As a researcher focusing on strategic management and green innovation, I have spent the past months investigating how digital intelligence, which I conceptualize as the convergence of digitalization and intelligent technologies, reshapes the innovation trajectory of EV battery enterprises. My interest in this topic stems from both the unprecedented growth of China’s new energy vehicle industry and the mounting pressure that EV battery producers face in maintaining environmentally sustainable competitiveness. In the following article, I report my research findings with a dual analytical strategy: first, a quantitative panel-data econometric analysis of the influencing mechanisms, and second, a fuzzy-set Qualitative Comparative Analysis (fsQCA) of the configurational pathways toward high sustained green innovation. Throughout my analysis, my central focus remains on EV battery enterprises.
1. Introduction
EV battery enterprises constitute the core of the new energy vehicle supply chain, and their capacity for sustained green innovation determines whether the transportation sector can achieve carbon neutrality. According to industry statistics, China produced roughly 7.06 million new energy vehicles in 2022, ranking first globally for eight consecutive years. Behind this remarkable achievement stands the EV battery industry. In 2022, domestic EV battery production increased by 148.5 percent year-on-year, demonstrating both market expansion and technological progress. Nonetheless, I observe several persistent vulnerabilities: more than 70 percent of lithium resources rely on imports; energy density is approaching physical limits; extreme-environment performance remains insufficient; and international competitors have enacted protectionist policies in response to China’s rising dominance.
Confronted with these vulnerabilities, I believe that digital intelligence may offer a critical solution. I define digital intelligence as the advanced stage of digitalization in which intelligent technologies—including big data, cloud computing, artificial intelligence, the Internet of Things, and 5G—not only collect and transmit data but also generate insights and enable autonomous decision-making. For EV battery enterprises, digital intelligence could transform production, R&D, supply chain coordination, and end-of-life recycling. Yet my literature review reveals a noticeable gap: while many studies discuss digital transformation and corporate innovation separately, few integrate these concepts to examine sustained green innovation specifically in EV battery enterprises. Moreover, existing work seldom combines regression-based mechanism testing with configurational pathway analysis. My dissertation therefore addresses two research questions:
- Through which mechanisms does digital intelligence affect the sustained green innovation level of EV battery enterprises?
- Which combinations of technical, organizational, and environmental conditions constitute sufficient pathways for high sustained green innovation in EV battery enterprises?
By addressing these questions, my research contributes threefold. First, I construct an econometric model that separates the direct effect of digital intelligence from its indirect effects through scientific and technological human capital. Second, I examine how government subsidies and the regional business environment moderate the digital intelligence–green innovation nexus. Third, I move beyond the conventional single-factor perspective and employ fsQCA to identify equifinal pathways, thereby offering more actionable guidance to managers and policy-makers in the EV battery sector.
2. Literature Review and Theoretical Foundations
2.1 Digital Intelligence and Sustained Green Innovation
In the digital economy era, digital intelligence is not merely a technology toolkit but a strategic capability that fosters operational efficiency and innovation. Scholars such as Qi Yudong and his colleagues describe digital intelligence as the process of using digital intelligent technologies to empower production and management, restructuring business models, and achieving collaborative evolution between technology and commerce. My own view aligns with the broader perspective that digital intelligence comprises data acquisition, real-time monitoring, predictive analytics, and smart decision-making. These capabilities allow an EV battery enterprise to optimize energy consumption, reduce material waste, and accelerate the R&D cycle of battery chemistry, thus contributing to sustained green innovation.
Green innovation itself originates from the concept of technologies or products that reduce pollution while creating economic value. Sustained green innovation, which I define following Xiang Gang and later authors such as Yang Lisheng, refers to the long-term process in which an enterprise continuously introduces multiple green innovation initiatives over a period of at least five to ten years, obtains economic returns from those initiatives, and improves environmental quality. Unlike one-off green patent applications, sustained green innovation emphasizes continuity, path dependence, and intertemporal resource allocation. In the EV battery context, this means continuous improvement in battery chemistry, production processes, and recycling technologies.
Several mechanisms connect digital intelligence to sustained green innovation in EV battery enterprises. First, the deep integration of digital intelligent technologies with domain-specific battery technologies produces knowledge spillovers, accelerates iterative upgrades, and stimulates a virtuous cycle of green innovation. Second, industrial Internet platforms and smart supply chains enable EV battery enterprises to coordinate dispersed upstream and downstream partners, thereby lowering transaction costs and information asymmetries. Third, digital intelligence fosters cross-boundary knowledge sharing, and it helps enterprises constantly evaluate environmental performance through real-time data. I thus propose my first hypothesis:
H1: Digital intelligence significantly improves the sustained green innovation level of EV battery enterprises.
2.2 Scientific and Technological Human Capital as a Mediator
Human capital theory, first articulated by Adam Smith and then formalized by Theodore Schultz and Gary Becker, emphasizes that investments in education, skills, and knowledge enhance individual productivity and organizational innovation. In EV battery enterprises, scientific and technological human capital embodies the engineers, chemists, data scientists, and process specialists who conduct R&D and transform knowledge into green products and processes.
Digital intelligence can increase both the demand for and the productivity of scientific and technological human capital. As intelligent devices replace repetitive manual tasks, enterprises must retrain and hire highly skilled personnel who can analyze complex data streams and implement intelligent production systems. In other words, digital intelligence strengthens the technological human capital base of EV battery enterprises. A strengthening technological human capital base then feeds directly into sustained green innovation because scientists and engineers accumulate knowledge, interact with external research partners, and explore new battery materials and recycling technologies. I hypothesize the following mediating effect:
H2: Digital intelligence enriches the scientific and technological human capital of EV battery enterprises, thereby promoting their sustained green innovation.
2.3 The Moderating Roles of Government Subsidies and Business Environment
Government intervention theory provides a framework for understanding how fiscal incentives influence innovation. Historically, subsidies significantly accelerated the development of China’s new energy vehicle industry. Recently, however, scholars have observed diminishing returns: as the EV battery sector matures, firms become less dependent on subsidies. I nevertheless believe that subsidies still shape innovation behavior by reducing R&D costs and enhancing the returns to human capital investment. I therefore set up two related hypotheses. First, the moderating hypothesis:
H3a: Government subsidies moderate the relationship between digital intelligence and sustained green innovation in EV battery enterprises.
H3b: Government subsidies positively moderate the relationship between digital intelligence and scientific and technological human capital in EV battery enterprises.
Beyond subsidies, the regional business environment—comprising macroeconomic conditions, market environments, infrastructure, and policy contexts—affects innovation through external governance, legal protection, and administrative efficiency. A favorable business environment facilitates entrepreneurial activity, reduces transaction costs, and protects intellectual property, thereby amplifying the contributions of digital intelligence. Hence I propose:
H4: The business environment positively moderates the relationship between digital intelligence and sustained green innovation in EV battery enterprises.
2.4 Heterogeneity in Enterprise and Regional Characteristics
I also anticipate that firm-specific and region-specific characteristics moderate the effects of digital intelligence on sustained green innovation. Ownership type, growth potential, and geographic location jointly shape access to resources, organizational agility, and strategic motivation. For instance, private enterprises often respond more nimbly to technological opportunities; high-growth enterprises enjoy superior capital abundance; and firms in central and western China may experience stronger marginal benefits from the late-mover advantage of digitalization. Thus I formulate a generalized hypothesis:
H5: The effects of digital intelligence on sustained green innovation in EV battery enterprises exhibit heterogeneity across ownership, growth ability, and geographical region.
3. Research Design and Models
3.1 Baseline Econometric Model
To test the above hypotheses, I define OIP as the sustained green innovation of an EV battery enterprise, which I measure using the continuity index of green patent applications OIN across adjacent periods, following He Yubing’s practice:
$$OIP_{t} = OIN_{t} \times \left[ \frac{1}{2} \times \frac{OIN_{t} – OIN_{t-1}}{OIN_{t-1}} + \frac{1}{2} \times \frac{OIN_{t} – OIN_{t-1}}{OIN_{t}} \right]$$
Where Dcg denotes the level of digital intelligence, measured by counting the frequencies of digital intelligence-related keywords in annual reports. I construct the following direct-effect model, where Controls indicates the vector of control variables:
$$OIP_{i,t} = \alpha_{0} + \alpha_{1} Dcg_{i,t} + \alpha_{2} Controls_{i,t} + Year + Industry + \varepsilon_{i,t}$$
For the mediation analysis of scientific and technological human capital (Hum), I apply the causal-steps approach of Wen Zhonglin and colleagues, estimating the following equations:
$$Hum_{i,t} = \beta_{0} + \beta_{1} Dcg_{i,t} + \beta_{2} Controls_{i,t} + Year + Industry + \varepsilon_{i,t}$$
$$OIP_{i,t} = \delta_{0} + \delta_{1} Dcg_{i,t} + \delta_{2} Hum_{i,t} + \delta_{3} Controls_{i,t} + Year + Industry + \varepsilon_{i,t}$$
For the moderating effect of government subsidies (Gov), I estimate the following models:
$$OIP_{i,t} = \gamma_{0} + \gamma_{1} Dcg_{i,t} + \gamma_{2} Gov_{i,t} + \gamma_{3}(Dcg_{i,t} \times Gov_{i,t}) + \gamma_{4} Controls_{i,t} + Year + Industry + \varepsilon_{i,t}$$
$$Hum_{i,t} = \eta_{0} + \eta_{1} Dcg_{i,t} + \eta_{2} Gov_{i,t} + \eta_{3}(Dcg_{i,t} \times Gov_{i,t}) + \eta_{4} Controls_{i,t} + Year + Industry + \varepsilon_{i,t}$$
For the moderating effect of the business environment (Be), I estimate the model below:
$$OIP_{i,t} = \theta_{0} + \theta_{1} Dcg_{i,t} + \theta_{2} Be_{i,t} + \theta_{3}(Dcg_{i,t} \times Be_{i,t}) + \theta_{4} Controls_{i,t} + Year + Industry + \varepsilon_{i,t}$$
3.2 Configurational Models Based on TOE
While regression analysis examines net effects, the actual innovation trajectory of EV battery enterprises may be determined by interdependent conditions. To account for the complexity of these influencing factors, I invoke the Technology-Organization-Environment framework, which parses the factors affecting innovation adoption into three dimensions. From the technology dimension, I choose the level of digital intelligence application; from the organization dimension, I select digital intelligent resource input and scientific and technological human capital; from the environment dimension, I select the digital intelligent policy environment and digital development environment. I then apply fsQCA to identify configuration paths.
4. Empirical Analysis: Variables and Data
4.1 Variables and Measurement
Table 1 summarizes the exact measurement methods of all variables used in my econometric analysis. I calculated sustained green innovation through the green patent application transition formula described above. For digital intelligence, I constructed a research-oriented keyword dictionary encompassing artificial intelligence, big data, cloud computing, blockchain, and digital applications. Using Python-based text mining of listed EV battery enterprises’ annual reports, I aggregated the word frequencies and normalized them with the natural logarithm of frequency plus one.
| Variable Type | Variable Name | Code | Measurement |
|---|---|---|---|
| Dependent variable | Sustained green innovation level | OIP | Continuity index of green patent applications |
| Core explanatory variable | Digital intelligence level | Dcg | Annual report keyword frequency; ln(1 + total frequency) |
| Mediating variable | Scientific and technological human capital | Hum | Employees with bachelor’s degrees or above / total employees |
| Moderating variable | Government subsidies | Gov | ln(amount of government R&D subsidies collected manually from financial statements) |
| Moderating variable | Business environment | Be | Composite score of provincial business environment using principal component analysis |
| Control variable | Firm size | Size | ln(total assets) |
| Control variable | Leverage ratio | Lev | Total liabilities / total assets |
| Control variable | Return on assets | Roa | Net profit / average total assets |
| Control variable | Board size | Board | ln(number of directors) |
| Control variable | Firm growth | Growth | Operating revenue growth rate |
| Control variable | Duality | Dual | Takes 1 when chairman and CEO are the same person |
| Control variable | Independent director ratio | Indep | Independent directors / total directors |
| Control variable | Top shareholder concentration | Top1 | Shares held by largest shareholder / total shares |
| Control variable | Regional GDP | Grp | ln(provincial GDP) |
4.2 Sample Selection and Data Sources
My sample consists of listed EV battery enterprises from the Chinese A-share market covering the period from 2012 to 2021. I deliberately selected this window because it captures both digital transformation acceleration and strict environmental regulation strategies in China. First, I excluded firms classified as ST, *ST, PT, or firms with financial anomalies and missing key data. Thereafter I obtained an unbalanced panel of 72 EV battery enterprises with 629 firm-year observations. Financial and human capital data derive from the CSMAR and Wind databases, whereas green patent data come from the Chinese Research Data Services Platform. Digital intelligence variables are extracted via Python crawler technology from annual reports. Business environment data come from the China Statistical Yearbook and provincial statistical yearbooks. To avoid the identification of individual firms, I do not report the complete list of enterprises.
4.3 Descriptive Statistics
Table 2 provides descriptive statistics for all variables. I noticed that the average value of OIP was 1.972 with a standard deviation of 1.734 and a maximum of 6.952. This reveals that the sustained green innovation level of EV battery enterprises is unevenly distributed, yet an overall average is relatively low. The mean of digital intelligence is 3.110, with a range from 0 to 6.873 and a standard deviation of 1.156, indicating that some EV battery enterprises have already undertaken sophisticated digitalization while others remain in the early stage. The mean of Hum is 0.178, which means only 17.8 percent of employees are scientific and technical talents with a bachelor’s degree or above. This is a relatively low ratio. The substantial variation in government subsidies—from 0 to 21.75—also suggests that policy support is highly asymmetric.
| Variable | Observations | Mean | Std. Dev. | Min | Max |
|---|---|---|---|---|---|
| OIP | 629 | 1.972 | 1.734 | 0.000 | 6.952 |
| Dcg | 629 | 3.110 | 1.154 | 0.000 | 6.873 |
| Hum | 629 | 0.178 | 0.101 | 0.000 | 0.568 |
| Be | 629 | 0.350 | 0.380 | -0.594 | 1.574 |
| Gov | 629 | 13.040 | 6.560 | 0.000 | 21.750 |
| Size | 629 | 22.360 | 1.274 | 19.870 | 26.450 |
| Lev | 629 | 0.434 | 0.185 | 0.047 | 0.989 |
| Roa | 629 | 0.034 | 0.062 | -0.613 | 0.249 |
| Board | 629 | 2.073 | 0.380 | -0.331 | 2.708 |
| Growth | 629 | 0.290 | 0.798 | -0.864 | 15.610 |
| Indep | 629 | 36.630 | 4.789 | 33.330 | 60.000 |
| Dual | 629 | 0.347 | 0.476 | 0.000 | 1.000 |
| Top1 | 629 | 26.710 | 13.870 | 3.003 | 75.250 |
| Grp | 629 | 9.584 | 0.406 | 8.859 | 10.780 |
To inspect potential multicollinearity, I obtained Pearson correlation coefficients for the principal variables. Digital intelligence displays a positive correlation with sustained green innovation of 0.305, while Hum and Be also correlate positively at 0.138 and 0.135 respectively. Importantly, none of the correlation coefficients exceed 0.8, implying no severe multicollinearity problem. Table 3 contains the relevant correlation coefficients.
| Variable | OIP | Dcg | Hum | Be | Gov | Size | Lev |
|---|---|---|---|---|---|---|---|
| OIP | 1.000 | ||||||
| Dcg | 0.305*** | 1.000 | |||||
| Hum | 0.138*** | 0.324*** | 1.000 | ||||
| Be | 0.135*** | 0.225*** | 0.120*** | 1.000 | |||
| Gov | -0.067* | -0.052 | 0.081** | -0.157*** | 1.000 | ||
| Size | 0.542*** | 0.087** | -0.178*** | 0.088** | -0.031 | 1.000 | |
| Lev | 0.283*** | 0.020 | -0.275*** | -0.104*** | -0.146*** | 0.574*** | 1.000 |
Note: *** p<0.01, ** p<0.05, * p<0.1. Similar significance levels are used in subsequent tables.
5. Empirical Findings on Impact Mechanism
5.1 Direct Effect of Digital Intelligence
Table 4 presents the baseline OLS regression results. In column (1), without any controls, digital intelligence positively and significantly affects sustained green innovation. In column (2), after adding all control variables, the coefficient on Dcg remains positive and statistically significant. Column (3) further adds year and industry fixed effects to mitigate omitted-variable bias. The coefficient on Dcg is 0.2186, which is significant at the 1% level. In terms of economic significance, a one-unit increase in digital intelligence is associated with an approximately 0.22 unit improvement in the sustained green innovation index. This evidence strongly supports H1: digital intelligence improves the sustained green innovation of EV battery enterprises.
| Variable | (1) OIP | (2) OIP | (3) OIP |
|---|---|---|---|
| Dcg | 0.4585*** (9.0020) |
0.3675*** (8.7515) |
0.2186*** (3.7782) |
| Size | 0.7312*** (12.4912) |
0.8039*** (13.4610) |
|
| Lev | 0.2734 (0.6005) |
-0.2092 (-0.4856) |
|
| Roa | 2.3594*** (2.9508) |
2.1894*** (2.7688) |
|
| Board | -0.3785** (-2.2659) |
-0.3262* (-1.8442) |
|
| Growth | -0.1321*** (-2.7658) |
-0.1275*** (-2.8944) |
|
| Indep | 0.0223* (1.8539) |
0.0243* (1.7982) |
|
| Dual | 0.3447*** (2.8600) |
0.3277** (2.4239) |
|
| Top1 | -0.0241*** (-6.4783) |
-0.0226*** (-5.2947) |
|
| Grp | 0.2581** (2.0713) |
0.1609 (0.9227) |
|
| Year / Industry fixed effects | Yes | Yes | Yes |
| N | 629 | 629 | 629 |
| R-squared | 0.093 | 0.424 | 0.495 |
| F-statistic | 81.0360 | 53.4294 | 25.2978 |
5.2 Mediating Effect of Scientific and Technological Human Capital
Table 5 presents the mediation regression results. The first column shows a significantly positive coefficient of digital intelligence on Hum, confirming that EV battery enterprises that actively pursue digital intelligence tend to accumulate more highly educated technical employees. Column (2) introduces Hum into the baseline model. Although the Dcg coefficient decreases relative to the baseline coefficient but remains significant, Hum itself exerts a positive and significant effect on OIP. This mixed evidence points to partial mediation: digital intelligence nurtures scientific and technological human capital, and that enriched human capital subsequently accelerates sustained green innovation. I therefore accept H2.
| Variable | (1) Hum | (2) OIP |
|---|---|---|
| Dcg | 0.0185*** (4.4606) |
0.1666*** (2.8407) |
| Hum | 2.8053*** (4.1473) |
|
| Size | 0.0019 (0.5685) |
0.7986*** (13.4459) |
| Lev | -0.1511*** (-5.0403) |
0.2145 (0.4852) |
| Roa | -0.0272 (-0.4072) |
2.2657*** (2.9889) |
| Board | -0.0065 (-0.6901) |
-0.3080* (-1.7337) |
| Growth | -0.0064** (-2.2688) |
-0.1096** (-2.5361) |
| Indep | 0.0028*** (3.0229) |
0.0166 (1.2316) |
| Dual | -0.0241*** (-2.7003) |
0.3953*** (2.9960) |
| Top1 | -0.0012*** (-4.3100) |
-0.0191*** (-4.3783) |
| Grp | -0.0122 (-0.9360) |
0.1953 (1.1434) |
| Year / Industry FE | Yes | Yes |
| N | 629 | 629 |
| R-squared | 0.316 | 0.514 |
To further verify the indirect pathway, I employed the bias-corrected percentile Bootstrap method with 1,000 replications. Table 6 reports the bootstrap result. The indirect effect equals 0.0741, with a 95% confidence interval from 0.0416 to 0.1175. Since zero is excluded from this interval, I have confidence in the statistical significance of the mediating pathway.
| Effect | Coefficient | Bootstrap Std. Err. | Z | P-value | 95% Confidence Interval |
|---|---|---|---|---|---|
| Indirect effect | 0.0741 | 0.0194 | 3.82 | 0.000 | [0.0416, 0.1175] |
| Direct effect | 0.2934 | 0.0467 | 6.29 | 0.000 | [0.1947, 0.3820] |
5.3 Moderating Effects
Table 7 reports the moderating effects of government subsidies and the business environment. Column (1) examines the interaction term, Dcg×Gov, but it is not statistically significant. This result indicates that the current level of government subsidies no longer serves as an effective moderator between digital intelligence and sustained green innovation in EV battery enterprises. Interestingly, in column (2), the interaction term between digital intelligence and government subsidies has a positive coefficient of 0.0010, which is significant at the 5% level when the dependent variable is Hum. In other words, subsidies continue to amplify the effect of digital intelligence on human capital accumulation, but the overall moderation effect on sustained green innovation is weak. H3a is not supported in my sample, whereas H3b is supported. In column (3), the interaction term of Dcg×Be is positive and statistically significant at the 1% level, confirming that regional business environments facilitate the translation of digital intelligence into sustained green innovation. H4 is therefore supported.
| Variable | (1) OIP | (2) Hum | (3) OIP |
|---|---|---|---|
| Dcg | 0.3503** (2.5801) |
0.0037 (0.5421) |
0.1448** (2.3767) |
| Gov | 0.0267 (0.8499) |
-0.0015 (-1.0945) |
|
| Dcg×Gov | -0.0094 (-1.0865) |
0.0010** (2.2795) |
|
| Be | -1.5827*** (-3.4986) |
||
| Dcg×Be | 0.3059*** (2.6106) |
||
| Controls | Yes | Yes | Yes |
| Year / Industry FE | Yes | Yes | Yes |
| N | 629 | 629 | 629 |
| R-squared | 0.497 | 0.329 | 0.504 |
5.4 Heterogeneity Analysis
To investigate heterogeneous responses, I re-estimated the baseline model for distinct subgroups. Table 8 subdivides the sample by ownership type. I observe that Dcg has a stronger effect in non-state-owned EV battery enterprises, with a coefficient of 0.3343, than in state-owned enterprises with a coefficient of 0.2569. In the Chinese institutional context, non-state enterprises frequently have greater autonomy to make timely investments in digital infrastructure and to attract top technical talent, which subsequently allows them to achieve higher sustained green innovation levels.
| Variable | (1) State-Owned OIP | (2) Non-State-Owned OIP |
|---|---|---|
| Dcg | 0.2569** (2.2249) |
0.3343*** (4.1877) |
| Controls | Yes | Yes |
| N | 143 | 486 |
| R-squared | 0.630 | 0.519 |
Table 9 divides the sample into high-growth and low-growth EV battery enterprises based on the median of revenue growth. In high-growth enterprises, the coefficient on digital intelligence equals 0.2692, significant at the 1% level, whereas in low-growth enterprises, the coefficient is positive but insignificant. High-growth EV battery enterprises possess flexible funding channels and stronger incentives to exploit digital intelligence for innovation breakthroughs.
| Variable | (1) High-Growth OIP | (2) Low-Growth OIP |
|---|---|---|
| Dcg | 0.2692*** (2.8497) |
0.1190 (1.4971) |
| Controls | Yes | Yes |
| N | 315 | 314 |
| R-squared | 0.528 | 0.518 |
Finally, Table 10 splits enterprises according to geographical location. Digital intelligence does not significantly affect the sustained green innovation level of enterprises in the eastern region, which may reflect diminishing marginal returns in developed areas. By contrast, the coefficient on Dcg for central and western China is positive and significant. Interestingly, EV battery enterprises in central and western regions can utilize the latecomer advantage and digital intelligence to leapfrog, assisted by lower land and labor costs and the proactive policies of local governments.
| Variable | (1) Eastern Region OIP | (2) Central and Western Regions OIP |
|---|---|---|
| Dcg | 0.1379 (1.3952) |
0.2306*** (2.9781) |
| Controls | Yes | Yes |
| N | 428 | 201 |
| R-squared | 0.557 | 0.495 |
5.5 Endogeneity and Robustness Checks
To address possible reverse causality and measurement bias, I conducted several robustness tests. First, I employed the lagged digital intelligence terms as instrumental variables, because past digitalization levels affect current levels but are predetermined with respect to current green innovation shocks. Column (1) in Table 11 presents the first-stage result, exhibiting a strong positive relationship. The Cragg-Donald Wald F statistic of 302.30 confirms that weak instrumental variables are not a concern. The second-stage result in column (2) still shows a positive and significant Dcg coefficient.
Second, I replaced the core explanatory variable with a comprehensive sub-category indicator based on artificial intelligence, blockchain, cloud computing, and big data keywords. Third, I substituted an OLS model with a Tobit model to account for data censoring. Fourth, I excluded samples headquartered directly in municipalities and provincial capitals to reduce regional-specific determinants. The outcomes, displayed in columns (3) through (5), remain qualitatively unchanged. Table 11 therefore provides consistent evidence for the main conclusions.
| Variable | (1) Dcg First-Stage | (2) OIP Second-Stage | (3) OIP Alternative Dcg | (4) OIP Tobit | (5) OIP Excluding Special Cities |
|---|---|---|---|---|---|
| Dcg | 0.328*** (3.50) |
0.1217** (2.1297) |
0.2796*** (3.0359) |
0.1729* (1.8359) |
|
| L.Dcg | 0.664*** (14.42) |
||||
| L2.Dcg | 0.116*** (2.59) |
||||
| Controls | Yes | Yes | Yes | Yes | Yes |
| N | 464 | 464 | 629 | 629 | 342 |
6. Configurational Pathway Analysis with fsQCA
6.1 Data and Calibration
Having verified the net effect of digital intelligence, my attention turned to understanding how multiple conditions combine to generate high sustained green innovation. For this purpose, I assembled a cross-sectional sample of 63 EV battery enterprises from the A-share market for the year 2021. Based on the TOE framework, I selected five antecedent conditions:
- Digital intelligence application (Dcg), representing the technology dimension.
- Digital intelligent input (Rd) and scientific and technological human capital (Hum), representing the organizational dimension.
- The digital intelligent policy environment (Gov) and the digital development environment (Dig), representing the environmental dimension.
In line with fsQCA practice, I calibrated raw data into fuzzy-set membership scores between 0 and 1. I set the 95th percentile as the full membership anchor, the 50th percentile as the crossover point, and the 25th percentile as the full non-membership anchor. Because fsQCA software cannot exactly process the 0.5 value, I adjusted membership values at the crossover anchor to 0.501, an innocuous operation. Table 12 displays the resulting calibration anchors.
| Variable | Code | Full Membership | Crossover | Full Non-Membership |
|---|---|---|---|---|
| Sustained green innovation | OIP | 5.805 | 2.780 | 0.000 |
| Digital intelligence application | Dcg | 5.132 | 3.714 | 2.565 |
| Digital intelligent input | Rd | 0.067 | 0.030 | 0.008 |
| Scientific and technological talent | Hum | 0.334 | 0.188 | 0.090 |
| Digital intelligent policy environment | Gov | 18.308 | 4.360 | 0.000 |
| Digital development environment | Dig | 0.707 | 0.685 | 0.487 |
6.2 Necessity Analysis
Before evaluating sufficient configurations, I performed a necessity analysis to test whether any single condition is indispensable for the outcome. As shown in Table 13, all consistency levels are below the conventional threshold of 0.9. I therefore conclude that no single condition alone is necessary for either high or non-high sustained green innovation. This result substantiates my premise that EV battery enterprises rely on the interplay among technical, organizational, and environmental factors rather than on any unique silver bullet.
| Condition | High OIP Consistency | High OIP Coverage | Low OIP Consistency | Low OIP Coverage |
|---|---|---|---|---|
| Dcg | 0.6916 | 0.6808 | 0.6163 | 0.6047 |
| ~Dcg | 0.5984 | 0.6101 | 0.6747 | 0.6856 |
| Rd | 0.6381 | 0.7128 | 0.5574 | 0.6207 |
| ~Rd | 0.6605 | 0.5995 | 0.7421 | 0.6715 |
| Hum | 0.6095 | 0.6556 | 0.5985 | 0.6417 |
| ~Hum | 0.6669 | 0.6249 | 0.6788 | 0.6341 |
| Dig | 0.6486 | 0.6383 | 0.6724 | 0.6596 |
| ~Dig | 0.6541 | 0.6670 | 0.6312 | 0.6417 |
| Gov | 0.7411 | 0.6862 | 0.5956 | 0.5497 |
| ~Gov | 0.5138 | 0.5604 | 0.6601 | 0.7176 |
6.3 Sufficiency Analysis and Configurational Solutions
I then constructed a truth table, set the frequency threshold at 1, and set the consistency cutoff at 0.85. The intermediate and parsimonious solutions allow me to classify core and peripheral conditions. In Table 14, I present the sufficient configurations for high and non-high sustained green innovation. Black circles “●” indicate the presence of a condition; crossed-out circles “⊗” indicate its absence. Large-sized circles indicate core conditions, and small-sized circles indicate peripheral conditions. Blank cells mean that the condition is irrelevant for that particular configuration.
| Condition | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|
| Outcome | High OIP | Low OIP | ||||
| Dcg | ⊗ | ● | ⊗ | ● | ⊗ | ⊗ |
| Rd | ⊗ | ● | ⊗ | |||
| Hum | ● | ⊗ | ● | ● | ⊗ | |
| Dig | ● | ● | ⊗ | ● | ● | |
| Gov | ● | ● | ● | ● | ⊗ | ⊗ |
| Consistency | 0.879 | 0.867 | 0.887 | 0.861 | 0.859 | 0.849 |
| Raw coverage | 0.356 | 0.342 | 0.309 | 0.250 | 0.358 | 0.349 |
| Unique coverage | 0.047 | 0.097 | 0.000 | 0.042 | 0.053 | 0.044 |
| Solution coverage | 0.522 | 0.402 | ||||
| Solution consistency | 0.849 | 0.857 | ||||
I identified two broad pathways that lead to sustained green innovation in EV battery enterprises: the digital talent-led pathway and the digital application-led pathway.
(1) Digital Talent-Led Configuration. Configuration 1 consists of high Hum and high Gov combined with absent Dcg and absent Rd. This configuration explains roughly 35.6% of the high-outcome cases. Configuration 3 similarly combines high Hum, high Dig, and high Gov, while digital intelligence application is weak. These configurations imply that when an EV battery enterprise cannot quickly deploy abundant digital applications, it can nevertheless achieve sustained green innovation by emphasizing scientific and technological human capital and obtaining policy support. The complementary role of talent and policy is therefore evident.
(2) Digital Application-Led Configuration. Configuration 2 comprises high Dcg, high Dig, and high Gov while Hum is absent, explaining 34.2% of high-outcome cases. Configuration 4 contains high Dcg, high Hum, high Rd, and high Gov even when digital development environment is absent. These results indicate that the role of organizational resources is context-dependent: in Environments 2 and 4, direct digital technology applications substitute for some missing conditions. These configurational pathways corroborate the regression conclusion that digital intelligence is an important engine for sustained green innovation in EV battery enterprises but also demonstrate that digital intelligence operates jointly with complementary conditions.
With respect to the non-high outcome, Configurations 5 and 6 show that enterprises located in an advantaged digital development environment still fail to achieve high sustained green innovation when they lack digital inputs, subsidy support, or talent. In Configuration 5, high Dig alone accompanied by the absence of Dcg and Rd and the absence of Gov causes low OIP. In Configuration 6, high Dig combined with absent Hum and absent Gov likewise causes low OIP. I thus infer that a favorable external digital condition cannot deliver its benefits without sufficient internal organizational investments and policy support. Only when EV battery enterprises and governments work together can resources flow toward continuous innovation.
7. Conclusions and Policy Implications
7.1 Main Conclusions
Supported by the econometric and configurational evidence, my research arrives at several major conclusions. First, digital intelligence exerts a genuine, statistically significant positive influence on sustained green innovation in EV battery enterprises. This finding remains robust after endogeneity treatment, replacement of the core variable, alternative model specification, and sample adjustment.
Second, scientific and technological human capital partially mediates the relationship between digital intelligence and sustained green innovation. Enterprises that utilize digital intelligence successfully tend to create more high-skilled jobs and strengthen knowledge-based capabilities. This, in turn, improves the continuity of green innovation. Digital intelligence is not merely a physical capital investment; it is a lever for intangible human capital empowerment.
Third, government subsidies do not significantly moderate the entire pathway from digital intelligence to sustained green innovation. Although subsidies still positively moderate the accumulation of technological human capital under digitalization, their overall effect has weakened since Chinese EV battery enterprises have entered a stage of more autonomous, market-driven innovation. The business environment, however, strengthens the digital intelligence effect. A stable, transparent, and efficient institutional environment allows EV battery enterprises to realize the full innovative returns of digital intelligence.
Fourth, heterogeneities abound. Private EV battery enterprises benefit more from digital intelligence than state-owned enterprises. High-growth enterprises gain more than low-growth ones. Digital intelligence is more effective for EV battery enterprises in central and western China than in eastern China. These findings have implications for the geographical optimization of the EV battery industrial chain.
Fifth, the fsQCA results reveal that the pathways to high sustained green innovation are multiple but require configuration. Talent-policy combinations and digital application-environment combinations both lead to high outcomes. Non-high outcomes occur when enterprises have uncertain environments but lack internal digital inputs and government funding. Thus, no silver-bullet policy exists; rather, a holistic configuration of technical, organizational, and environmental capacities delivers sustained green innovation in EV battery enterprises.
7.2 Policy and Managerial Recommendations
My research generates several actionable recommendations for managers and policy-makers seeking to enhance sustained green innovation among EV battery enterprises. First, EV battery enterprises should consolidate a digital intelligence empowerment mechanism, covering smart manufacturing, industrial internet, data analytics, and virtual simulation. In my view, this mechanism is now a prerequisite for advancing green innovation in the EV battery sector. Second, enterprises should formulate a human-capital development strategy to train and attract engineering and data science talent. This can mitigate a key bottleneck in the technology–organization–environment configuration identified in the fsQCA results. Third, policy-makers should re-examine government subsidies in the current phase of industry maturity. Although subsidies have weakened in their moderating capacity, they may maintain a threshold effect under international competition. Future policy research should simulate subsidy thresholds to guarantee the global competitiveness of EV battery enterprises. Fourth, local governments should continuously optimize the business environment by promoting administrative efficiency, strengthening legal protection for intellectual property, and providing transparent information services. My results show that the business environment positively moderates digital intelligence effectiveness, and I thus recommend a coordinated policy portfolio that combines digital infrastructure construction with business environment improvement. Fifth, a national-level digital innovation ecosystem should be established for EV battery enterprises. The ecosystem should advance the flow and value-added use of data elements, enabling enterprises in different regions to share knowledge and production resources. In particular, given the stronger digital intelligence effect in central and western China, an ecosystem approach can support industrial relocation, reduce regional disparities, and improve the resilience of the entire EV battery supply chain. Last but not least, policy design should consider the interdependent and equifinal pathways identified by fsQCA. Because multiple configurations produce the same result, tailored policy mixes should be created for different categories of EV battery enterprises rather than applying a one-size-fits-all set of incentives.
7.3 Limitations and Future Research Directions
Although this research has delivered meaningful results, I acknowledge several limitations. First, the measurement of digital intelligence and sustained green innovation relies on indicators derived from annual reports and patent applications. These indicators may carry some degree of subjectivity and may not capture unobserved forms of green innovation such as process-level eco-innovation. Second, I selected a limited number of moderating variables and configurational conditions. Future research should expand the analytical framework by considering other relevant conditions such as consumer green demand, corporate ESG performance, and the digital capabilities of upstream suppliers. Third, my empirical sample focuses on A-share listed EV battery enterprises in China. Future studies may extend the model to unlisted small and medium-sized EV battery enterprises or to cross-country samples. Fourth, longitudinal configurational analysis may be used to observe how configurations evolve across different stages of the EV battery industry life cycle. Given the rapid development of digital and intelligent technologies, my overriding hope is that future research continues to investigate how EV battery enterprises can reconcile competitiveness with environmental stewardship through digital intelligence.