), (3), and (5), a 0.1 decrease in DE will cause GTFP to decline by approximately 0.04. Columns (4) and (6) present the estimation results for the nonlinear model of Eq. 2 and show that the coefficient of the quadratic term of DE is significantly positive, which proves that DE has a U-shaped relationship with GTFP, which is in line with the Kuznets curve. By calculating the marginal effects in column (6), it can be concluded that the turning point nearly occurs at DE equal to 0.6. The development of the digital economy has all the time inhibited the improvement of GTFP in B&R countries. While the level of the digital economy is below 0.6, as DE increases, the negative effect of DE on GTFP is increasing. With the development of the digital economy, when it reaches a certain level, for example, 0.6, the inhibition effect of the digital economy on GTFP will start to diminish. This conclusion is also in line with the existing literature on the study of the Internet, digital economy, and economic efficiency (Guo and Liang, 2021). In the meantime, over the sample period of the sample countries, although the coefficient of the quadratic term of DE is significantly positive, the total contribution of the digital economy to GTFP is consistently negative. Our findings are also in line with Zhao et al. (2022), who proved that the coefficient for the effect of digital economy development on green total factor energy efficiency (GTFEE) is significantly negative (Zhao et al., 2022a). There are two plausible explanations for these findings. First, massive investment in the construction of digital infrastructure such as network sites and data centers, as well as matching digital technologies, drives investment in ICT equipment manufacturing, chips, steel, optical fiber, and other industries, thus increasing energy consumption. Agricultural and land surveys have been carried out with the help of the Beidou navigation system, while projects such as digital connectivity, digital railways, ports, roads, energy, and water resources have been rapidly developed, driving power consumption up.

As for the control variables in column (4), the estimated coefficients of industry structure (i.e., lnService), Urbanization level (i.e., lnUrban), industry concentration (i.e., lnIC), and GDP per capita (i.e., lnRGDP) are all significant, and their signs mostly coincide with those 38 B&R countries’ actual conditions (as two cross-sections are dropped). Specifically, the GDP per capita has a negative impact on the growth rate of GTFP in all the columns of Table 9. The results may be explained in the following ways: rapid economic expansion is typically accompanied by high energy consumption, which has a positive impact on the rise in household CO₂ emissions. These results are consistent with previous studies (Nasir et al., 2019; Pham et al., 2020; Shahbaz et al., 2020; Nasir et al., 2021). Large CO2 emissions lower the GTFP of the country. As rural residents rely heavily on coal consumption (Dou et al., 2021), the urbanization process contributes to the centralized utilization of energy, which can effectively improve energy utilization efficiency and, thus, improve GTFP. Since the energy consumption per unit output value of industry exceeds that of the service industry, and the scale effect brought by industrial concentration can improve the utilization efficiency of energy, capital, and labor, the increase in the proportion of tertiary industries (i.e., lnService) and the increase in the concentration of secondary industries (i.e., lnIC) can both lead to the improvement of GTFP.

In the early days of the digital economy, investment in digital infrastructure required the consumption of limited local resources. Moreover, the optimization of the production and organization modes of the secondary industry is often ignored in the early stages of the development of the digital economy. In addition, the “enabling” of the digital-based economy requires a certain amount of time for technological precipitation and penetration so as to play the role of the digital economy in optimizing the industrial structure and production mode. These factors may together lead to the reverse effect of the early development of the digital economy on GTFP. On the other hand, due to the “digital” nature of the digital economy, it is usually reflected in the improvement of the efficiency and output value of the tertiary industry, especially in the aspects of platform economy, digital currency, digital finance, etc. Although high efficiency in these areas contributes to economic development, it has a limited effect on energy conservation and emission reduction. In order to further explore the lagged effect brought by DE and confirm the research conclusions, we conducted regression with lagged terms. We find that the three-order lag term of DE has a significantly positive effect on GTFP, and the result is listed in column (3) of Table 10. This suggests that there is a significant lagged effect from DE on GTFP, supporting Wei and Hou (2022), who also reached the same conclusion (Wei and Hou, 2022). Figure 4 shows the intuitive trend. It has been shown that the improvement of GTFP depends on the level of environmental regulations and the optimal allocation of resources among secondary and tertiary industries, and the penetration of “digital” in these aspects usually requires a longer “enabling” process (Zhang et al., 2022c).

Muhammad Shahbaz et al. (2022) found that the digital economy positively affects energy transition; therefore, we explore the mediating effect of energy transition on the digital economy–green efficiency relationship. Following Shahbaz et al. (2022), the renewable energy consumption structure and the renewable energy generation structure are used as proxies for energy transition in this article, including data from the EIA for renewable energy consumption (REC) and renewable energy generation (REG) (Shahbaz et al., 2022). Table 6 presents the mediating effects results.

Columns (1) and (2) of Table 11 present that the digital economy can significantly raise REC and REG, meaning that the digital economy promotes energy transition in the sample countries. While columns (3) and (4) show that energy transitions can significantly contribute to the growth of GTFP, the results mean that B&R sample countries have been undergoing energy transitions with the help of the digital economy, and the decline of GTFP brought on by the digital economy may be partially offset by the intermediary effect and become a nonlinear relationship. Table 11 verifies Hypothesis 2.

According to Muhammad Shahbaz et al. (2022), countries or regions with more mature renewable energy transition, the digital economy will contribute more obvious effects on that transition. Developed countries are more experienced in improving the application scope of renewable energy through digital technology. However, most B&R countries are less developed, which means that they are less skilled at energy transition and are not technologically advanced. Moreover, they use fewer renewable energies, therefore having a limited ability to raise GTFP. When the digital economy plays a highly vital role in improving industrial production efficiency and raising the economy, those less experienced B&R countries then have additional carbon emissions, which have a negative total effect on GTFP.

Moran’s I test is used to analyze the spatial distribution of GTFP.

The global Moran’s I index is calculated by Eq 10. M o r a n ′ s I = n ∑ i = 1 n ∑ j = 1 n W i j y i − y ¯ y j − y ¯ ∑ i = 1 n ∑ j = 1 n W i j ∑ i = 1 n y i − y ¯ 2 (13)

Where y _i denotes the value of the indicator for country i, n is the total number of countries in the sample, and W _ij is the weight matrix. Moran’s I is within the range of [-1, 1], and a value larger than zero implies a positive geographical correlation, while a value less than zero indicates a negative spatial correlation, and a value equal to 0 indicates no spatial correlation. In this paper, we use the matrix of inverse bilateral distance between countries (Wang, 2013).

The values of Moran’s I are shown in Table 12.

It can be seen that Moran’s index of GTFP of countries along the Belt and Road is significant in 2011, 2015, 2019, and 2020, which proves that there is not always spatial correlation in the GTFP of countries along the Belt and Road; however, there is a significant spatial correlation between the variables of the digital economy from 2006 to 2021. Therefore, this paper discusses the spatial spillover effect of the DE-GTFP nexus.

The LM, Robust LM, Wald, and Hausman tests suggest that the SDM model should be used. In column (3) of Table 13, the SDM model estimation results prove that the coefficient of DE is significantly negative and the coefficient of DE2 is significantly positive. In addition, within the range of sample values, the total effect of the digital economy on GTFP is negative, which is consistent with our benchmark regression results. Urbanization is significantly positive, while GDP per capita is significantly negative, which is also consistent with our benchmark regression.

The direct and indirect effects are shown in Table 14. The regression results prove that DE has both a significant direct effect and a significant indirect effect. That is, the development of DE will cause the decline of domestic GTFP and the rise of GTFP in neighboring countries through spillover effects. This inconsistent result may be because it takes time for digital technology to spread. On the other hand, the majority of infrastructure investment is borne by the home country, so other countries can enjoy the benefits of digital progress without having to bear a large cost. As the foundation of the digital economy, the advancement of Internet technology has increased the flow of information, cut the cost of information transmission, and considerably reduced the spatiotemporal distance between regions. The increased usage of Internet technology has boosted management efficiency, broadened the market, and improved the structure of energy use. Therefore, the development of the digital economy promotes the improvement of GTFP in neighboring countries by improving the quality of innovation and upgrading the industrial structure.

The same phenomenon has been captured in previous studies (Su et al., 2021; Zhao et al., 2022a).

As can be seen in Table 15, the contribution of the digital economy to GTFP is significantly negative in European countries and ASEAN (Association of Southeast Asian Nations) countries, while the effect is positive but not significant in Asian countries other than ASEAN countries (mainly Central, South, and West Asian countries).

For high- and middle-income countries, DE progress brings a decline in GTFP, while in low-income countries, the coefficient of DE is not significant (see Table 16). At the same time, we can see that for middle-income countries, the negative impact of the digital economy on GTFP is greater than that of high-income countries, which is consistent with our previous analysis of industrial stages and household consumption in different countries.

The classification of countries can be found in Appendix Tables A1; Table B1. This paper uses the World Bank’s annual classification of national income levels to classify the income levels of countries along the Belt and Road. The national income classification changes dynamically from year to year.

Table 17 presents the robustness tests. Since there are numerous ways to measure the digital economy, this paper replaces the digital economy index with the Online service index in the UN e-government report. The Online service index is based on an overall synthesis of service delivery, technology, institutional frameworks supporting e-government development, content delivery, and e-participation. Online service indexes can also reflect the development of the digital economy (Zhang et al., 2022b). The regression results prove that the model and conclusions of this paper are robust.

Meanwhile, in order to exclude the influence of COVID-19, this paper also conducts a regression on the data from 2006 to 2019, and the regression results show that our conclusions still hold.

In addition, this paper uses the dynamic least squares (DOLS) method to test the robustness of the benchmark empirical models. Columns (5)–(6) in Table 15 show the results. The DOLS regression coefficients have the same direction and have similar values with FMOLS, which also proves that our benchmark regression model is reliable.