How Can a Firm Innovate When Embedded in a Cluster?—Evidence from the Automobile Industrial Cluster in China

In the era of the knowledge economy, knowledge management is increasingly important. Knowledge management ability is one of the core factors influencing enterprise competitiveness, affecting innovation performance and sustainable development. To test the impact mechanism of the knowledge management of enterprises on innovation performance, a multilevel structural equation model was established using data from the automobile industry in China, with “knowledge management” (KM) as the independent variable, the three dimensions of absorptive capacity as the mediating variables, and “innovation performance” (IP) as the dependent variable at the firm level. At the cluster level, the innovation milieu of the cluster was introduced into the model. The results show that the three dimensions of absorptive capacity all significantly mediate the relationship between knowledge management and innovation performance. The innovation milieu of the cluster had a direct cross-level effect on the innovation performance of enterprises and a positive cross-level moderated effect on the relationship between explorative learning and innovation performance. These results support the promotion of enterprise innovation ability and the creation of an innovation milieu in the automobile industry in China.


Introduction
In the era of the knowledge economy, innovation ability is a key factor for enterprises thriving and withstanding the fierce market competition.Enterprise innovation involves a series of reforms in products, technologies, management measures, and marketing strategies.The innovation ability of enterprises is closely related to their knowledge management (KM) level, absorption ability, and external innovation atmosphere [1][2][3].Enterprises provide a knowledge base for innovation activities through a series of knowledge management activities, including knowledge acquisition, generation, storage, sharing, and updating.Enterprises can only perform low-level imitative innovation activities if their absorptive capacity is insufficient, making it difficult to form their own core competitiveness.The enterprises need to internalize external knowledge into their own innovative achievements via explorative learning (EL), transformational learning (TL), and developmental learning (DL) if they want to improve their innovation level.Furthermore, isolated enterprises struggle to ensure sustainable development in the highly segmented industrial chain.The vast majority of enterprises are embedded in industrial clusters.They can exchange knowledge and information by interacting with the cluster environment and other enterprises to improve their innovation ability.
In recent years, related studies focused on the impact mechanism of knowledge management and absorptive capacity on innovation performance.However, most of them only considered the impact of firm-level variables on innovation performance, lacking an analytical perspective on the interaction between firms and clusters.The studies on the relationship between industrial clusters and innovation often focused on cluster factors, ignoring micro-enterprise characteristics.From a methodology perspective, there is a nested relationship between enterprises and clusters that violates the principle of error independence if only firm-level data are analyzed [4].This greatly reduces the credibility of the results because the estimation results are inaccurate when using classical regression models under these circumstances [5].
In this study, a multilevel structural equation model was established in order to produce more credible results, with "knowledge management" as the independent variable, "explorative learning", "transformational learning", and "developmental learning"-the three dimensions of absorptive capacity-as the mediating variables, and "innovation performance" as the dependent variable at the firm level.At the cluster level, the innovation milieu of the cluster was introduced into the model to study its cross-level impact mechanism on innovation performance.We found that the three dimensions of absorptive capacity-explorative learning, transformational learning, and developmental learning-all significantly mediate the relationship between knowledge management and innovation performance.We also found that the innovation performance of enterprises could be affected by the innovation milieu, which had a positive cross-level moderated effect on the relationship between explorative learning and innovation performance.
The contributions of this paper are as follows.First, from the perspective of both the cluster and firm, we innovatively found a cluster-firm interactive innovation mechanism, which expanded previous theories on the innovation performance of the firms.Second, we found that the innovation milieu of clusters has important cross-level effects on the innovation performance of the firms, including a direct effect on the innovation performance and a positive cross-level moderating effect on the relationship between explorative learning and enterprises' innovation performance.Third, this paper sheds light on the policy-making of the automobile industrial cluster in China, which should encourage them to form a strong and open innovation atmosphere.
The rest of this paper is structured as follows.Section 2 provides a review of relevant papers on industrial cluster and innovation, knowledge management, absorptive capacity, innovation performance, and innovation milieu.Section 3 introduces the automobile industry in China, questionnaire design, and sample.Section 4 includes the analysis of the multilevel structural equation model.We discuss the results in Section 5.The conclusions and implications are drawn in Section 6.

Industrial Cluster and Innovation
The agglomeration of economic activities is the focus of economics, geography, and management [6][7][8][9].Marshall introduced the concept of "industrial district" and found that firms were clustered together due to input-output relationships, sharing labor pool, and knowledge spillover [6].Subsequently, many scholars summarized and analyzed the phenomenon of agglomeration.Based on this, Porter introduced the concept of the "industrial cluster" into the analysis of international competitiveness, thinking that industrial clusters can reduce transaction costs, create knowledge, and promote the formation of new firms [10].This marked the beginning of exploratory research on the relationship between industrial clusters and innovation.Numerous scholars preliminarily explored the innovation mechanism of industrial clusters from the perspective of knowledge spillover.Audretsch and Feldman found that innovation activities tend to cluster in the industry where knowledge spillover played a decisive role [11].Lawson and Lorenz believed that tacit knowledge is difficult to transfer.Hence, tacit knowledge is regarded as the basic element of regional sustainable competitiveness [12].
Early studies mainly explored the innovation mechanism of industrial clusters from a regional perspective, ignoring micro-firm characteristics.The innovation behavior of industrial clusters is the result of the interaction between knowledge flow and the innovation ability of the firms.The learning and absorptive ability of firms cannot be overlooked, which are key factors to innovation [13].Therefore, the research focus shifted to knowledge management and multi-agent interaction.Based on knowledge management theory, Bathelt et al. found that both tacit and coding knowledge can be exchanged through the local buzz and global pipelines [14].Giuliani indicated that the highest-level enterprises in industrial clusters played the role of a knowledge gatekeeper in Chile's experience [15].Guo and Guo found that innovation institutions played multiple roles in the knowledge system of industrial clusters, including technology gatekeepers, technical problem solvers, and innovation resource integrators [16].
Scholars have gradually begun to pay attention to the impact of knowledge management capability of firms and cluster environment on the innovation performance of firms.Lai et al. reported that the knowledge management capability of firms could significantly improve their innovation performance [17].Müeller and Jungwirth stated that the contextual, structural, and function determinants all played roles in the performance of specific clusters [18].
In summary, the related papers focus on either firm or cluster factors, lacking a synergistic analysis.Hence, we establish a multilevel structural equation model including both firm variables and cluster variables to explore the collaborative innovation mechanism of the firms embedded in clusters.

Knowledge Management
In the era of the knowledge economy, the intangible intellectual capital is more important than tangible assets [19].Knowledge management is regarded as a new wealth of firms [20].Knowledge management includes knowledge acquisition, creation, storage, sharing, and updating, which influences the innovation ability of firms [21].Knowledge management can promote communication between firms and increase knowledge benefits [17], and is also conducive to the creation of new knowledge through interaction between firms [21][22][23].The knowledge management capability of firms also affects their innovation performance [24][25][26].

Absorptive Capacity
Absorptive capacity was introduced by Cohen and Levinthal [27], which was defined as the business capability to acquire, digest, transform, and develop new external knowledge.Lane et al. deemed that absorptive capacity was mainly composed of the ability to understand and absorb external knowledge [28], which was a necessary dynamic capability for the sustainable development of enterprises [29].Lane et al. illustrated that absorptive capacity included explorative learning, transformational learning, and developmental learning [30].Absorptive capacity reflects the firm's ability to identify, assimilate, and exploit knowledge from its environment [2].Absorptive capacity prompts enterprises to exchange existing knowledge and combine it with new knowledge to promote innovation [31].Some scholars hold the view that absorptive capacity should be seen as a special capacity, which is different from knowledge management [32] and reflects a real capacity to transfer knowledge [32].

Innovation Performance
The innovation performance of firms involved the innovative achievement after internalizing and absorbing knowledge.Early papers on innovation performance mainly focused on product innovation performance [33,34].Then, studies presented multi-dimensional characteristics, including product innovation, process innovation, organizational innovation, service innovation, and other aspects.We measured innovation performance from four aspects: The number of patents, the success rate of new product research and development, the technical content of new products, and the proportion of new product revenue based on previous papers [35,36].

Knowledge Management and Absorptive Capacity
Absorptive capacity is regarded as a bridge for enterprises to create new knowledge through knowledge management activities [37].Enterprises improve their knowledge spillover tendency by learning external knowledge to enhance their absorptive capacity [27].Enterprises deepen their understanding of the value of knowledge via the management of internal and external knowledge [38].Absorptive capacity is a social interaction mechanism and a re-creation of existing achievements based on knowledge management [29,39].How to improve their absorptive capacity through knowledge management activities is the key to maintaining the core competitiveness of enterprises [40].Knowledge acquisition, transfer, sharing, and other knowledge management activities are all closely related to absorptive capacity [31,41,42].Flor et al. [31] pointed out that external knowledge search could influence the firm's absorptive capacity, which consists of potential and realized absorptive capacity.The former has a positive effect on the linkage between external search breadth and radical innovation, while the latter moderates the impact of external search breadth.Camisón and Forés [41] put forward three mechanisms to explain that a firm's particular knowledge creation process will lead to specific absorptive capacity.The first is that the diversity of knowledge base could provide sufficient frames of reference and standards to obtain better ideas and valuable tacit knowledge.The second is that it will improve a firm's abstract ability to assimilation and integration of new information.The third is that it will build new relationships between new knowledge and existing concepts.These mechanisms provide possibilities for enhancing a firm's absorptive capacity.Bloodgood [43] deemed that a firm could enhance its own competitiveness via incremental imitation, radical imitation, incremental innovation, and radical innovation based on absorptive capacity to acquire different kinds of knowledge.Rafique et al. [44] found that knowledge sharing is an important factor for developing absorptive capacity, which could create new ideas after assimilating the existing knowledge and expressing them clearly within the firm.Soto-Acosta et al. [45] discovered that the knowledge management capacity of a firm will influence the explorative and exploitive innovation process through the mediated effect of absorptive capacity.The knowledge management process of a firm could provide a diverse knowledge base and network relationships for developing absorptive capacity.
Therefore, the following hypotheses are proposed: Hypothesis 1 (H1).The knowledge management level of enterprises has a significantly positive impact on explorative learning.

Hypothesis 2 (H2).
The knowledge management level of enterprises has a significantly positive impact on transformational learning.

Hypothesis 3 (H3).
The knowledge management level of enterprises has a significantly positive impact on developmental learning.

Absorptive Capacity and Innovation Performance
Numerous papers indicated that explorative learning, transformational learning, and developmental learning all affect the innovation performance of enterprises [38,46,47].Enterprises need to possess a certain absorptive capacity in order to internalize external knowledge into the resources needed for innovation activities.Hence, the absorptive capacity of enterprises can significantly affect their innovation performance [29,[48][49][50][51][52][53][54].Chaudhary and Batra [55] put forward that absorptive capacity helps the firms form proper knowledge strategies to enhance innovation performance.These firms with better abilities to acquire and assimilate knowledge could form a stronger entrepreneurial atmosphere, which is positive for innovation.Albort-Morant et al. [56] found that the absorptive capacity of firms, obtaining knowledge from others and combining it with existing knowledge via novel methods, is a key factor to improve the effectiveness of innovation.Kohlbacher et al. [47] deemed that absorptive capacity has a positive effect on both exploitative and explorative innovation, and the firms should develop the absorptive capacity to gain the fruits of agglomeration effects.Lau and Lo [49] pointed out that absorptive capacity could improve a firm's innovation performance via knowledge acquisition, assimilation, transformation, and exploitation.Tzokas et al. [50] found that absorptive capacity could lead to better performance in terms of new product development, market performance and profitability, and form strong relationships with customers to gain new knowledge.
Meanwhile, many studies also indicated that absorptive capacity mediates the relationship between knowledge management and innovation performance [27,29,48,52,53,57].Ferreras-Méndez et al. [48] found that absorptive capacity, which is of great significance to assimilate existing valuable knowledge and increase innovation performance, has a full-mediated effect between external knowledge search and innovation performance.Xie et al. [58] discovered that both knowledge transformation and knowledge exploitation capacity have a mediated effect between knowledge acquisition and innovation performance.
Therefore, the following hypotheses are proposed: Hypothesis 4 (H4).The explorative learning capacity of enterprises has a significantly positive impact on innovation performance.

Hypothesis 5 (H5).
The transformational learning capacity of enterprises has a significantly positive impact on innovation performance.

Hypothesis 6 (H6).
The developmental learning capacity of enterprises has a significantly positive impact on innovation performance.

Hypothesis 7 (H7).
The explorative learning capacity of enterprises mediates the relationship between knowledge management and innovation performance.

Hypothesis 8 (H8).
The transformational learning capacity of enterprises mediates the relationship between knowledge management and innovation performance.

Hypothesis 9 (H9).
The developmental learning capacity of enterprises mediates the relationship between knowledge management and innovation performance.

Cross-Level Effect of Innovation Milieu
The vast majority of enterprises are embedded in industrial clusters.They are inevitably affected by their environment.Numerous scholars focus on the integration of cluster and firm perspective to study a firm's innovation performance in a regional environment [49,[59][60][61].Lau and Lo [49] verified the positive impact of regional innovation initiatives on innovation performance using the single level structural equation model [4], which violates the principle of error independence from the perspective of methodology.In addition, some papers focused on the impact of innovation milieu within firms on innovation performance [1,62,63].However, papers seldom pay attention to the mechanism of the cross-level impact of the innovation milieu of the cluster on innovation performance using the multilevel equation model, despite its importance.How innovation milieu as an environmental variable impacts a firm's innovation performance?First, the innovation milieu of the cluster may directly affect the innovation performance.Second, the innovation milieu of the cluster may interact with firm characteristics, which then influences innovation performance.Third, from the perspective of methodology, more accurate results will be obtained if the multilevel structural equation model is adopted.
To explore this complicated cross-level impact mechanism, the following hypotheses are proposed: Hypothesis 10 (H10).The innovation milieu of the cluster has a significantly positive cross-level impact on innovation performance.

Hypothesis 11 (H11).
The innovation milieu of the cluster moderates the relationship between explorative learning capacity and innovation performance.
Hypothesis 12 (H12).The innovation milieu of the cluster moderates the relationship between transformational learning capacity and innovation performance.
Hypothesis 13 (H13).The innovation milieu of the cluster moderates the relationship between developmental learning capacity and innovation performance.
In summary, the study framework is shown in Figure 1.

Hypothesis 11 (H11):
The innovation milieu of the cluster moderates the relationship between explorative learning capacity and innovation performance.

Hypothesis 12 (H12):
The innovation milieu of the cluster moderates the relationship between transformational learning capacity and innovation performance.

Hypothesis 13 (H13):
The innovation milieu of the cluster moderates the relationship between developmental learning capacity and innovation performance.
In summary, the study framework is shown in Figure 1.

Study Industry
China's automobile industry cluster, including six clusters-the northeast, central and southwest region, Beijing-Tianjin, Yangtze River Delta, and the Pearl River Delta region-was chosen as the study industry, which is characterized by a high degree of agglomeration.In terms of production and sales, China has become the largest automobile producer and seller in the world, with automobile production and sales reaching 30 million and 29 million vehicles in 2017, respectively.However, from the perspective of innovation ability, the automobile industry is at the low end of the global automobile industry chain.The independent development ability is too weak, and the core technology and market are dominated by multinational companies.The automobile industry occupies a pivotal position in the whole industrial system of a country.Hence, it is of practical significance to study the impact mechanism of innovation performance of automobile enterprises.

Study Industry
China's automobile industry cluster, including six clusters-the northeast, central and southwest region, Beijing-Tianjin, Yangtze River Delta, and the Pearl River Delta region-was chosen as the study industry, which is characterized by a high degree of agglomeration.In terms of production and sales, China has become the largest automobile producer and seller in the world, with automobile production and sales reaching 30 million and 29 million vehicles in 2017, respectively.However, from the perspective of innovation ability, the automobile industry is at the low end of the global automobile industry chain.The independent development ability is too weak, and the core technology and market are dominated by multinational companies.The automobile industry occupies a pivotal position in the whole industrial system of a country.Hence, it is of practical significance to study the impact mechanism of innovation performance of automobile enterprises.

Questionnaire Design
The questionnaire in this study included two parts.The first covered information about the enterprises under investigation.A five-point Likert scale, ranging from "strongly disagree" (1) to "strongly agree" (5), was adopted to measure knowledge management, absorptive capacity, and innovation performance.The measurement items for each variable were adapted according to the mature scale used in prior studies.A pre-survey was conducted to test the applicability of questionnaires.The final questionnaire was formed after revising several words according to the feedback of the enterprises.
The first part included contact addresses, the number of employees, annual revenue, and age of these enterprises.The second part included 21 items.Five items measured knowledge management (KM), which were adapted from Lai et al. [17].The items were as follows: KM1: The enterprise has coded the collected information and established a knowledge system, KM2: The enterprise records and integrates work knowledge in the form of a database, KM3: The enterprise can completely preserve professional technology and work knowledge, KM4: The enterprise has a perfect management mechanism of professional technology and knowledge, and KM5: The enterprise has established a special knowledge feedback mechanism to improve future performance.Twelve items measured absorptive capacity, including explorative learning, transformational learning, and developmental learning, which were adapted from previous papers [64,65].The items for explorative learning (EL) were as follows: EL1: The enterprise has institutions for screening external knowledge, information and technology.EL2: The enterprise can widely search for the latest market information and technology.EL3: The enterprise can quickly assess and acquire the latest market information and technology.EL4: The enterprise often communicates with partners.The items of transformational learning (TL) were as follows.TL1: The enterprise organizes new knowledge, TL2: The enterprise can quickly understand and transform new knowledge, TL3: The enterprise can discover the value of acquired new knowledge, and TL4: The enterprise integrates original knowledge with new knowledge.The items of developmental learning (DL) were as follows: DL1: The enterprise pays attention to better ways that make use of acquired new knowledge and technology, DL2: The enterprise effectively applies acquired new knowledge and technology to new product development, DL3: The enterprise has established systems of using external knowledge, information, and technology, and DL4: The knowledge that the enterprise has digested and used can effectively enhance its competitive advantages.Four items measured innovation performance (IP), which were adapted from previous studies [66,67]: IP1: The number of enterprise patents has increased significantly, IP2: The success rate of new product development of the enterprise has significantly improved, IP3: The technology content of new products is higher, and IP4: The proportion of new product revenue increases year over year.
The data used to measure cluster innovation milieu included patent data provided by the China National Knowledge Internet.We used the application programming interface (API) provided by Google Maps (Mountain View, CA, USA) to obtain the longitude and latitude of these patents according to the contact address provided in the patent description.Then, we aggregated all patents at the district and county levels.We regarded the number of patents in the districts or counties where the enterprises were surveyed as the innovation milieu of the cluster in which the enterprises were located.

Sample
The questionnaire was issued from 1 September to 15 November 2018, which included four stages.In the first stage, we obtained the list of automobile companies from the China automobile statistical yearbook.Then, we used the web crawler technology to obtain the emails of these enterprises on Qichacha.com,a website providing business information of Chinese enterprises.In the second stage, we numbered all of the automobile companies.Then, we obtained 1000 random numbers via a random number generator.In the third stage, we sent the questionnaire to their emails and set some reverse questions to eliminate unqualified questionnaires.In the last stage, we deleted some invalid questionnaires.We collected 768 questionnaires with a recovery rate of 76.8%.Of these, 658 valid questionnaires were obtained with a valid recovery rate of 85.7%.In general, to establish an effective structural equation model, the sample number should be ten times greater than the number of items [68].Therefore, the data obtained in this study satisfied this requirement.The specific data composition is shown in Table 1.

Analysis of Normality and Common Method Bias
The normality of the data was analyzed via the kurtosis and skewness tests, which showed that the data were distributed normally.We tested several methods including the partial correlation method, the multi-trait-multimethod model, and Harman's single factor test to avoid the impact of common method bias [69,70].The results indicated that there was no common method bias in the model.

Exploratory Factor Analysis
We adopted the principal component method for factor rotation.The final results are shown in Table 2.The factor loading of all items were greater than 0.6, the items for each variable were all gathered together (see the bold figures in Table 2), and the eigenvalues of the five components were all greater than one, which means that the data in this study had good construct validity.

Reliability and Validity Analysis
We used Cronbach's α to test the reliability of the data.The Cronbach's α values of all variables were greater than 0.7 (Table 3), which indicated that the data were reliable.We used the composite reliability (CR) and average variance extracted (AVE) to test the convergent and discrimination validity.Table 3 shows that the CR values were all greater than 0.7, and the AVE values were all greater than 0.5, which meet the criteria proposed by Hair et al. [71].Fornell and Larcker [72] deemed that a variable has a good discrimination validity if the root mean square of the AVE value of the variable is greater than its correlation coefficients with other variables.Table 4 shows that each variable had good discrimination validity.

Model Fit Analysis of the Structural Equation Model at the Firm Level
We used the chi-square divided by degrees of freedom (CMIN/DF), standardized root mean squared residual (SRMR), root mean square error of approximation (RMSEA), comparative fit index (CFI), and Tucker Lewis index (TLI) to test the model fit.The results are shown in Table 5.All the indicators met the classic criteria.Hence, the model fit was sufficiently good.Note: CMIN/DF is the chi-square divided by degrees of freedom, SRMR is the standardized root mean squared residual, RMSEA is the root mean square error of approximation, TLI is the Tucker Lewis index, and CFI is the comparative fit index.

Path Analysis of Multilevel Structural Equation Model
This section includes four subsections.In Section 4.4.1, the null model is established for analyzing the heterogeneity between groups.Then, we determined whether this study is suitable for multilevel model analysis.In Section 4.4.2, a structural equation model is established with only firm-level variables.In Section 4.4.3,we tested the mediated effect of the structural equation model.In Section 4.4.4,a multilevel model is established to test the cross-level effect of innovation milieu on the innovation performance.

Null Model
Establishing the null model involved calculating the intraclass coefficient (ICC), which is used to test the heterogeneity between groups of data.We used MPLUS7.4(Muthén & Muthén, Los Angeles, CA, USA) to obtain the ICC value.ICC was 0.756, which meant that the heterogeneity between groups could not be ignored.Hence, the multilevel model was more suitable.

Structural Equation Model at the Firm Level
The structural equation model at the firm level was established without considering the impact of cluster factors.The results are shown in Table 6.We can see that the standardized coefficient between knowledge management and explorative learning was 0.497 (p = 0.000), which means that the knowledge management level of enterprises has a significantly positive impact on explorative learning.Hence, the results support H1.The standardized coefficient between knowledge management and transformational learning was 0.375 (p = 0.000), which means that the knowledge management level of enterprises has a significantly positive impact on transformational learning.Hence, this supports H2.The standardized coefficient between knowledge management and developmental learning was 0.433 (p = 0.000), which means that the knowledge management level of enterprises has a significantly positive impact on developmental learning, supporting H3.The standardized coefficient between explorative learning and innovation performance was 0.370 (p = 0.000), which means that the explorative learning capacity of enterprises has a significantly positive impact on innovation performance.Hence, this supports H4.The standardized coefficient between transformational learning and innovation performance was 0.216 (p = 0.000), which means that the transformational learning capacity of enterprises has a significantly positive impact on innovation performance, supporting H5.The standardized coefficient between developmental learning and innovation performance was 0.251 (p = 0.000), which means that the developmental learning capacity of enterprises has a significantly positive impact on innovation performance.Hence, the results support H6.

Mediated Effect Analysis
We adopted the product of coefficients method and the bootstrapping method, including the bias-corrected and percentile method, to test the mediated effect.From the aspect of product of coefficients, the specific mediated effect is significant if the Z value is larger than 1.96.From the aspect of bootstrapping, the specific mediated effect is significant if zero is not in the interval between the minimum and maximum.From Table 7, we can see that all the Z values were larger than 1.96, and zero was not in the interval between the minimum and maximum of any path.Therefore, the explorative learning capacity, transformational learning capacity, and developmental learning capacity of enterprises all mediate the relationship between knowledge management and innovation performance.These results support H7, H8, and H9.

Cross-Level Effect Analysis
In previous studies, cluster-level variables were regarded as equivalent to firm-level variables.There is a nested relationship between enterprises and clusters that violates the principle of error independence if only firm-level data are analyzed [4].The estimation results are inaccurate when using classical regression models [5].Hence, the multilevel model was needed to solve these problems.First, we tested the direct effect of the innovation milieu of the cluster on innovation performance.Then, we tested the cross-level moderated effect of innovation milieu of the cluster on the relationships between the three dimensions of absorptive capacity and innovation performance.The results are shown in Table 8.The coefficient of the direct effect of innovation milieu of the cluster on innovation performance is 0.254 (p < 0.01), which supports H10.M2 meant that the innovation milieu of the cluster significantly and positively moderates the relationship between explorative learning capacity and innovation performance (p < 0.05).This supports H11.M3 and M4 meant that the moderated effects of the innovation milieu of the cluster on the relationships between transformational learning capacity and innovation performance, and developmental learning capacity and innovation performance were not significant (p > 0.05).The results do not support H12 or H13.

Discussion
The knowledge management level of enterprises significantly affects their innovation.The results verified the importance of knowledge management on a firm's innovation performance again, which is consistent with the researches of Ferraris et al. [24], Hussinki et al. [25], and Obeidat et al. [26].It shows that the set of a perfect knowledge management system is still valuable and effective for the automobile firms in China, which can help them acquire, integrate, create, dissemination, and share knowledge to build up a knowledge base.In the era of the knowledge economy, knowledge as an intangible asset is more important than tangible assets [19].Enterprises turn knowledge into wealth through knowledge acquisition, creation, storage, sharing, updating, and other knowledge management activities.The importance of knowledge management has been recognized increasingly by more enterprises since it provides knowledge support to the innovation activities of enterprises.
The absorptive capacity of enterprises significantly mediates the relationship between knowledge management and innovation performance, which verified the researches of Cohen and Levinthal [27], Zahra and George [29], and Ferreras-Méndez et al. [48].Specifically, we found that the three dimensions of absorptive capacity-explorative learning, transformational learning, and developmental learning-all significantly mediated the relationship between knowledge management and innovation performance.In order to transform knowledge wealth into innovative achievements, enterprises also depend on their absorptive capacity.First, in the process of explorative learning, enterprises could find more new technologies and knowledge to enrich their knowledge base via communicating and interacting with suppliers, customers, competing enterprises, universities, scientific research institutions, and government departments.Second, enterprises with stronger transformational learning ability are better at transforming knowledge into innovative achievements.Third, developmental learning ability is the key to enterprises' independent innovation and is also an important index to test whether enterprises really understand new knowledge.The ultimate transformation of knowledge into innovative achievements of enterprises mainly depend on the developmental learning ability of enterprises.Therefore, for China's automobile industry, in order to improve the innovation performance of enterprises, it is necessary to not only strengthen the level of knowledge management, but also to strengthen the absorptive capacity of enterprises by encouraging employees to explore new knowledge, enhancing information communication between enterprises and others, standardizing the process of knowledge transformation, and cultivating employees' developmental learning ability.
The impact of knowledge management on innovation performance involves a cluster-firm interactive innovation mechanism, which partly supports the hypotheses in this paper.Previous studies focused more on the impact of firm-level factors on innovation performance [17,48,49], but we found that the impact of cluster-level factors on enterprise innovation performance cannot be ignored.The innovation milieu of the cluster had a cross-level direct effect on the innovation performance of the enterprises.The stronger the innovation milieu of a cluster, the better the innovation performance of enterprises, which sheds light on the integrative research of both cluster and firm perspective.That is to say, the innovation performance of a firm is the integrative outcome of its own characteristics and the external environment.Some scholars discussed the cluster factors on innovation performance using the single level structural equation model [4], which violates the principle of error independence from the perspective of methodology.Hence, a multilevel structural equation model was established in this paper to overcome this problem.From a practical point of view, creating a good innovation milieu for China's automobile industry cluster would be an effective measure to improve the innovation performance of these enterprises.The innovation milieu of the cluster also had a positive cross-level moderated effect on the relationship between explorative learning and enterprises' innovation performance.This indicates that the process of enterprises using their explorative learning ability to promote the transformation of new technologies and knowledge into innovative achievements was also affected by the innovation milieu of their clusters.A good innovation milieu could effectively speed up the transformation process and provide more possibilities for enterprises to acquire new technologies and knowledge.Therefore, the local government should strive to improve the innovation and entrepreneurship motivation and create a better innovation milieu for the automobile enterprises.
However, the innovation milieu did not have a cross-level moderated effect on the relationship between transformational learning and innovation performance or on the relationship between developmental learning and innovation performance, which does not support the hypotheses in this paper.This shows that the mechanisms of transformational learning and developmental learning on innovation performance are relatively simple, as they were not affected by the innovation milieu of the cluster.This also means that the main role of the innovation environment is to provide enterprises with opportunities to acquire more new knowledge and technologies to improve the diversity of their own knowledge.We did not find evidence that the innovation milieu of the cluster affected the transformational and developmental learning.All in all, we innovatively found a cluster-firm interactive innovation mechanism.It verified that the integrative research from both cluster and firm perspective, which was put forward by Romijn and Albaladejo, Caloghirou et al., and Yam et al. [59][60][61], is of great significance.It also expanded previous theories on innovation performance of firms which only includes the firm's characteristics using single level models.

Conclusions
A multilevel structural equation model was established using the data of the automobile industry in China, with knowledge management as the independent variable, explorative learning, transformational learning, and developmental learning-the three dimensions of absorptive capacity-as mediating variables, and innovation performance as the dependent variable at the firm level.At the cluster level, the innovation milieu of the cluster was introduced into the model.We verified the importance of knowledge management on the innovation performance of enterprises and explored the micro mechanism of knowledge management on innovation performance.We found that the three dimensions of absorptive capacity-explorative, transformational, and developmental learning-all significantly mediated the relationship between knowledge management and innovation performance.We also found that the innovation performance of enterprises could be affected by the innovation milieu.The impact of knowledge management on innovation performance was a cluster-firm interactive innovation mechanism.The innovation milieu of the cluster had a cross-level direct effect on the innovation performance of the enterprises.The innovation milieu of the cluster had a positive cross-level moderating effect on the relationship between explorative learning and enterprises' innovation performance.However, we did not find any evidence that the innovation milieu of the cluster affected transformational and developmental learning.
Considering the weak independent innovation ability of the automobile industry in China, we provide some recommendations.For China's automobile industry, in order to improve the innovation performance of enterprises, it is necessary to not only strengthen the level of knowledge management, but also to strengthen their absorptive capacity.Our results also clarify the role of the local government, which creates a better innovation milieu and promotes communication between enterprises.

Theoretical Implications
We innovatively found a cluster-firm interactive innovation mechanism, which verified the importance of the integrative research of both cluster and firm perspective and expanded previous theories on the innovation performance of firms.The innovation milieu of the cluster has a cross-level direct effect on the innovation performance of the enterprises and a positive cross-level moderating effect on the relationship between explorative learning and enterprises' innovation performance.It indicates that a cluster-firm interactive innovation mechanism exists, which mainly plays a role in the process of explorative learning, and does not affect the processes of transformational and developmental learning.These results shed light on the integrative research on innovation performance from both the cluster and firm perspective.We also found that the three dimensions of absorptive capacity all significantly mediated the relationship between knowledge management and innovation performance, which means that the existing theories on absorptive capacity still play an important role in explaining the relationship between knowledge management and the innovation performance of a firm.

Practical Implications
The practical implications of this paper are as follows: The automobile firms should not only strengthen the level of knowledge management but also strengthen their absorptive capacity.The automobile industry in China is at the low end of the global automobile industry chain due to its weak research and development (R&D) capability, which is a long-standing and urgent problem.To solve this problem, some measures should be taken at both the cluster and firm level.The weak research and development (R&D) capability is due to the lack of sufficient available knowledge.Therefore, these enterprises should not only strengthen the construction of the knowledge management system but also pay attention to improve their absorptive capacity.
The local government should provide policy support that creates a better industrial cluster innovation environment.The external conditions of weak research and development (R&D) capability were that the cluster had not yet formed a strong and open milieu of innovation.The local government should play a role in creating a better innovation milieu and promoting communication between enterprises, customers, universities, scientific research institutions, and government departments.Creating an innovative milieu is a long-term and complex task, which requires not only the guidance of government departments but also the collaborative participation of relevant enterprises, universities, research institutions, and consumers.
The central government should further expand the opening of the automobile market.Free competition is the cornerstone of innovation.The market access of the automobile industry should be relaxed including many policies, such as allowing foreign enterprises to enter the automobile market freely, reducing import tariffs on automobiles, encouraging foreign new energy automobile enterprises to invest in China, and reducing government subsidies for state-owned auto firms to stimulate them to innovate in order to maintain sustainable competitiveness.

Limitations and Further Research
We verified the indirect influencing mechanism of knowledge management on innovation performance, and the cross-level mechanism of the innovation milieu of the cluster on innovation performance, which has certain theoretical and practical significance.However, there are some shortcomings in this paper.First, the data from the automobile industry in China could reflect the innovation mechanism of these enterprises, but whether there are differences among industries requires further testing.Second, we used the number of patents as the indicator of innovation milieu, which has a certain degree of representativeness.However, the composition of innovation milieu is complex and diverse, including innovation culture, entrepreneurship, social inclusiveness of failure, etc.Whether other dimensions have a cross-level impact on innovation performance still needs further study.The cluster-firm interactive innovation mechanism is a meaningful research direction in the future.We will collect data from multiple industries to explore this mechanism.

Table 3 .
Reliability and convergent validity of overall data.
Note: FL is factor loading, SFL is standard factor loading, SE is standard error, *** p < 0.001.

Table 5 .
Model fit analysis.
Note: CI is confidence interval.