This study used fsQCA to study how the entrepreneurship policies of 31 provinces in China jointly affect innovation capability. QCA can fully compare and analyze cases using Boolean logic and algebra, it can explore the joint effect of the interaction of multiple antecedents on specific phenomena (Ragin, 2008). QCA studies involve learning about unknown facts by using what information people do know (Thomann & Maggetti, 2020); the present study follows this logic. According to the types of variables, QCA can be divided into three types: crisp-set QCA (csQCA), multi-value QCA (mvQCA), and fuzzy-set QCA (fsQCA). The reasons for choosing fsQCA are as follows. First, compared with csQCA and mvQCA, it adopts a higher consistency standard and has more stringent results (Rihoux & Ragin, 2008). In other words, fsQCA can better preserve the richness of the original data in a case (Herrmann & Cronqvist, 2009) and make the research results reproducible and transparent. Second, fsQCA is rarely used in social science, and existing studies (López-Cabarcos, Vázquez-Rodríguez & Piñeiro-Chousa, 2016) show that fsQCA is feasible when analyzing causal complexity in social science. It is consistent with the analytical logic of this study. Third, it is suitable for small sample research (Fiss, 2011; López-Cabarcos, Vázquez-Rodríguez & Quiñoá-Piñeiro, 2022), especially when using countries or regions as the analysis unit (Beynon, Jones & Pickernell, 2020), which is again suitable for this study's purpose. The following steps are followed in fsQCA: calibration, necessity analysis, and sufficiency analysis (Schneider & Wagemann, 2010).

Previous studies have used fsQCA to analyze the impact of entrepreneurship policy on innovation activities (Jabeur, Mefteh-Wali & Carmona, 2021; Xie, Wang, Xie, Dun & Li, 2021), but viewed entrepreneurship policy as a single antecedent. Different from previous studies, this study focuses on the interaction of various entrepreneurship policies and discusses how the combination of entrepreneurship policies affects RIC, which is an innovative approach.

Taking 31 provinces in China as the sample, this study used Python3, a data capture software, to retrieve data from the websites of provincial governments. The study collected continuous and effective entrepreneurship policies from July 27, 2017, to December 11, 2021 (from the date of document release to the date of obtaining the result variable data).

The conditional variable of this study comes from the idea of “strengthening the implementation of innovation driven development strategy and further promoting the in-depth development of mass entrepreneurship and innovation” issued by the State Council of China in 2017. Five antecedent conditions are set for the five policies: technology transfer policy (TTP), fiscal and tax policy (FTP), digital transformation policy (DTP), talent policy (TP), and government innovation management policy (GIMP). This study takes the specific policy terms as the basic analysis unit and encodes them according to the five entrepreneurship policies. After excluding invalid clauses, the policy clauses are counted.

The RIC index used in this study comes from China's regional development evaluation report, titled “Evaluation Report of China's Regional Innovation Capability in 2021”. With the support of the Ministry of Science and Technology of China, the report was prepared by the China Science and Technology Development Strategy Research Group in conjunction with the University of the Chinese Academy of Sciences. The report establishes a four-level index system as the main evaluation method. The first level index includes knowledge creation, knowledge acquisition, enterprise innovation, innovation environment, and innovation performance, reflecting enterprise innovation, basic research and original innovation, achievement transformation, etc. in all regions of China. Table 1 provides more details on the outcome and conditions.

Under fsQCA, calibration is the process of assigning collective membership to cases, where the calibrated collective membership is between 0 and 1 (Ragin, 2008). Commonly used calibration methods may be either direct or indirect (Pappas & Woodside, 2021). For direct calibration, exactly three qualitative breakpoints are needed to define the membership levels of full input, intermediate, and full output in the fuzzy set. For indirect calibration, there is the need to rescale the measurement range accordingly. Studies usually use three thresholds of 0.95, 0.5, and 0.05 for direct calibration (Ragin, 2008). However, the setting of a threshold should be adjusted according to the situation and should not be selected mechanically (Pappas & Woodside, 2021). Because the data in this study do not completely follow a normal distribution and are slightly skewed, if calibrated according to the above threshold, the effectiveness of the configuration will be reduced. Referring to existing studies (Fiss, 2011; Romero-Castro, López-Cabarcos & Piñeiro-Chousa, 2022), this study sets the full membership points of five antecedents and one result variable as the upper quartile (75%) of the case data, the intersection points as the median (50%) of the case data, and the complete non-membership points as the lower quartile (25%). See Table 2 below for the anchor point calibration.

The fsQCA software was used to detect the necessary conditions, and the results are shown in Table 3. The necessary condition facilitates the occurrence of an event. According to the literature (Amara, Rhaiem, & Halilem, 2020; Beynon et al., 2020), the consistency threshold of the necessity analysis is set at 0.9. Table 3 lists the results of the necessity analysis. The necessary consistency of the single antecedent condition on high or low RIC is below 0.9, indicating that no single policy type is necessary for high or low RIC. In other words, a single policy category has insufficient explanatory power for high RIC. Therefore, it is necessary to further analyze the combined effect of antecedent conditions.

After determining that a single antecedent condition does not constitute a necessary condition for high RIC, the study further analyzes the conditional combination of the five conditional variables to obtain the entrepreneurship policy combination path. This study adopts the consistency threshold in the range of 0.75–0.85, as established by Ragin (2006), or the natural discontinuity value of consistency score for truncation (Crilly, Zollo & Hansen, 2012). Based on Ragin (2006), this study sets the consistency threshold to 0.75 and the case frequency threshold to 1 to obtain the truth table. The core solution is generally used to determine the number of configurations and antecedents, while the reduced solution is used to determine the intermediate conditions. According to Fiss (2011), the condition appearing in the reduced solution is the core condition, indicating a strong causal relationship with the result variable. The condition that appears in the core solution, but not in the reduced solution, is the peripheral condition, which has a weak relationship with the result variable. Additionally, some definitions need to be provided as follows. Consistency refers to the degree to which a case corresponds to the set-theoretic relationship expressed in the solution (Fiss, 2011). It is an important indicator to measure the necessary conditions. Coverage indicates how many cases have high membership in the outcome condition (Skarmeas, Leonidou & Saridakis, 2014). Coverage measures the degree to which the subset covers the target set and is used to measure the empirical correlation of the necessary conditions. Coverage includes raw coverage, unique coverage, and solution coverage (Ragin, 2008). Raw coverage refers to the proportion of cases covered by a given configuration; unique coverage refers to the degree to which a single configuration explains the results after excluding the common parts with other configurations; solution coverage refers to the proportion of cases in the outcome that is explained by the complete solution. The results show that there are three configurations (see Table 4) that can produce high RIC, namely H1, H2, and H3. Since the core conditions of H1a and H1b are the same, they constitute a second-order equivalent configuration. Additionally, the coverage of the solution is 0.652, indicating that the solution covers most cases with high RIC.

Configurations H1a, H1b, H2, and H3 are the combinations that yield high RIC. Configurations H1a and H1b are second-order equivalent configurations. According to Configuration 1a, 50.3% of the cases suggest that RIC is high when a region has a large amount of TTPs, FTPs, DTPs, and GIMPs. Among them, TTP and GIMP are the core conditions, and they play the most important roles. FTP and DTP are peripheral conditions that also affect high RIC. In Configuration H1a, TP has no effect on the outcome. The consistency score is 81.5%. According to Configuration 1b, 50.3% of the cases indicate that high RIC is achieved when there is a large number of TTPs, FTPs, TPs, and GIMPs. TTP and GIMP are core conditions, and they are most important for high RIC. FTP and TP are peripheral conditions and do not play a major role. DTP has no effect on the outcome. The consistency score is 85%.

According to Configuration H2, 51.7% of the cases show that RIC is high when there is a large number of FTPs, DTPs, TPs, and GIMPs. GIMP is the core condition, and it plays the most important role. FTP, DTP, and TP are peripheral conditions and do not play a major role. TTP is not significant for the results. The consistency score is 82.6%.

According to Configuration H3, 10.5% of the cases in this study suggest that RIC is high even if it lacks FTPs, DTPs, TPs, and GIMPs, as long as it has enough TTPs. The consistency score is 91.8%.

There are three common robustness tests: changing calibration, changing the consistency threshold, and adding and deleting cases (Schneider & Wagemann, 2012). In this study, the consistency threshold is increased to 0.8, the frequency threshold is still 1, and the configuration obtained from the analysis remains unchanged, which shows that the analysis results are reliable.

Many studies cited in this paper showed an obvious positive correlation between the transformation of scientific and technological achievements and regional innovation capability (Gong et al., 2019; Lin et al., 2020; Peng et al., 2020). This study builds on these studies by investigating the impact of TTP on RIC and how TTP relates to other types of entrepreneurship policies. Specifically, the intellectual property and marketization policies of scientific and technological achievements in TTP promote the transformation of scientific and technological achievements of enterprises into real productivity. This corroborates previous research (Guan, Mok, Yam, Chin & Pun, 2006) in that technology transfer is a key factor in innovation performance and economics. Chinese enterprises should focus on technology absorptive capacity (Cohen & Levinthal, 1989), and the government should introduce more policies to improve the ability of enterprises in this regard.

According to some studies (Dai & Chapman, 2022), tax incentives for high-tech enterprises stimulate their R&D and help emerging economies achieve innovation-led economic growth. The results also suggest that FTP is related to RIC as it helped companies stimulate more aggressive and substantive innovation activities, especially small and medium-sized enterprises that have been hit by the COVID-19 pandemic.

Some studies have used panel measurement models to examine the relationship between digital transformation and firm performance (Peng & Tao, 2022). They believed that digital transformation could stimulate business innovation. Our results agree with them. The results show that high RIC can be produced when DTP is combined with TTP, FTP, and GIMP. Therefore, the results support propositions 1, 2, and 3. These conditions positively affect RIC.

According to the literature cited in this paper, talent plays an important role in innovation activities (Bajwa et al., 2017; Chen et al., 2020; Wang, 2020). As the results of this study show, provinces with a large amount of TP have high RIC. In addition, research on corporate green innovation (Lian, Xu & Zhu, 2022) showed that, limited by factors such as talent, compared to large companies, it was more difficult for small companies to conduct green innovation. Therefore, the government can retain high-quality talent by creating employment opportunities and a good living environment and help small and medium-sized enterprises to gather talent to improve regional innovation capability. Therefore, proposition 4 is supported.

Government innovation management is related to RIC (Oluwatobi et al., 2015). Research pointed out that the construction of digital government promotes enterprise innovation, allowing firms to interact with the government on digital innovation issues (Wang, Li, Tian & Hou, 2022). This study agrees with this view. The results show that provinces with a large number of GIMPs have high RIC. The government facilitates enterprise innovation by simplifying government approval procedures and building a digital government. Therefore, proposition 5 is supported.

The findings of this study enrich theoretical research on RIC. Specifically, this study examines how a mix of entrepreneurship policies affects Chinese RIC. Consistent with previous research that entrepreneurship policies can drive innovation (Oluwatobi et al., 2015; Guo & Niu, 2021; Hao & Zhang, 2021; Liu & Bai, 2021; Liu, Xu, Zhao, Xiao & Li, 2021), this study clearly identifies three entrepreneurship policy combinations. Inspired by research on RIS in East Asian economies (Yoon et al., 2015), this study explores the entrepreneurship policy mix that promotes RIS entrepreneurship in East Asian economies, and uses data to confirm the transition of RIS to mature RIS in China, thus further refining previous research findings. Finally, this study confirms the important role of TTP and GIMP on RIC enhancement. In conclusion, this study echoes previous research related to the intersection of innovation and entrepreneurship concepts, that is, further research on the intersection of innovation, entrepreneurship and knowledge (IEK) will help policymakers to develop more effective policy measures to promote regional growth (Piñeiro-Chousa et al., 2020).

Another important contribution of this study is related to the research method and the scope of the research object. From a methodological point of view, previous research regards entrepreneurship policy as a single antecedent condition and does not consider the combined effect of different policies. This has imposed certain limitations on our understanding of the role of policies. This study uses fsQCA to study the comprehensive effects of five types of entrepreneurship policies and reveals the role of different entrepreneurship policy combinations in promoting knowledge flow and improving innovation capability. As an extension of previous research, it especially emphasizes the important role of GIMP. In terms of the scope of research objects, previous studies on China's regional innovation capability are confined to some regions (He, Li, Zhang & Wang, 2018; Li, He & Zhao, 2020). This study takes policy texts from 31 provinces in China as the research object, covering both economically developed and underdeveloped regions, and the results are therefore more comprehensive.

This study has practical management significance for multiple regional policy makers. First, economically developed provinces (e.g., Guangdong and Hubei) or underdeveloped provinces (e.g., Shaanxi and Anhui) can achieve high RIC through different paths. When formulating entrepreneurship policies, policymakers should fine-tune the policies in line with the local economic and technological level to narrow the regional innovation gap and build an innovative country.

Second, in economically underdeveloped areas, policymakers should formulate policies to transform scientific and technological achievements, including the transformation of achievements, and the protection of intellectual property rights, to prevent the loss of innovation resources to other areas. In economically underdeveloped areas, the market demand and interest drive are insufficient, and the accumulation of innovation is weak (Xu, Qiu, Jin, Cheng & Zhu, 2022). TTP can alleviate the loss of innovation resources and promote new innovations. Third, policymakers should not ignore the role of GIMP, especially in East Asian economies that are deeply influenced by government efficiency and decision-making. Such countries should strive to simplify government processes and create a culture of innovation, which is conducive to creating an effective and transparent innovation environment.

This study has some limitations. First, the policies captured in this study are not comprehensive enough. Some entrepreneurship policies may only be available offline or deleted due to expiration, and researchers may not be able to access all entrepreneurial policy texts. This is a reminder for future researchers to fully consider various obstacles before obtaining policy samples and to try to overcome them to obtain comprehensive samples.

Second, in terms of research methodology, due to the lack of time contingency factors in fsQCA (Beynon et al., 2020), this study only discusses the static impact of entrepreneurship policy on RIC. Future studies could include time-varying factors to explore the dynamics of RIC. This study uses fsQCA to confirm the synergistic effect of entrepreneurship policies on high RIC, but the results do not show the strength of the impact of each policy. This could be measured by other methods in future studies. This study focuses on configurations that produce high RIC and does not present configurations that produce low RIC. However, the results related to low RIC configuration can also provide lessons for policymakers. Therefore, studying the entrepreneurship policy configuration of low RIC in the future could have practical significance.

Third, the role of other factors is not carefully considered in this study. Therefore, for a more in-depth analysis of RIC, future research should fully consider the relationship between various factors such as culture, institutional background, educational policy, and regional development.