Explained Variable

The super-SBM undesirable model is employed to evaluate China’s GIE in this paper. The model involves 3 types of variables simultaneously: input variables, and desirable and undesirable output variables. Input variables mainly include labor, capital, and energy. Hence, in this paper, input variables include R&D expenditure, R&D personnel, and energy consumption. We select intramural expenditure on R&D (10000 yuan) to represent R&D expenditure and choose the full-time equivalent of R&D personnel (man-year) to denote R&D personnel. Also, electricity consumption (100 million kWh) is selected as a proxy for energy consumption. In addition, desirable output variables consist of 3 parts: patent applications, green patent grants, and sales revenue of new products, considering both quantity and quality. We choose domestic patent applications (piece), green invention patent grants (piece), and sales revenue of new products of industrial enterprises above the designated size (10000 yuan) as proxies for desirable outputs, respectively. Furthermore, undesirable output variables are composed of 3 parts: wastewater discharge, waste gas emissions, and solid waste generated. In this paper, the chemical oxygen demand discharge in wastewater (10000 tons), sulfur dioxide emissions in waste gas (10000 tons), and common industrial solid wastes generated (10000 tons) are chosen to represent the undesirable output variables.

Explanatory Variables

As mentioned previously, GIE is influenced by many factors from various aspects. Hence, considering the availability and applicability of data, referring to [15, 19, 22, 23, 32], we choose economic development, marketization level, industrial structure, digital economy, government financial support, and trade openness as the influencing factors and explore the heterogeneity of their effects on GIE from temporal and spatial perspectives. Specifically, GDP per capita (GDPPC, 10000 yuan) is selected as the proxy of economic development; the digital financial inclusion index (DFI) is applied to represent the digital economy. Besides, the proportion of R&D expenditure to GDP (RDE) is used to denote government financial support, and the marketization index (MI) is chosen to represent the marketization level. Moreover, the ratio of the output value of the tertiary industry to the output value of the secondary industry (INS) is used to measure industrial structure, and the proportion of total import and export trade in goods to GDP (IET) is applied to measure trade openness.

The Super-SBM Undesirable Model

The super-SBM undesirable model has been demonstrated to be superior to the traditional DEA models in many research areas. It takes into account both undesirable outputs and allows for more in-depth comparisons of effective DMUs. Hence, it was adopted to evaluate China’s GIE in this paper. Referring to [33], the super-SBM undesirable model can be constructed as follows:

where ρ represents China’s GIE. The parameter λ represents the constant vector. Besides, x_j^t, y_j^t, and b_j^t denotes inputs and desirable and undesirable outputs, respectively. In addition, m, l, and h represent the number of inputs, desirable outputs, and undesirable outputs, while s_m^x, s_l^y, and s_h^b are the corresponding slack vectors, respectively.

Spatial Markov Chain Model

When exploring the distributional evolution of variables, as an extension of the traditional Markov chain model, the spatial Markov chain model can demonstrate the dynamic evolutionary characteristics of variables and explore spatial spillover effects [34]. Currently, the spatial Markov chain model has been widely employed in various academic fields, such as air pollution [35], urban health development efficiency [36], and urban resilience [37]. Hence, a spatial Markov chain model is used in this paper to explore the dynamic evolutionary characteristics of GIE. Besides, before executing the spatial Markov chain model, we first divided GIE into 5 levels based on the natural breaks (Jenks) method, i.e., the levels were employed to represent GIE of different provinces in different years. In the traditional Markov chain model, p_ij is usually used to denote the probability that this stage is in rank i, and the next stage transitions to rank j. Therefore, referring to Du et al. [34], p_ij(k) is used to represent the probability that this stage is in rank i, and the next stage transitions to rank j while the adjacent regions are in rank k at this stage. Moreover, the spatial weight matrix involved in the spatial Markov chain model is generated on the basis of the queen adjacency principle in this paper.

Geographically and Temporally Weighted Regression

Traditional econometric methods, such as geographically weighted regression and spatial econometric models, ignore the role of temporal or spatial factors in the analysis. In order to account for both temporal and spatial nonstationarity and to compensate for the shortcomings of traditional econometric methods, Huang et al. [38] proposed the GTWR model to analyze house prices. Currently, the GTWR model has been widely applied in the fields of environmental science [39, 40] and economic analysis [41, 42], etc., and has been demonstrated to be superior to other traditional econometric models. Thus, referring to Chen et al. [43], the GTWR model can be constructed as follows:

where u_i, v_i, and t_i represent the longitude coordinate, latitude coordinate, and specific year, respectively; β_i denotes the constant and regression coefficients, while ε_it is the error term.

Evaluation Results of GIE

The super-SBM undesirable model is employed to evaluate each province’s GIE in China during 2011-2022 in this paper. The evaluation results of GIE in China by province are shown in Table 2. The lowest GIE was 0.137 in Inner Mongolia in 2011, while the highest GIE was 1.269 in Hainan in 2016. The latter is 9.263 times higher than the former, showing a very significant difference in GIE among provinces. In addition, GIE in most of the provinces does not show a consistent and clear trend, which fully illustrates the complexity of the evolution of GIE. This finding is consistent with Huang et al.’s [10] study that GIE in Hainan is almost always above 1.000, while the vast majority of China’s provinces have GIEs below 1.000, i.e., a state of inefficiency, and GIEs in less developed regions are relatively low.

Spatiotemporal Evolution of GIE

The temporal evolution of GIE in China by region is shown in Fig. 1. Ordinarily, China’s economic zones, which contain 31 provinces (excluding Hongkong, Macao, and Taiwan), can be divided into 4 regions: the eastern, central, western, and northeastern regions. The trends in the temporal evolution of the nationwide GIE and GIE in the individual regions remain highly consistent. For each region, the year 2016 is a critical juncture for significant changes in GIE. The nationwide GIE exhibits 3 distinct phases in the time dimension: 2011-2014, 2014-2016, and 2016-2022. GIE at these 3 phases successively shows a steady, sharp rise and fluctuating upward trend, respectively. In the long run, such a trend reflects the continued improvement of green innovation in China. This is consistent with the conclusions of Zhang et al. [25], which concluded that China’s GIE has generally shown an increasing trend through time. The temporal evolution of GIE in the western region is strongly characterized by similar characteristics as the nationwide GIE, but the former always remains at a relatively low level after 2019 and does not exceed 0.5 until 2022.

In the eastern region, GIE is consistently and significantly higher than GIE in other regions, except in 2011. This finding confirms the conclusions of Zhao et al.’s [14] study that the eastern region has the highest average GIE. This is due to the fact that the eastern region has the best economic base and the highest level of technology, as well as being a magnet for high-level talents from various regions. After 2016, GIE in the eastern region has been slowly and intermittently rising on the whole, and its dominance has become more pronounced, i.e. the phenomenon of the Matthew Effect has emerged. Additionally, the temporal evolution of GIE in the central region can be divided into 3 stages: 2011-2016, 2016-2019, and 2019-2022, with an accompanying rising-falling-rising characteristic. The economic base of the central region is second only to that of the eastern region, and it is also attractive to high-level technical talents. Simultaneously, the central region is endowed with abundant resources, such as minerals, and GIE may be further substantially enhanced in the process of industrial green transformation. Fig. 1 also shows the significant potential for the central region to increase its GIE beyond 2019, which is different from the slow rise in the western region. This finding is in contrast to that of Zhao et al. [14] who concluded that the difference in GIE between the central and western regions is decreasing. One possible explanation is that the latter only calculates GIE up to 2020, and the regional division criteria for China are not consistent with this paper. The temporal evolution characteristics of GIE in the northeastern region are more complex compared to other regions. GIE underwent a downward-upward trend with a clear U-shaped pattern before 2016. By contrast, after 2016, GIE in the northeastern region has trended downward with significant fluctuations. The mass outflow of advanced technicians from the northeastern region and the large number of highly polluting heavy industries have inhibited local green innovation.

From a spatial perspective, Hainan has the highest average GIE of 1.056, the only province in the country with a GIE that exceeds 1. Recently, Hainan has been capitalizing on its natural tropical monsoon oceanic climate and island advantages to develop tourism resources, reduce heavily polluting heavy industries, and enhance the advantages of technological innovation in terms of greenness and sustainability. As indicated in Table 2, the average GIE is followed by Tianjin, Zhejiang, Shanghai, and Beijing, with the former two exceeding 0.9 and the latter two exceeding 0.8. Of the 10 provinces in the eastern region, all 7 except Shandong, Fujian, and Hebei are ranked in the top 10 of GIE, which further confirms that GIE in the eastern region is much higher than in other regions as depicted in Fig. 1. Among the 10 provinces at the bottom of the rankings, Hebei belongs to the eastern region, Henan and Shanxi belong to the central region, Heilongjiang and Liaoning are from the northeastern region, while the other provinces are from the western region. The GIEs of these 10 provinces are below 0.45. The average GIE of Inner Mongolia is the lowest nationally, reaching only 0.261, accounting for only 24.716% of Hainan’s GIE. While Inner Mongolia’s average GIE is consistently the lowest nationwide until 2019, it is rising significantly, even surpassing 13 provinces to reach 0.518 in 2022.

Fig. 1.

The temporal evolution of GIE in China by region.

Traditional Markov Chain Model

The transition probability matrix of GIE based on the traditional Markov chain model is shown in Table 3. With the exception of those located at level 4, the GIEs of the provinces located at other levels in this stage remain stable, i.e., the probability of not experiencing a transition in the next stage exceeds 0.5. Provinces located at level 4 in this stage would jump to level 5 in the next stage with a probability of 0.591. Besides, there is a clear trend of upward transfer of level in the majority of provinces, but also a risk of level decline in some individual provinces, especially those whose GIE is at a high level at this stage. Overall, there is much more upward level transition than a downward level. Also, some of the transition probabilities away from the diagonal are not equal to zero, implying that there is a transition of GIE across levels in some provinces. This clearly provides a factual basis for those provinces that have taken appropriate measures with the expectation that GIE will be upward transfer across levels in the short term. Similar to this finding, the conclusions of the study by Xin et al. [44] confirm that green innovations across provinces stabilize to maintain the initial state. The difference, however, is that the latter confirms that there is no overstepping transfer in the level of green innovation in the provinces. This may be due to the fact that only four levels of provincial green innovation were categorized in Xin et al.’s [44] study.

Spatial Markov Chain Model

In comparison to the traditional Markov chain model, the spatial Markov chain model fully takes into account the spatial spillover effect due to geographic proximity and spatial lag conditions. In Table 4, the transition probability matrix of GIE is generated based on a spatial Markov chain model. Under different spatial lag conditions, the spatial Markov transition probability of GIE is different from the transition probability in Table 3, indicating that the transition probability of the level of GIE in the local region to the next stage is affected by the level of neighboring regions. Similarly, there are differences in the spatial Markov transition probabilities for GIEs at the same level under different spatial lag conditions. For instance, the probability that the local region has a GIE rating of level 2 and moves to level 3 in the next stage is 0.083, 0.146, 0.239, 0.077, and 0.000 for different spatial lag conditions, respectively. Additionally, in contrast to Table 3, there is still a high probability that a region’s GIE will remain at its original level or jump upwards after accounting for spatial factors. For instance, when GIE of the local region and neighboring regions are at the same level in the current stage, the probability that GIE of the local region would remain at the same level in the next stage is 0.742, 0.854, 0.621, 0.100, and 0.778, respectively. Also, the level of GIE in the local region is influenced by neighboring regions, and there are relatively few circumstances in which there is a jump across levels and a transition downward. In some regions, the probability of a downward shift in the level of the local region’s GIE is relatively high when adjacent to a province with a low level; conversely, the probability of an upward shift in the local region is also relatively high when adjacent to a province with a high level. The fact that the probability p₂₃ (0.167) is less than p₂₁(1) (0.250) and that p₂₃(3) (0.239) is greater than p₂₃ (0.167) proves this conclusion. To summarize, these findings confirm Xin et al.’s [44] conclusions that the state of China’s provincial green innovation is significantly influenced by geospatial patterns.

Overall Characteristics of Influences

The GTWR model is adopted to explore the spatiotemporal heterogeneity of influences from factors on China’s GIE in this paper. The GTWR model has the values of adjusted R², AICc, and residual sum of squares of 0.817, -312.122, and 4.643, respectively, both of which are significantly superior to the global OLS regression, geographically weighted regression model, and temporally weighted regression model. In regression results on the basis of the GTWR model, there is a separate regression coefficient for each variable for each province at each time point. Table 5 depicts the descriptive statistics of regression coefficients based on the GTWR model. As shown in Table 5, in general, GDPPC, MI, and INS promoted GIE, while DFI, RDE, and IET inhibited GIE. In Table 5, the average of the regression coefficients of RDE on China’s GIE is -0.032, indicating a negative effect of the former on the latter. One possible explanation is the relatively great differences in economic development, infrastructure, and resource endowment among provinces, leading to differences in the impact of R&D expenditure on GIE. There are 8 provinces where RDE consistently shows a positive effect on GIE, which are Henan, Anhui, Jiangsu, Hebei, Shanxi, Shanghai, Zhejiang, and Shandong. In comparison, in other provinces, the impact of RDE on GIE is sometimes negative, and in 8 of them, it is consistently negative. Geographically, the 8 provinces in which RDE exhibits a positive effect are spatially closely connected, showing obvious clustering characteristics. The 8 provinces that are consistently negative are mainly located in the less economically developed regions, i.e., the northeastern and western regions of China. These findings are different from Li et al. [22], possibly because we embed both temporal and spatial factors in the regression model, which is clearly a more scientific and rational treatment. Indeed, there are also conclusions from many studies, such as Song et al. [19], that support the findings of this paper that there are significant regional differences in the impact of influencing factors on China’s provincial GIE. Therefore, an in-depth spatiotemporal heterogeneity analysis of impacts is necessary.

Temporal Heterogeneity of Influences

Fig. 2 displays the temporal heterogeneity of factors’ influences on China’s GIE. With the exception of 2021, the positive influence of GDPPC on GIE diminishes over time, from 0.087 in 2011 to 0.017 in 2022. Similarly, INS has consistently demonstrated a positive influence in any given year and has shown a V-shaped trend, with 2020 being the turning point. Moreover, with the exception of 2019, the impact of MI on GIE has also been consistently positive and shown a V-shaped trend. These results indicate that, on the one hand, the enhancement of the GIE is influenced by more and more factors, so that the influence of some traditional factors is gradually weakening. On the other hand, due to the impact of COVID-19 epidemiology, the marketization level and the tertiary sector are becoming increasingly critical to GIE.

Besides, the direction of DFI’s effect on GIE appears to alternate negatively and positively. However, the negative effect of DFI on GIE tends to be close to zero, implying a very weak causal relationship between DFI and GIE. This is different from the findings of Li et al. [22] for the reason that Li et al. [22] focused on analyzing the influence of the digital economy on the industrial GIE, whereas this study does not differentiate between industries. According to Liang et al. [24], the digital economy contributes to sustainable innovation, but the adequacy of the digital economy infrastructure significantly affects the full development of digital green innovation. Obviously, in China, the digital economy serves more as a crucial infrastructure to improve the efficiency of social operations and production, while its impact on advanced green technologies is minimal. This implies that China’s provinces need to continue to promote the digital economy in order to effectively realize its contribution to GIE in the future. In addition, in Table 5, while both RDE and IET show an overall negative impact on GIE, Fig. 2 displays that the impacts of both RDE and IET on GIE shift from negative to positive, with 2019 as the turning point for the former and 2018 for the latter. China’s continued investment in R&D in line with the current situation and the expansion of international exchanges and cooperation are effective ways to enhance GIE.

Fig. 2.

The temporal heterogeneity of influences.

Fig. 3.

The spatial heterogeneity of influences.

Spatial Heterogeneity of Influences

The spatial heterogeneity of factors’ influences on China’s GIE is illustrated in Fig. 3. The province with the largest positive effect of GDPPC on GIE is Heilongjiang, followed by Jilin and Liaoning, all located in the northeastern region. Other provinces with positive effects are mainly found in the west. In the eastern and central regions, where the economic development level is relatively high, GDPPC has a mainly negative effect on GIE. Indeed, the excessive pursuit of short-term economic benefits by local governments leads to development at the expense of environmental pollution and ecological damage, which ultimately inhibits GIE. Economically underdeveloped regions need to continue to develop economies, while other regions need to curb the excessive pursuit of economic benefits. In terms of spatial distribution, the direction of DFI’s impact on GIE is almost exactly opposite to that of GDPPC. The positive impact of DFI on GIE is mainly in the eastern and central regions, while the negative impact is mainly in the western and northeastern regions. Therefore, the western and northeastern regions need to strengthen and improve the infrastructure of the digital economy, so as to realize the role of the digital economy in the enhancement of GIE.

Similarly, the effect of RDE on GIE shows significant spatial heterogeneity. In the western and northeastern regions, RDE plays a predominantly negative role, while in other regions it plays a positive role. This suggests that the western and northeastern regions should improve the system of science and technology development and establish a good incentive mechanism to attract a large number of technical talents, high-tech enterprises, institutions of higher education, and other technological innovation subjects. As for other regions, it is imperative to continue to invest in R&D expenditure in order to sustain the important role of R&D in green technology innovation. As depicted in Fig. 3, MI positively contributes to GIE in all provinces except Ningxia, Gansu, and Sichuan. This provides a practical basis for local governments to enhance the development of product and factor markets and strengthen their linkages with markets.

The provinces with large positive contributions of INS to GIE are Guangdong, Hainan, and Guangxi, while the provinces with large negative contributions are Shanghai, Jiangsu, and Shandong. Most of the provinces where INS shows a negative impact on GIE are located in the eastern, central, and northeastern regions, while the western region mainly shows positive impacts. Hence, the northeastern region should take into account its natural environment and resource endowment and actively develop tourism resources, such as winter tourism programs, to reduce the long-term overdependence on the secondary industry. The eastern and central regions should also appropriately reduce highly polluting industries and improve the greenness and sustainability of technological innovation. In addition, the negative impact of IET on GIE is mainly found in Qinghai, the northeastern region, and Xinjiang, all of which are relatively remote regions that are weak in terms of import and export trade. Therefore, these provinces can take advantage of national-level development strategies, such as the Belt and Road Initiative, to strengthen international trade exchanges and cooperation.