1. Introduction
With the emergence of the Porter Hypothesis, the relationship between environmental regulations (ERs) and technological innovation has received wide attention from researchers over the last two decades [
1,
2,
3]. To investigate the role of regulation, scholars have carried out a great deal of studies, upon which no consensus has been reached yet, as scholars continue to find conflicting evidence [
4,
5,
6]. New research has also emerged investigating the indirect effects of ERs on technological innovation [
7]. However, these studies have typically ignored the spatial agglomeration of technological innovation activities. The purpose of this study is to examine whether ERs affect technological innovation efficiency at the provincial level from the viewpoint of geography.
Figure 1a–d details the spatial disparity of technological innovation activities in Chinese industrial enterprises for the year 2015. To be specific, provinces with high innovation inputs (i.e., R&D personnel—research and development personnel—and R&D expenditure) are mainly concentrated in eastern China; provinces with high innovation outputs (i.e., the number of patent applications and new product sales revenue) are also predominantly concentrated in eastern China. Overall, from the perspective of geography, there is a spatial agglomeration in Chinese technological innovation activities. Therefore, spatial location should be taken into consideration when examining the relationship between technological innovation and ERs.
Taking a cue from
Figure 1, the panel data of China’s 30 provinces is used to explore the relationship between ERs and technological innovation through spatial econometric methods. In contrast to traditional econometric methods, spatial econometric methods can investigate whether technological innovation is affected by the ERs of adjacent areas. Furthermore, the development of technological innovation in Chinese industrial enterprises is measured by an improved data envelopment analysis (DEA) model. To the best of our knowledge, no study to date has explored the Porter Hypothesis by combining the DEA model and spatial econometric model.
As a result, the contributions of this study have the following four aspects. First, this study applies three methodologies—statistical, spatial, and visual—to examine the relationship between technology innovation efficiency and ERs, which may develop a bridge between the geography of innovation and the ecological-economic literatures, and thus is one of the major contributions of this study. Second, this study considers the spatial agglomeration of technological innovation activities when examining the Porter Hypothesis, which will extend the literature on the Porter Hypothesis to geography. Third, for all we know, this study is the first empirical measurement of technological innovation efficiency through the game cross-efficiency DEA model. There are two reasons for using this model: (1) In the traditional DEA model, if the decision-making units (DMUs) are effective, their efficiency scores are 1, which leads to unavailable discrimination of the DMUs. (2) The improved cross-efficiency DEA model takes into account the game relationship between DMUs. Fourth, fully considering the spatial agglomeration of technological innovation efficiency at the provincial level, spatial econometric methods are used to test the Porter Hypothesis; these methods may avoid the estimation bias due to ignoring spatial effects.
The structure of this study is as follows.
Section 2 provides literature review.
Section 3 presents the improved cross-efficiency DEA model and the spatial econometric methods.
Section 4 and
Section 5 discuss the relationship between ERs and technological innovation efficiency.
4. Results
4.1. Technological Innovation Efficiency
The game cross-efficiency model is used to calculate China’s province-level technological innovation efficiencies from 2006 to 2015. As shown in
Table 1, the results demonstrate that the game cross-efficiency model performs well in differentiating the DMUs.
There are differences in the development trends of the technological innovation efficiencies in each province. For example, the provinces whose efficiency scores show a fluctuant descending tendency are mainly concentrated in eastern and central China, such as Beijing, Fujian, Guangdong, Henan, and Hunan. Western provinces, such as Gansu, Ningxia, Shanxi, Sichuan, and Xinjiang, appear to rise in the relative advantages of efficiency scores. The average efficiency scores of Zhejiang, Tianjin, Chongqing, Shanghai, and Guangdong are all above 0.8, indicating that the technological innovation inputs and outputs in these provinces are more reasonable. Of all the 30 provinces, 26.7% of them are below 0.4 (Hainan, Qinghai, Hebei, Shaanxi, Xinjiang, Yunnan, and Gansu) in terms of efficiency scores, indicating that industrial enterprises in these provinces need to reduce emissions in addition to increasing the innovation inputs.
We select three time points in 2006 (the start time of our datasets), 2010 (the end time of the 11th Five-year plan—for the development of China’s economy, the government will formulate a plan every five years, which is the five-year plan—and 2015 (the end time of our datasets) to show the three-dimensional space patterns of efficiency scores, as shown in
Figure 2. It can be seen that China’s province-level technological innovation efficiencies are characterized by multiple peaks. These peaks are mainly located at 30°–40° north latitude and 120°–140° east longitude, while the largest valley is around located at 35°–45° north latitude and 100°–120° east longitude. These results indicate that the technological innovation efficiency in southeastern China is more efficient. The number and height of peaks have displayed a downward trend over time, indicating that the overall gap in the technological innovation efficiency at provincial level tends to narrow.
Overall, the efficiency scores deviate from a uniform distribution, indicating that there may be spatial autocorrelation in China’s province-level technological innovation efficiencies. This speculation will be investigated in the next-step test.
4.2. Spatial Autocorrelation Test
The spatial autocorrelation test is used to investigate the global
Moran’s I of China’s province-level technological innovation efficiencies in the sample period. According to the approach above, three spatial weights matrices are constructed to determine whether there are adjacent dependencies (M
A), geospatial distance dependencies (M
S), or economic distance dependencies (M
E) in China’s provincial technological innovation efficiencies (see
Table 2).
As shown in
Table 2, China’s provincial technological innovation efficiencies of industrial enterprises show a positive spatial autocorrelation. Moreover, there are differences in the development trends of the global
Moran’s I in each spatial weight matrix. For example, in general, the values of M
A and M
E display a fluctuant descending tendency during the period studied, while the value of M
S displays a fluctuant ascending tendency. This result indicates that the adjacent dependence and economic distance dependence of province-level technological innovation efficiencies are weakened during the sample period.
Although there are some spatial autocorrelation findings for China’s provincial technological innovation efficiencies, the annual global
Moran’s I can only detect spatial dependence in global regions. It seemingly overlooks the spatial autocorrelation in local regions. Taking the adjacent spatial weight matrix as an example, the
Moran’s I scatter plots for 2006, 2010, and 2015 are drawn to further investigate whether China’s province-level technological innovation efficiencies are clustered, as shown in
Figure 3.
Figure 3 reports that in 2006, 2010, and 2015, 18 sample points (60%), 18 sample points (60%), and 16 sample points (53.3%) are located in the first and third quadrants, indicating that China’s province-level technological innovation efficiencies seem to cluster together in space. In addition, corresponding local indicators of spatial association (LISA) also provide visual evidence of spatial clustering for the efficiency of technical innovation in industrial enterprises (see
Figure 4).
As shown in
Figure 4, the provinces belonging to the H-H clustering type are mainly concentrated in eastern China, while the provinces belonging to the L-L clustering type are mainly centralized in northwest China. Specifically, in 2006, Gansu (
p = 0.01) belonged to the L-L clustering type; Jiangxi (
p = 0.05) belonged to the L-H clustering type. In 2010, Inner Mongolia (
p = 0.05) joined the L-L clustering type; Jiangsu (
p = 0.05), Zhejiang (
p = 0.05), and Shanghai (
p = 0.05) belonged to the H-H clustering type. In 2015, Jiangsu (
p = 0.05) and Zhejiang (
p = 0.01) belonged to the H-H clustering type; Xinjiang (
p = 0.05) joined the H-L clustering type; Jiangxi (
p = 0.05) joined the L-L clustering type. These findings again demonstrate the existence of spatial clustering of China’s province-level technological innovation efficiencies.
4.3. Effects of ERs on Technological Innovation Efficiency
4.3.1. The Traditional Panel Econometric Model
For the sake of comparison, a traditional panel model is first constructed, as shown in
Table 3. As supplementary analysis, the multicollinearity test for the panel data is conducted. The results of variance-inflating factors (VIFs) show that the multicollinearity among the indicators can be ignored (see
Table 3). Before performing econometric analysis, the Hausman test is conducted to determine whether the econometric model is optimal under the fixed effect. According to
Table 3, the result of the Hausman test is 25.674 (
p = 0.019), implying that the econometric model is more appropriate to use the fixed effect. Moreover, the results of the likelihood ratio (LR) test show that the values of the individual fixed effect and the year fixed effect are significant at the 1% level, indicating that the traditional panel model under the fixed effect is optimal.
It can be seen that the coefficient of NOEPA is positive but only significant under the individual fixed effect. The coefficient of NOEAPC ranges from −0.004 to 0.039, and none of them pass the significance test except for the OLS model. These results suggest that there is no discernible relationship between regulations and technological innovation efficiency through the traditional panel model. In addition,
Table 3 reports that at least one of the LM test is statistically significant in whatever fixed effects incorporated into our model, which indicates that the meta-hypothesis (there is no spatial correlations) does not hold, and the robust LM tests are the same. As a result, when examining the relationship between ERs and technology innovation efficiency at the provincial level in China, the traditional panel model is not enough, and the spatial panel model needs to be considered.
4.3.2. The Spatial Panel Econometric Model
According to LeSage and Pace (2009) research [
47], the steps for the spatial panel econometric model are as follows: (1) The panel
Moran’s I needs to be calculated to determine whether the panel data has spatial dependence. (2) The Wald and LR test need to be conducted to determine whether the SPDM model can be transformed to a SPLM model or a SPEM model. It can be seen from
Table 4 that the estimation results of the Wald and LR test are statistically significant at the 5% level, indicating that
H0:
ρ = 0 and
H0:
δ = −
ρβ are rejected. The SPDM model could not be simplified into the SPLM model or the SPEM model. Therefore, we use the SPDM model to analyze the effects of ERs on the technological innovation efficiency, and the estimation results of the SPLM model and the SPEM model are also listed for reference.
Just focusing on the individual fixed SPDM model (see
Table 4), the coefficient of
ρ is statistically significant at the 5% level. This result suggests an obvious spatial spillover in the technological innovation efficiency of industrial enterprises. The coefficient of NOEPA is statistically significant at the 5% level, and an increase of 1 in NOEPA will lead to an increase of 0.098 in the technological innovation efficiency. This result indicates that voluntary regulation is positively associated with the efficiency of technological innovation in industrial enterprises. The coefficient of NOEAPC is not statistically significant, indicating that there is no significant correlation between mandatory regulation and technological innovation efficiency.
As for the control variables, the foreign investment is positively associated with the efficiency of technological innovation in industrial enterprises. The coefficient of GFS is negative and statistically significant, suggesting that the government’s financial support might not help industrial enterprises with developing new products in the long term. In addition, the coefficient of spatial lag variable (W * NOEPA) has evident characteristics, indicating that voluntary regulation may have a spatial spillover effect. Therefore, we use the partial differential method to investigate the direct, indirect, and total effects of ERs on the technological innovation efficiency in industrial enterprises, as shown in
Table 5.
It can be seen that the coefficient of NOEPA shows a statistically significant direct effect, indicating that if a province promulgates voluntary regulation, its technological innovation efficiency will be improved. Furthermore, the indirect effect of NOEPA on the technological innovation efficiency is statistically significant at the 5% level, suggesting that the technological innovation efficiency of one province is affected by the voluntary regulation of neighboring provinces. In contrast, the direct and indirect effects of NOEAPC on the technological innovation efficiency are not statistically significant at the 10% level, which is consistent with our expectations. It can be seen from
Table 5 that mandatory regulation has no significant influence on the efficiency of technological innovation in industrial enterprises.
5. Discussion
Strengthening ERs so that they stimulate technological innovation is an academic and practical concern. This study contributes to the grasping of technological innovation efficiency in terms of (1) how the spatial disparity of the efficiency of technological innovation in industrial enterprises and (2) how the relationship between ERs and technological innovation efficiency from a spatial perspective.
As demonstrated empirically, the technological innovation efficiency of industrial enterprises in southeastern China is generally higher than those of provinces in northwestern China. This finding bolsters existing literature that emphasizes the importance of geographic context for technological innovation research [
9,
12]. This is because industrial enterprises in the southeastern regions usually have relatively rich innovation resources and developed platforms for technology transfer, which induces new product development. Consistent with extensive research on the spatial correlation of similar innovation activities [
13,
16], our findings demonstrate the existence of spatial dependence in China’s province-level technological innovation efficiencies. Some scholars believe that leading industrial enterprises may tend to avoid clusters in the clustering process, because their relative advantages may be impaired [
48]. However, our speculation is that provinces with similar technological innovation efficiencies tend to agglomerate within a single cluster in geospatial space, possibly because sub-clustered provinces perform more efficiently through leveraging local knowledge spillovers in relation to new product development [
13]. These findings can assist the government in allocating limited fiscal resources more effectively through accurately pinpointing some provinces with high technological innovation efficiency.
The design of ERs plays a key role in promoting the efficiency of technological innovation in industrial enterprises. For example, voluntary regulation can stimulate the technological innovation efficiency of industrial enterprises at the provincial level, possibly because voluntary regulation only specify pollution prevention goals, but provide enterprises with discretion concerning the ‘how’, which can develop a mutual trust between environmental protection agencies and industrial enterprises [
29,
31]. In our study, if the government adopts mandatory regulation to restrict industrial enterprises, the enthusiasm of enterprises for technological innovation may be dampened, possibly because mandatory regulation forces enterprises to meet pre-specified environmental standards or else face administrative penalties, then enterprises have to pay for the pollution control technique [
6]. These findings extend the Porter’s Hypothesis to the case of China’s provincial industrial enterprises. Furthermore, our findings provide empirical evidence of the relationship between ERs and technological innovation efficiency, which represents a departure point for future qualitative or quantitative explorations into the Porter’s Hypothesis associated with spatial location. These findings can also help the government to promulgate more advanced environmental decision.
The empirical results show that the spatial panel econometric model not only outperforms the traditional panel model but also reveals the spillovers effects of dependent and independent variables. In our case, the efficiency of technological innovation in industrial enterprises has a spatial spillover effect at the provincial level, possibly because the spatial proximity to knowledge can contributes to the exchange of information among provinces improving the efficiency of knowledge transfer [
9]. In our study, the technological innovation efficiency of one province is influenced by local voluntary regulation and the neighboring provinces due to the spillovers effect. This is possibly because if a province strengthens ERs, the productive investment of its industrial enterprises is likely to be crowded out, causing some industrial enterprises to transfer to neighboring provinces. In this study, statistical, visual, and spatial methods are combined to test the Porter Hypothesis, which develops a bridge between spatial metrology and the Porter Hypothesis. These findings can also help guide location decisions for industrial enterprises.
The results have several policy implications. First, local governments should capitalize on the agglomeration of innovation activities in industrial enterprises to maximize innovation efficiency. For instance, provinces with high innovation efficiency in southeastern China can increase investment in research funding and experimental equipment to stimulate innovation. Second, local governments should engage in establishing a scientific and effective environmental regulation system. Third, the spillovers effect of innovation activities is also a great way to meliorate technological innovation efficiency. Local governments can establish a cross-provincial innovation demonstration area to encourage knowledge spillovers among provinces. For example, provinces with spatial proximity (i.e., Anhui, Shanghai, Jiangsu and Zhejiang) can set up special funds to guide collaborative innovation.