Can Cooperation Stimulate Firms’ Eco-Innovation? Firm-level Evidence from China


 Firms’ collaborative activities have created increasing opportunities for eco-innovation in modern society. Based on unbalanced panel data from the Chinese National Innovation Survey between 2011 and 2015, this paper explored the influences of different modes of cooperation, i.e., vertical cooperation, horizontal cooperation, and mixed cooperation, on the eco-innovation of Chinese manufacturing firms. Results indicated that three types of cooperation all had positive and statistically significant effects on the firms’ eco-innovation, and mixed cooperation had promoted eco-innovation more dramatically. The extent of such impacts may vary depending on the heterogeneity of the characteristic of enterprises. We also verified that knowledge spillovers from cooperative partners have played a mediating role between cooperation and eco-innovation. Our results suggest the potential benefits of diversified collaborative activities and appropriate intellectual property protection for firms’ eco-innovation in China.


Introduction
As a key to decoupling environmental pressure from economic growth, eco-innovation has gained increasing attention by governments, scienti c communities, and rms, to uncover the underlying determinants for sustainable growth owing to its "double externalities" (Barbieri et al. 2016; Del Río et al. 2016). Eco-innovation can not only effectively reduce the environmental negative externalities, but also produce positive knowledge externalities (Rennings 2000). However, such positive knowledge externalities may render the rms, which are original with high enthusiasm in eco-innovation, less motivated, mainly because the knowledge and technology originated from eco-innovation can also bene t other rms or even its competitors by knowledge and technological spillovers. On the other side, the complex and uncertain nature of eco-innovation sets a higher standard for cross-disciplinary knowledge and diversi ed skills (Horbach et al. 2013). The multitudinous and wide-distributed knowledge, skills, and resources essential for eco-innovation are generally unavailable for one individual enterprise, but can only be obtained from external sources. As a consequence, rms concentrating on their internal environment may lose the opportunity for achieving the potential eco-innovation and the consequent bene ts (Laursen and Salter 2006). Moreover, the high investment of diverse knowledge and resources for eco-innovation may bring about high risk inherently, and thus prevent one individual enterprise from the potential eco-innovation due to its high concern about the returns of investment (Acemoglu et al. 2016).
In this case, cooperation aimed to explore the innovation resources outside the rms has been regarded as an effective solution for the problems arising from the high and diverse requirements for knowledge for rms' eco-innovation (Marzucchi and Montresor 2017). Currently, open cooperation includes the horizontal collaboration with partners beyond the supply chain (such as universities, research institutions, the government, and even competitors, et, al.) and vertical collaboration with partners within the supply chain (including consumers and suppliers), and such collaborative activities for innovation has evolved as a new way that helps rms maintain their competitive advantage (Chesbrough 2003 ). Moreover, partners can also afford the cost-and risk-sharing in the presence of potentially high risk accompanied with innovation (Miotti and Sachwald 2003). Based on an investigation of 77 telecom equipment manufacturers in 13 countries, Phelps (2010) found that collaborative networks have bene ted the innovative rms a lot from their partners with diversi ed technologies, and thus promoted their exploratory innovation.
Such bene ts of open cooperation can also go for eco-innovation, which is characterized by complexity and high knowledge demands, for external knowledge (De Marchi 2012; Awan et al. 2021). In the face of increasingly stricter environmental regulation, heterogeneous partners can help innovative rms to much more rapid access to the diversi ed knowledge, resources, and the transformation and application of ecoinnovation technologies (Del Río et al. 2016). As a consequence, increasingly more rms choose to develop their own inter-organizational collaborative networks to promote eco-innovation for better environmental and economic performances with relatively lower cost and risk ( In the past decades, rms in China are facing "the strictest environmental regulation" ever, and thus have to seek their optimal behavior decisions to strive for a balance between economic outcomes and environmental performance. Eco-innovation has also been highly favored by rms and the Chinese government. However, the rm-level eco-innovation in China is di cult to meet the practical requirements, mainly due to the insu cient eco-innovation and achievement transformation activities (Miao et  have issued the Guidelines on building an eco-innovation system, and aim to construct a market-oriented eco-innovation system by 2022 by enhancing the complementary collaboration among industries, universities, research institutions, nancial service institutions, and intermediary agents. Hofman et al. (2020) collected data in the Chinese automotive, electronics, and textiles manufacturing sectors through a self-designed questionnaire, and found that cooperation with suppliers is conducive to green process innovation. Yang and Lin (2020) found that supply chain cooperation has played an important role in eco-innovation strategies by studying a case of an automobile rm in southwestern China. Although the effect of collaboration on the eco-innovation in China has been involved by previous literature, its inherent in uencing mechanism, and the endogeneity problems accompanied by the lack of rm-level data, still need further investigation. Such in-depth exploration may help to identify the key factors in promoting the eco-innovation practices, and further facilitate the win-win development of environmental improvement and economic outcomes in China.
To ll the above-mentioned gap in the existing literature, this paper contributes to the existing literature in the following ways. First, as a complement to the existing perspectives between collaboration and ecoinnovation, mixed cooperation was also considered for innovation, rather than only that within or beyond the supply chain. Second, by constructing reliable two-period panel data from China National Innovation Survey, we constructed a panel xed effect model to provide micro evidence of the impact of different types of cooperation on eco-innovation in China. Third, the mechanism of the impact of cooperation on eco-innovation at the micro-level has also been clari ed by using knowledge spillover as the mediating channel.
The remainder of the paper is structured as follows. Section 2 proposes the inherent mechanism between cooperation and eco-innovation, and Section 3 introduces the estimation strategy and the data. Section 4 presents the main results of the baseline regression, the heterogeneous analysis, robustness analysis, and further analysis. The nal section concludes with policy implications.

Different types of cooperation and eco-innovation
Previous literature has summarized that cooperators for innovative rms mainly include partners beyond the supply chain and that within the supply chain, according to the bene ts of cooperation for innovation with different types of organization (Tether 2002). Due to the diversities in knowledge type, management mode, and cooperative motivation, cooperation with different partners may affect innovative rms' ecoinnovation differently (Awan et al. 2021;De Marchi 2012).
For partners within the supply chain, suppliers and consumers can help innovative rms to complement different types of resources for further eco-innovation. On one hand, due to suppliers' specialized resources of the technical requirements, component speci cations, cooperation with suppliers can bene t the innovative rms with strategic knowledge and technologies for their internal green R&D efforts (Du et al. 2018;Hofman et al. 2020). In addition, their mutual technological dependence on skills and resources can make their collaborative activities with environmental-friendly factor inputs in the whole production process, to further reduce the adverse impact on the environment (De Marchi 2012). On the other hand, due to the potential uncertainties of consumer preferences of emerging technologies, consumers can contribute to obtaining complementary knowledge on users' behavior, and balancing the environmental and economic outcomes with lower prices (Kammerer 2009;Chatterji and Fabrizio 2014). Therefore, such vertical cooperation with partners along the supply chain may positively in uence the propensity of rms to innovate because they can provide crucial information on technologies and markets (Miotti and Sachwald 2003).
In terms of horizontal cooperation with partners beyond the supply chain for innovation, various partners can contribute diversi ed resources and technical capabilities to supplement the limited internal resources of the innovative rms (Lichtenthaler 2011). Universities and research institutions can effectively guarantee the acquisition of cutting-edge knowledge and emerging technologies to complement their internal knowledge with much lower R&D expenditures (De Marchi 2012), and governmental partners can provide long-term research funding support and unique material resources for rms in technology development alliances (Doblinger et al. 2019). Especially, the industry-universityresearch cooperation can also deliver a lot of intellectual and human capital for rms' innovative activities (Horbach 2014). Based on a probit model analysis, Triguero et al. (2013) concluded that rms, which attached importance to the industry-university-research cooperation in European countries, are more active in eco-innovation activities. In addition, cooperating with other rms and institutions can help to share the high risks and costs of eco-innovation with rms, so that rms are more motivated to conduct the eco-innovation. However, it is noteworthy that innovative rms collaborated with competitors may be affected by their mutual competitive pressure and knowledge spillovers, which is not conducive to rms' eco-innovation (Cainelli et al. 2012).
It should be also noteworthy that the above-mentioned types of cooperators are more than substitutes, but can also be complements due to their potential additive effects (Howells 2006). Due to the huge requirements on the cutting-edge, diversi ed, and interdisciplinary knowledge and abilities, it is hard for one single rm to be pro cient in all the skills required for eco-innovation (Doran and Ryan 2012). Cooperators beyond the supply chain, especially universities and research institutions, can provide innovative rms with the acquisition of advanced knowledge, but they are generally insensitive or less responsive to the market demand (Tether 2002). As a complementarity, partners within the supply chain can help to better understand the target market needs, and thus realize the most suitable exploitation of knowledge, skills, and resources necessary for eco-innovation. Using data from Community Innovation Surveys (CIS) conducted in the Netherlands, Belderbos et al. (2006) found that cooperation strategies with competitors and customers, and with customers and universities can play complementary roles in the innovation process. By such mixed cooperation, namely, complementing their respective collaboration advantages from both scienti c communities and supply-chain partners, rms can be more motivated to the engagement in eco-innovation activities and the consequent transformation into new and high-quality products with responsive market orientation (Haus-Reve et al. 2019).
Thus, the rst hypothesis was posited as follows: H1: The establishment of various types of collaborative partnerships can help rms to develop ecoinnovation in China.

The in uencing mechanism of cooperation on ecoinnovation
Cooperation can promote eco-innovation through knowledge spillovers. The establishment of stable collaborative networks can create the possibility for knowledge spillovers by the potential exchange of information and resource, and thus improve knowledge networks (Wang et al. 2014). Speci cally, cooperation between rms and different types of partners will directly trigger knowledge acquisition and knowledge integration, which can be easily achieved by their daily mutual communication (Wang 2016), to affect the innovation performance of rms (Hagedoorn et al. 2018;Liao and Liu 2021). Moreover, such knowledge spillovers can be sustainable, even under the situation that there is no cooperation at present, the researchers who have cooperated in the past may continue to exchange knowledge in an informal way (Frenken et al. 2010). The consequent complementarity of internal R&D activities and external knowledge acquisition is conducive to the rms' innovation activities (Cassiman and Veugelers 2006). Overall, such direct and sustainable knowledge spillovers from collaborative activities can promote rms' eco-innovation by knowledge sharing (Song et al. 2020). Chatterji and Fabrizio (2014) found that rms' external cooperation is conducive to the use of external knowledge from other rms, universities, and users to promote innovation. In particular, for the rms with more knowledge, which generally tend to have strong learning ability, the potential cooperation can not only enrich the rm's knowledge, but also enable them to quickly respond to changes in the market environment to reduce the uncertainty of ecoinnovation (Liao 2018).
Cooperation can also make tacit knowledge explicit, thereby facilitating the exchange of knowledge. The complex technologies and technical know-how required by eco-innovation is generally tacit knowledge, which is di cult to be simply transferred by documents but more reliable on close social communication (Becerra et al. 2008). Gertner et al. (2011) concluded that cooperation based on personal social interaction generally contains more tacit knowledge, which could become more explicit and more easily transferred during collaborative engagements (Willoughby and Galvin 2005). Furthermore, successful innovative entrepreneurs are less likely to exchange their tacit knowledge with unfamiliar rms (Singh et al. 2016). Such a situation necessitates collaborative networks with these innovative entrepreneurs directly or indirectly with mutual trust to obtain the highly tacit knowledge, and such partners with the common purpose, are more inclined to share knowledge in a positive attitude to allow rms to absorb the knowledge.
Above all, the second hypothesis was formulated as follows: H2: The cooperation between rms and partners generates knowledge spillovers, and thus promotes their potential eco-innovation.

Estimation strategy and the variables
To investigate the impact of cooperation on eco-innovation, a xed-effects model, which can effectively address the statistical concerns that might not be tackled by an ordinary least squares method (Wooldridge 2013), has been applied to undertake the baseline regression analysis in this paper. The xed-effects model is speci ed as follows: eco_inno it = α 0 + α 1 vertical it + α 2 horizontal it + α 3 mix it + β i ∑X it + γ c + δ y + λ i + μ it 1 where, eco_inno it is the number of types of eco-innovation introduced by rm i in year t. vertical it , horizontal it , and mix it are dummy indicators that equal 1 if rm i conducts the vertical cooperation, horizontal cooperation, and mixed cooperation in year t, respectively, with α 1 ,α 2 andα 3 representing their coe cients. X it is a vector of control variables (i.e., rm characteristics, entrepreneur characteristics, city characteristics). We also control for (a) industry xed effects (λ i ), (b) year xed effects (δ y ), and (c) city xed effects (γ c ). μ it is the error term. In addition, to modify the potential heteroscedasticity caused by the correlation among different rms in different sectors, we further clustered the standard errors at the industry level.

Dependent variable
In this study, the self-reported data about the impact of process innovation on the environmental bene ts of rms from the CNIS have been employed to measure eco-innovation, as used in previous literature (Horbach et al. 2012;Antonioli et al. 2013). In the CNIS, which is similar to the widely used CIS, rms were requested to report their achieved bene ts, mainly in material saving, energy saving, or emission reduction, from introducing process innovation. If one of the bene ts in material saving, energy saving, or emission reduction has been achieved through the process innovation by rms, we de ned this as ecoinnovation, and we further used the sum of the values for each type of bene t as the eco-innovation variable, following the method in Ghisetti et al. (2015). Similar to the method in Horbach et al. (2012), for each type of the achieved bene ts, the answer of "very high" receives a value of 1, and the answer of "medium", "low", or "none" receives a value of 0. It should be noted that the assignment of a value of "0", rather than "1" in Horbach et al. (2012), to the answer of "medium" mainly aimed to prevent the entrepreneurs from overstating their potential achievements.

Independent variables
According to the response of which kind of partners rms cooperate with to innovate in the CNIS, we de ned the key independent variables, i.e., different types of cooperation, as three dummy variables, following the method used by Haus-Reve et al. (2019). Horizontal cooperation is measured as a dummy with value 1 when the rms cooperate with any one of the following partners, including other rms, universities, research institutions, governments, associations, competitors, consultants, intermediaries, venture capital institutions, or other partners in their a liated group. Similarly, the vertical cooperation is represented as a dummy with value 1 if the rms' partner is one of their suppliers or customers. Furthermore, if rms' partners include both the above horizontal and vertical partners, we de ned this kind of cooperation as the mixed cooperation with the dummy value as 1, otherwise with the value as 0.

Control variables
In this paper, several rm-, entrepreneur-, and city-level variables, which may potentially affect the exploratory innovation, have been controlled to isolate the marginal effects of the explanatory variables only.
Firm-level control variables. (a) Generally speaking, larger rms have more capital and human resources to support the implementation of eco-innovation (Rehfeld et al. 2007; Liao and Liu 2021). We measured the rm size as the natural logarithm of the number of employees by rm i in year t. (b) The rm age is expected to in uence the knowledge base, thus affecting the development of eco-innovation (Barbieri et al. 2016). Firm age is de ned as the natural logarithm of the differences between the foundation date of rm i with its surveyed date in the CNIS. (c) Previous studies have shown that the ownership of a company, state-owned or private-owned, is increasingly considered as a key factor, which may signi cantly affect the development of technology innovation (Clò et al. 2020). In this paper, we controlled for rm ownership by using three dummy variables, i.e. state-owned enterprises(SOEs), domestic private, and foreign-invested, that take the value of 1 if a rm belongs to the corresponding rm type or 0 otherwise. (d) Firms with more xed assets tend to operate more stably, which is more conducive to the adoption of eco-innovation, and we de ned the per capita xed assets as the natural logarithm of the total xed assets per employee. (e) Previous studies have shown that exports have little in uence or negative in uences on eco-innovation for European rms (De Marchi 2012), while rms in China are more motivated to carry out eco-innovation to meet the increasingly strict environmental standards of the foreign market for exports. Therefore, export is considered as another important control variable, and is de ned as a dummy variable, with the value as 1 if the export delivery value of rm i in year t is not 0.

Entrepreneur-level control variables
Factors related to entrepreneurs will create innovation (Larson 2000). (a) Gender is measured as a dummy with a value of 1 when the entrepreneur is male according to the relevant survey response. (b) To control for entrepreneurs' age, ve dummy variables (i.e. younger than 29 years, 30-39 years old, 40-49 years old, 50-59 years old, and that older than 60 years) have been used, which receive the value 1 if they fall under the corresponding interval, or 0 otherwise. (c) Similar method has been employed for controlling entrepreneurs' education level by using ve dummy variables, namely, Ph.D., master, undergraduate, junior college, and others.
City-level control variables (a) Previous studies have shown that environmental regulation is the main driving factor of ecoinnovation (Horbach 2008). In this paper, a city-level comprehensive indicator has been designed by incorporating the city-level air pollutants (SO 2 , and soot and dust) removal rate, the water pollutant treatment rate, and the industrial solid waste utilization rate to represent the city-level environmental regulation (regulation) in the studied cities, by following the approach in Cheng et al. (2018). (b) The urban economic level is also an external factor that affects the rms' eco-innovation (Horbach 2014), and the natural logarithm of satellite nightlight density is used as a proxy to re ect urban economic level (nightlight) in this paper, mainly because that the nightlight data can not be confounded by the price factors among different regions (He et al. 2020).

Sample and data
To investigate the effect of cooperation on rm eco-innovation, we constructed data samples from three major sources. First, data on the innovation activities of Chinese manufacturing rms are from two waves of innovation surveys on manufacturing rms in four developed regions (namely Jiangsu, Zhejiang, Guangdong, and Shanghai) in the China National Innovation Survey (CNIS) 1 . Two surveys were jointly conducted by China's National Bureau of Statistics and the Ministry of Science and Technology in 2011 and 2015. According to the requirements of the survey, all "above-scale" manufacturing rms, with annual sales income above 5 million CNY before 2011 and 20 million CNY thereafter, were required to respond to the questionnaire. Consistent with the CIS and OECD's Oslo manual, the survey consisted of a rm innovation questionnaire and an entrepreneur questionnaire. The former includes detailed coverage of rm-level information on innovation activities (inputs, outputs, sources, effects, obstacles, modalities, etc.), and the latter asked the entrepreneurs to provide personal information like gender, age, education levels, and their attitudes and perception of government policies to support innovation. Therefore, the survey is the fundamental data source to identify the impact of cooperation on rms' eco-innovation. Based on the above-mentioned data sources, we rst obtain our sample by selecting rms located in Jiangsu Province, Zhejiang Province, Guangdong Province, and Shanghai City according to their administrative location codes at the city level. Then, we use both the legal identi cation code and the rms' names to match the samples between the CNIS and ASIF. In the regression model, rm-level control variables, environmental regulation, and satellite nightlight densities are lagged by one year to alleviate problems from the potential reverse causality to the best as we can.
Finally, we construct unbalanced panel data regarding rm eco-innovation, cooperation innovation, and other covariates both at the rm and the city level in two years: 2011, and 2015. The total data for 2011 and 2015 consist of 43,132 observations, and the number of observations from Jiangsu, Zhejiang, Guangdong province, and Shanghai city is 17185, 14414, 8660, and 2873, respectively. Table 1 presents a summary of descriptive statistics for dependent and main independent variables. Ecoinnovation (ei for short in the following) has a mean of 0.981, suggesting that all covered rms have a relatively low level of eco-innovation between 2011 and 2015. For main independent variables, on average, vertical, horizontal, and mix are about 0.066, 0.149, and 0.178, respectively, indicating a visible difference among various types of cooperation. Furthermore, considering the regional distribution of eco-innovation activities, Table 2 shows the ecoinnovation activities in each province during the two waves of investigation. Firms introducing ecoinnovation, which achieved material-saving, energy-saving, and emission reduction, accounted for about 45% of the total surveyed rms, and rms that introduce innovation to obtain all the three types of bene ts, accounted for more than 20% for each province, indicating that rms have paid more attention to the innovative technologies that can achieve material-saving, energy-saving, and emission reduction, simultaneously. In this paper, only the data from Jiangsu, Zhejiang, Shanghai, and Guangdong has been used, mainly due to the data limitation and the representativeness of active innovation engagements in these four developed regions of China. Table 3 reports the impact of cooperation on eco-innovation. From columns (1)  More precisely, horizontal cooperation and mixed cooperation are statistically signi cant (at the 1% level), while vertical cooperation is statistically signi cant (at the 10% level). Firms that choose cooperation display a stronger likelihood to introduce eco-innovation. Therefore, H1 was veri ed.

Cooperation and eco-innovation
Mixed cooperation had a more dramatic impact on promoting eco-innovation. As shown in Table 1, mixed cooperation has a mean score of 0.178, which is greater than that of the other two modes, suggesting that rms covered in our sample tended to choose mixed cooperation. Such choice has also shown better performance on eco-innovation promotion. By comparing the coe cients of vertical cooperation (0.042), horizontal cooperation (0.1874), and mixed cooperation (0.2682), we found that mixed cooperation has the highest positive impact on eco-innovation, suggesting that mixed cooperation can reap the bene ts of collaboration from external partners within and beyond the supply chain to the fullest.
In addition, we also found that gender, age, and education of entrepreneurs are important drivers of ecoinnovation, as shown in Column (3) of Table 3, and younger, highly-educated male entrepreneurs are more inclined to conduct eco-innovation statically. At the rm level, the rm size has positive impacts on eco-innovation statistically, and economies of scale would make rms more active in the development of eco-innovation with the expansion of rms' scale. Moreover, the per capita xed assets also have positive impacts on eco-innovation statistically, and its potential increase may bring more new energy-saving and emission-reduction equipment to rms and promote eco-innovation. In terms of rms' ownership, foreigninvested rms may have a negative in uence on eco-innovation statistically. Although these foreigninvested rms can provide rms with opportunities to learn and imitate, the knowledge and technologies cannot be utilized due to the limitations of the rm's absorptive capacity. Such high dependence on foreign technologies also inhibits their inherent willingness and activities to conduct innovation.  Table 4 shows that other rms, universities and research institutions, associations, governments, competitors and consultants, and intermediaries all play an important role in promoting eco-innovation. This alternative measure produces similar results as our baseline results in Table 3.

Alternative dependent variable
We perform the basic regression by using the dependent variable de ned by the number of ecoinnovation introduced. However, this measurement does not consider its speci c classi cation. Previous studies have shown that different types of eco-innovation may also bring heterogeneous economic performance to rms (Ghisetti and Rennings 2014), and consequently, cooperation may also have heterogeneous effects on different types of eco-innovation, given the pro t-maximization goal of rms.
Here, the eco-innovation has been further classi ed into two types, i.e., the Energy and Resource E ciency Innovation, for the reduction in material and energy input per unit of output, and the Externality Reducing Innovation, for reducing production externalities, by following Ghisetti and Rennings (2014). Speci cally, the Energy and Resource E ciency Innovation (erei) is measured as a dummy with a value of 1, if the rms have introduced innovation with a high impact on material and energy reduction. The Externality Reducing Innovation (reei) is also assigned as a dummy with a value of 1, if rms have introduced the innovation with a high impact on emission reduction, otherwise with a value of 0. As shown in columns (2) and (3) of Table 4, we found that the impacts of three types of cooperation on the Energy and Resource E ciency Innovation and the Externality Reducing Innovation are both positively signi cant, in line with the results in Table 3, indicating that the results are robust.

Alternative estimation method
According to the descriptive statistics in Table 1, the dependent variable is a count variable, and its variance is signi cantly greater than the mean value. Such overdispersion makes the negative binomial regression model also applicable (Cameron and Trivedi 2005). Therefore, we have also applied the negative binomial regression as another robustness check. As shown in column (4) of Table 4, the coe cient of vertical cooperation is insigni cant, but it is still positive. This alternative method produces results similar to the results of the panel xed-effect model in Table 3. It is proved that vertical cooperation, horizontal cooperation, and mixed cooperation are important drivers for eco-innovation.  (1)

Heterogeneity analyses
Generally, rms in energy-intensive and emission sectors may be more likely to conduct eco-innovation, which is also highly impacted by the level of intellectual property protection (Del Río et al. 2016; Brüggemann et al. 2016). Therefore, in this section, heterogeneous analyses have been further conducted.

Firms with different energy intensities. Based on the Statistical Communique of the People's Republic of
China on the 2010 National Economic and Social Development, six high energy-consuming industries 2 , have been identi ed. Firms have been then classi ed into high energy-consuming rms and low energyconsuming rms according to their sector attributes. The high energy-consuming rm is measured as a dummy with a value of 1, if the rm belongs to one of the six high energy-consuming industries, otherwise with a value of 0, representing a low energy-consuming rm. As shown in column (1) of Table   5, there is a statistically signi cant and positive association between the horizontal/mixed cooperation and eco-innovation, consistent with the results in Table 3 Table   5, we can obtain statistically signi cant and positive relations between vertical/horizontal/mixed cooperation and eco-innovation, respectively. Furthermore, the signi cantly negative coe cients of interaction between horizontal/mixed cooperation and heavy-pollution rms have shown that the promotion of horizontal cooperation and mixed cooperation on eco-innovation will be weakened in heavypollution rms, compared with that in light-pollution rms. This may be attributable to the reason that, in comparison to light-pollution rms, heavy-pollution rms are always reliable to their emission-intensive production patterns, and thus less motivated to conduct eco-innovation of production technology due to the existed technology lock-in and the potentially high costs. In addition, the cooperation may cause technology spillovers and make their innovative technology been easily imitated and even copied, and thus lead to relatively lower returns on investment, and further, alleviate the motivation for cooperation on eco-innovation.
Firms with different levels of intellectual property protection. In collaborative activities, knowledge can be spread and further be used by rms to meet their requirements of eco-innovation. However, the quantity and quality of such knowledge spillovers may be highly affected by the level of intellectual property protection in the city, where rms are located (Arora et al. 2021). Therefore, we measured the level of citylevel intellectual property protection (intproperty) by the number of documents on intellectual property protection in each city, and then we interacted that with the different types of cooperation. As displayed in Table 5, we found that vertical cooperation, horizontal cooperation, and mixed cooperation all have statistically signi cant and positive effects on eco-innovation. We also observed the statistically signi cant and negative association between vertical cooperation and intellectual property, while statistically insigni cant and negative coe cients of interaction between horizontal/mixed cooperation and intellectual property. In other words, the improvement of the level of intellectual property protection in a city may weaken the promotion of cooperation on eco-innovation. The main reason is that overstrict intellectual property protection may not be conducive to knowledge and technology sharing in cooperation by inhibiting the imitation and learning of rms. Such circumstances would make rms more reliable on their previous innovation, and have adverse effects on the engagements of more complex and valuable innovation (Brüggemann et al. 2016).

Further analysis
Based on the benchmark analysis, this paper constructs a mediating effect model to investigate how cooperation affects rms' eco-innovation. Based on the question "Which kind of information and knowledge have high impacts on rms' innovation activities" in the CNIS, the knowledge spillovers were de ned as the meditating variable to conduct mechanism analysis. Given the fact that the knowledge spillovers may be affected by different types of partners, the knowledge spillovers were further classi ed based on the above-mentioned different types of cooperation, by following the method in Martínez-Ros rst constructed a regression model to test whether these three types of cooperation can promote the corresponding knowledge spillovers, and then we veri ed whether knowledge spillovers can promote ecoinnovation, as follows: verknow it = α 0 + α 1 vertical it + β i ∑X it + γ c + δ y + λ i + μ it 2 horknow it = α 0 + α 2 horizontal it + β i ∑X it + γ c + δ y + λ i + μ it 3 mixknow it = α 0 + α 3 mix it + β i ∑X it + γ c + δ y + λ i + μ it 4 verknow it , horknow it , and mixknow it refers to vertical, horizontal, and mixed knowledge spillovers useful for the innovation of rm i in the year t, respectively. The other variables are consistent with the de nition in the benchmark model, and the standard errors are also clustered at the industry level.
As shown in columns (1) -(3) of Table 6, we observed statistically signi cant and positive associations between vertical/horizontal/mixed cooperation and vertical/horizontal/mixed knowledge spillovers, respectively, demonstrating that cooperation can generate knowledge spillovers, as we discussed in the theoretical analysis. Cooperation can be regarded as the co-production of knowledge, and knowledge spillovers are generated as a by-product (Frenken et al. 2010). In addition, business contacts and exchanges among partners could enable rms to timely grasp the most advanced tacit knowledge needed for eco-innovation.   Results have highlighted the importance of cooperation on rms' eco-innovation. Firms, aiming to conduct eco-innovation, should actively seek cooperation with different organizations with new ideas on eco-innovation. Through the establishment of a long-term and stable relationship, rms can make full use of the relevant professional knowledge generated by diversi ed partners to supplement the internal knowledge base, and thus their internal resource allocation can be further optimized and the ecoinnovation can be carried out more smoothly with relatively low cost and risk. In addition, the government should accelerate the promotion of horizon cooperation, especially the industry-university-research cooperation, which can provide diversi ed talents and resources for rms' eco-innovation, by supporting the construction of university talent incubation bases and science and technology industrial parks.

Conclusions
Finally, the Chinese government has proposed that in the 14th ve-year plan period (2021-2025), intellectual property rights should be protected, with both public interests and innovation incentives guaranteed. The heterogeneous analysis on different levels of intellectual rights protection has indicated that, the future intellectual property protection system should not only protect the core knowledge and technology of the rms from being leaked, but also ensure partners being not constrained by overstrict content of intellectual property protection for further knowledge and technology share.
As an initial attempt to explore the relations between different types of cooperation and eco-innovation in China, there are several limitations in this study. First, due to the data limitation, rms' eco-innovation in China has been measured by using the self-reported data of entrepreneurs in the China National Innovation Survey, which may be subjective to some extent. Further measurement can be extended by using more objective data, such as green patent data. Secondly, although the merge of surveys allows a time lag between the dependent variable and the independent variables to alleviate endogeneity issues to the best of our abilities, there may still exist endogenous problems in the regression model caused by missing variables, reverse causality, etc. However, the selection of appropriate instrumental variables for three types of cooperation is really di cult and can be further explored in future works. Finally, this paper is based on two waves of innovation surveys in 2011 and 2015, and the future extension of long-time series data can provide more valuable policy insights for the eco-innovation in China.

Declarations
Ethical Approval: The manuscript does not report on or involve the use of any animal or human data.
Consent to Participate: Not applicable.