How Does Digital Innovation Empower the Development of New Quality Productive Forces? An Empirical Study Based on Double Machine Learning

Abstract

New quality productive forces (NQPFs) are a key driver for sustainable and high-quality development, where digital innovation (DI) plays a crucial role in promoting the evolution of NQPFs. Based on this, this paper takes 2740 A-share listed companies from 2011 to 2022 as research samples and utilizes double machine learning to explore the impact and transmission mechanisms of DI on NQPFs. The study finds that DI significantly empowers the development of NQPF; mechanism-wise, DI achieves this through industry–university–research cooperation (IURC), increasing market concentration (MC) and enhancing government innovation subsidies (GISs); heterogeneity analysis reveals that the empowering effect of DI on NQPFs is stronger in large cities, small cities, the region northwest of the Hu Line, and the old industrial bases, whereas in megacity behemoths, megacities, regions along the Hu Line and the southeast region, and non-old industrial base enterprises, the effects are relatively smaller. This study provides both theoretical and empirical insights into how DI drives the development of NQPFs and supports sustainable economic growth, offering valuable guidance for future development strategies.

1. Introduction

In September 2023, General Secretary Xi Jinping introduced the concept of “new quality productive forces” during his research visit to Heilongjiang. Unlike traditional productivity, NQPFs represent a qualitative transformation in productivity driven by scientific and technological innovation, aligning with the principles of the new development concept and sustainable, high-quality growth [1]. As a comprehensive concept, NQPFs facilitate an overall improvement in production efficiency and effectiveness by innovating factors and enhancing quality. The essence of this concept lies in the profound leap in production activity capacity, which not only transcends mere technological innovation but also represents the modern realization of the qualitative change and leap within the theory of productive forces in a Marxist political economy [2]. Compared with eco-innovation, which primarily focuses on the internalization of environmental costs, or digital innovation, which emphasizes the use of technological tools, the core of new quality productive forces resides in the systemic and qualitative transformation of productive forces. This transformation is distinctly reflected in two critical aspects: “new” and “quality”. “New” refers to the composition of new elements and the new economic phenomena they generate, including new laborers, new production materials, and new labor objects, as well as their novel combinations; “quality” underscores new quality and new forms, including the high efficiency and low consumption in the production process, and the high quality and sustainability of the final production outcomes [3]. In this context, according to Schumpeter’s theory of innovation, technological innovation can enhance productive forces through “creative destruction”, thereby precipitating a profound transformation in the structural composition of productive forces [4]. Therefore, as a key manifestation of scientific and technological innovation in the digital era, DI may emerge as a crucial factor in establishing new production relationships and enhancing the development of NQPFs.

However, existing literature primarily explores the relationships between various factors, such as scientific and technological innovation, digital transformation, digital-real integration, and institutional mechanism reform, and the development of NQPF. There is limited research specifically addressing the impact of DI on the development of NQPF and the mechanisms involved. NQPFs are an advanced form of productivity, driven by technological innovation, and are generated through breakthroughs in technology, the innovative configuration of production factors, and deep industrial transformation [5]. NQPFs represent the new development concept, with a particular focus on green development. Leveraging breakthroughs in green technologies and the low-carbon transformation of industrial structures, they drive the greening of both production and lifestyle, contributing to the balanced and sustainable development of the economy and environment [6]. Aligned with this, DI, through the integration of big data, cloud computing, artificial intelligence, the Internet of Things, and other cutting-edge technologies, drives revolutionary technological advancements [7]. Moreover, DI is not only the core driving force for digital transformation but also a key factor in supporting the green, low-carbon transition and promoting sustainable development [8]. As a result, DI not only establishes the technical foundation for NQPFs but also facilitates the restructuring of industrial ecology through the spillover effects of green technologies, promoting the formation of more sustainable productivity models. Despite the various potential advantages of DI, can it effectively accelerate the development of NQPFs? What is the relationship between the two, and how do their mechanisms function? Does the relationship differ across diverse regional or industrial contexts? An in-depth investigation of these mechanisms not only bridges the gaps in existing research on the empowering effects of DI but also provides a theoretical foundation for businesses to strategically optimize their DI initiatives and for policymakers to develop precise interventions, which holds considerable practical significance in promoting the rapid cultivation of NQPFs.

In view of this, this paper takes A-share listed companies as the research object, comprehensively evaluating their NQPF levels from 2011 to 2022 based on both substantial and permeable elements to examine the role of DI in empowering the development of NQPF and its transmission mechanism. The possible contributions of this paper are as follows: First, in the research theme, this paper links DI with NQPFs, delving into how DI enables the enhancement of NQPFs, thus enriching research on the driving effects of DI. Second, in terms of mediating mechanisms, this paper introduces three key variables, namely industry–university–research cooperation, market concentration, and government innovation subsidies, to clarify their mediating roles between DI and NQPFs and investigates how DI empowers NQPFs. Finally, in terms of research methodology, this paper leverages the unique advantages of double machine learning methods in handling high-dimensional data and non-parametric prediction, which effectively avoids the “curse of dimensionality” and model setting bias, thereby enhancing the reliability of research results.

2. Literature Review and Research Hypothesis

2.1. Literature Review

2.1.1. Research on Digital Innovation

With the rapid progress of digitalization, DI has emerged as a core driving force, enabling companies to enhance competitiveness and achieve sustainable growth [9,10]. Against this backdrop, DI has become a prominent research focus, with studies primarily exploring its connotation, evaluation methods, driving factors, and impact effects.

First of all, DI is defined as the integration of digital technologies, such as big data and cloud computing, into the innovation process to generate new products, production processes, organizational models, and business models [11]. On this basis, DI not only optimizes production processes and management models but also facilitates the continuous iteration and upgrading of products and services [12]. Furthermore, by guiding companies to transition from traditional industrialized production to data-driven intelligent production modes, DI induces profound transformations in organizational structures and operational models [13]. As digital technologies continue to evolve, Gupta et al. (2019) argue that DI is increasingly characterized by openness, diversification, and decentralization, with a clear trend toward platformization. This evolution has shifted innovation activities beyond a single innovation entity, progressively transforming them into a digital innovation ecosystem collaboratively constructed by multiple stakeholders [14].

Secondly, the assessment methods of DI are mainly divided into two categories: quantitative analyses based on patents and qualitative reconstructions based on innovation capability. In terms of quantitative analyses, Huang et al. (2023) conducted keyword-based text analyses to count the number of digital patents within companies’ patent portfolios and used this as a metric for measuring DI [15]. Tao et al. (2023) identified digital technology-related patents based on International Patent Classification (IPC) information and calculated the number of digital patents from the perspective of technological fields [16]. Zhang et al. (2023) used the number of patent applications of the digital economy industry to evaluate the level of DI [17]. In terms of qualitative reconstructions, Liao et al. (2023) redefined the connotation of “DI capability” using the grounded theory approach and developed a measurement scale for DI capability by combining qualitative and quantitative methods [18].

Thirdly, the drivers of DI can be categorized into two main groups: internal and external factors. In terms of internal factors, Lokuge et al. (2019) identified flexible human capital as a key determinant of companies’ adaptability to change and their ability to implement innovations [19]. Nasiri et al. (2023) argued that the primary drivers of DI include human, collaborative, technological, and innovative capabilities [20]. Dahms et al. (2023) emphasized that organizational agility and digital capability are critical drivers of innovation performance [21]. In terms of external factors, Liu et al. (2021) revealed that companies with market positions at either end of the spectrum are more inclined toward DI [22]. Jun et al. (2021) found that SMEs in Pakistan accelerate DI by enhancing digital platform applications and strengthening organizational resilience [23]. Li et al. (2023) argued that companies’ DI behaviors are significantly shaped by peer influence, particularly in social environments characterized by strong networks and advanced fintech [24].

Finally, the effects of DI can be divided into two main categories: advancing traditional company performance and promoting high-quality company performance underpinned by green and low-carbon development. With respect to traditional company performance, Lyytinen et al. (2016) highlighted that DI facilitates the reconfiguration of innovation networks, enhances product innovation capabilities, and fosters cross-organizational knowledge integration and collaborative innovation [25]. Yu et al. (2017) argued that the embedding of digital capabilities significantly accelerates the innovation process of products and services, enhances synergistic cooperation among heterogeneous innovation entities, and redefines how value is created for company products and services [26]. Liu et al. (2021) found that DI capabilities significantly enhance company competitive advantages and enable companies to achieve sustainable growth during the dual transformation process [27]. In terms of high-quality company performance, cryptocurrency-based digital innovations drive sustainable investments by enhancing eco-friendly portfolios and exhibiting dynamic patterns across diverse market environments [28].

2.1.2. Research on New Quality Productive Forces

The high-quality development phase of China emphasizes the need to cultivate NQPFs, which are essential for driving sustainability and reconciling economic advancement with environmental preservation [29,30]. Against this backdrop, the academic discourse has increasingly recognized the importance of understanding NQPFs. Consequently, existing studies often focus on their theoretical basis, definition, driving factors, and impact effects.

The NQPF is an advanced productive force characterized by a leap in total factor productivity [30]. The essence of NQPFs is the systemic restructuring of the laborer–labor material–labor object combination, which is driven by the synergistic effects of digital technologies and institutional innovation [31]. This restructuring disrupts the traditional linear growth of productivity and shows the theoretical distinction from concepts such as digital innovation and eco-innovation. Digital innovation is primarily concerned with the application of technologies like big data and cloud computing to optimize existing production processes [14]. Eco-innovation, however, focuses on reducing the negative environmental impacts of production and service processes through technological and managerial improvements, enhancing efficiency in areas like energy use and waste disposal [32]. In the context of a circular economy, the “dematerialization of goods and services” aims to reduce material resource consumption by promoting product–service systems and the sharing economy, fostering resource recycling. In contrast, NQPF’s revolutionary nature manifests in three key dimensions: new technological, new energy, and the digital economy. In the technological dimension, disruptive innovations drive industrial upgrades and transform economic development models; in the energy dimension, NQPF promotes a transformation of the energy structure and the establishment of new energy systems to ensure green and sustainable development; in the economic dimension, it centers on the digital economy, facilitating the integration of digital technologies with the real economy, which sparks new sources of economic growth [33]. Unlike partial optimization, the NQPF stands out in its emphasis on the holistic integration of new technologies, new energy, and the digital economy, promoting a more comprehensive transformation. The theoretical innovation of NQPFs lies in converting Schumpeter’s notion of “creative destruction” into an orderly process of transformation, amplifying the effects of factor allocation through the creation of a unified national market [34], thereby providing institutional solutions for latecomer countries to overcome the “middle-income trap”.

The existing literature on the drivers of NQPFs mainly focuses on three aspects: scientific and technological innovation, data elements, and government–market collaboration. Xu et al. (2023) pointed out that scientific and technological innovation is the central impetus behind NQPFs, marking a clear distinction from traditional productivity models [35]. Zhang et al. (2024) argued that generative artificial intelligence (AIGC) technology, as a significant form of scientific and technological innovation, empowers the development of NQPFs through strategic decision-making innovation, production and manufacturing innovation, marketing service innovation, and organizational structure innovation [36]. Hong et al. (2024) found that data elements, leveraging their powerful permeability and integration, interact with traditional production factors to achieve disruptive breakthroughs in technological innovation driven by algorithms and computational power, thereby accelerating the formation and development of NQPF [37]. Zhou et al. (2023) believe that under government–market collaboration, the government provides policy support guarantees, while the market stimulates vitality through innovation-driven efforts. Together, these efforts improve the efficiency of scientific and technological transformation as well as resource integration, ultimately enhancing the NQPF [6].

In the context of the accelerated transformation of the global economy toward digitalization and intelligence, NQPFs are regarded as a key factor influencing high-quality development, economic structure optimization, and the construction of a modern industrial system. Jia et al. (2024) pointed out that the development of NQPFs can fully leverage the comparative advantages of various regions, facilitate the flow and agglomeration of production factors, enhance innovation-driven effects, and achieve high-quality development [38]. Zhang et al. (2024) argued that by advancing the process of digital-real integration, NQPFs can promote the deep integration of technological innovation with traditional industries, support the transformation and upgrading of the economic structure, and shift the development model from quantitative growth to qualitative improvement [39]. Hong (2024) argued that as an innovation-led form of advanced productive forces, NQPFs can contribute to the construction of China’s modernized industrial system by fostering technological innovation and upgrading production factors [33].

2.1.3. Research on the Relationship Between DI and NQPFs

In the context of the digital age, DI has become a fundamental force in promoting high-quality development [40]. It not only optimizes production processes but also provides vital technological support for the creation of NQFPs through the promotion of productivity [41]. However, although previous research has separately examined the connotations and mechanisms of DI and NQPFs, the relationship between the two remains ambiguous and warrants more thorough investigation. This section will analyze the relationship from three perspectives: micro, meso, and macro, focusing on how DI influences the formation of NQPFs through various mechanisms.

At the micro level, DI facilitates the qualitative transformation of productivity by optimizing the allocation of production factors. The transformative application of digital technologies, particularly big data, cloud computing, and artificial intelligence, is capable of markedly enhancing firms’ total factor productivity [15]. Sun et al. (2023) argue that DI fosters productivity growth by enhancing factor allocation efficiency and accelerating the accumulation of human capital [42]. In addition, Zhang and Ma (2024) discovered that with the widespread adoption of generative artificial intelligence (AIGC), firms can digitize and automate processes such as data perception, intelligent recognition, dynamic decision-making, and precise execution, thereby fully leveraging the empowering effects of AIGC technology in generating and developing NQPFs [36]. These studies indicate that DI enhances the production efficiency of micro-level entities, thereby providing the foundation for the formation of NQPFs.

At the meso level, DI serves as a driving force for NQPF development by fostering industrial collaboration and cultivating innovation ecosystems. Li et al. (2024) argue that DI constructs innovation ecosystems rooted in digital technologies and data-driven resources, where organizations and key actors collaborate and interact dynamically within innovation-oriented environments, thus offering critical support for NQPF growth [43]. Moreover, Luo and Xiao (2024) illustrate that the agglomeration dynamics of core digital economy sectors substantially reinforce regional NQPF development through technological breakthroughs, production factor reallocation, and industrial structural evolution [44]. Wu et al. (2024) further elucidate that the spatial diffusion effects of the digital economy dissolve interregional barriers, fostering cross-regional innovation networks that significantly accelerate regional NQPF advancement [45]. Collectively, these studies highlight the indispensable role of DI in strengthening industrial coordination and regional innovation ecosystems, both of which are fundamental to the formation of NQPFs.

At the macro level, the widespread adoption of DI hinges on a sound institutional environment and robust support for digital governance. Wei et al. (2024) found through their research on the reform of big data institutions that the deregulation of institutions can effectively unlock the potential of data elements, promoting DI application and creating a conducive institutional environment for NQPF formation [46]. Jiao and Qi (2024) noted that the digital economy, by optimizing the widespread proliferation and application of digital technologies, enhances societal operational mechanisms and accelerates the formation of NQPFs [47]. Zhai and Xia (2024) further pointed out that at the macro level, the digital economy contributes to NQPF development by optimizing knowledge production and diffusion, fostering the integration of industrial and innovation chains, promoting high-quality national economic growth, and shaping the new dual-circulation development strategy [48].

To summarize, in the analysis of micro, meso, and macro pathways, DI optimizes factor allocation, promotes industrial collaboration, and accelerates institutional deregulation, thereby providing technological, ecological, and policy support for the formation of NQPFs. However, current research faces several limitations: First, while the effects of the digital economy and data elements on productivity improvement are widely recognized, the specific role of DI in advancing NQPFs remains underexplored, and the relationship between the two has not been fully elucidated. Second, there is insufficient detail regarding the mechanism analysis, as the promotion of NQPFs by DI has not been sufficiently explored from the market and government perspectives. Third, the application of traditional empirical methods is limited, as existing econometric models struggle with high-dimensional data and complex causal mechanisms, making it difficult to precisely explain the influence of DI on NQPF development. In contrast, double machine learning methods present distinct advantages in handling such data and non-parametric prediction, effectively addressing the limitations of traditional models. Therefore, this paper employs double machine learning methods to explore the effects and mechanisms of DI on the development of NQPFs, aiming to unravel the “mechanism black box” between DI and NQPFs.

2.2. Research Hypotheses

DI drives the cultivation and development of NQPFs. On the one hand, DI leverages advanced digital technologies in product design, process management, organizational structures, and business models to optimize production processes, enhance management efficiency, and reduce operating costs [9]. It also strengthens companies’ ability to withstand uncertainties [49], thereby boosting innovation capacity, increasing labor productivity, and injecting vitality into the sustained growth of NQPFs. On the other hand, DI fosters industrial integration through the deep embedding of digital technology, breaking traditional industrial boundaries, building cross-industry ecosystems, facilitating inter-industry collaborative innovation, and providing favorable conditions for the formation and development of NQPFs [26]. Therefore, this paper proposes the following hypotheses:

H1: 

DI can drive the rapid development of NQPFs.

DI establishes a digital sharing platform for companies, universities, and research institutions, breaking traditional collaboration barriers and enabling all parties to participate in project cooperation across different fields more flexibly and efficiently, thereby expanding the breadth of Industry–University–Research cooperation (IURC) [50]. Meanwhile, the introduction of virtual experiments and digital twin technology allows collaboration parties to complete experiments and validations in a virtual environment, accelerates the technology development process, and reduces experimental uncertainty, encouraging active participation and further deepening the depth of cooperation [51]. DI drives the deepening of IURC and creates favorable conditions for the development of NQPF. IURC enables companies to acquire the latest cutting-edge technologies and R&D outcomes more rapidly and apply these results flexibly to innovations in production processes and manufacturing technologies, thereby enhancing NQPFs. More importantly, the continuous promotion of IURC transforms R&D activities from one-time tasks into a sustainable development system, ensuring that companies maintain their technological advantages over the long term, thereby promoting the sustained improvement of NQPFs. Therefore, the following hypothesis is proposed:

H2: 

DI empowers the development of NQPFs by promoting IURC.

DI strengthens the market position of leading companies by enhancing technological advantages and network effects, significantly raising market entry barriers and thereby increasing market concentration (MC) [52]. At the same time, the non-competitive nature of DI allows it to significantly reduce marginal costs through economies of scale, enabling successful companies to rapidly expand their market share, creating a winner-takes-all scenario and further promoting a high MC [53]. The increase in MC not only reshapes the competitive landscape but also exerts a significant impact on production efficiency. As Schumpeter noted in his 1943 study, there is a strong link between the rise in MC and production efficiency. According to the “relative market power hypothesis”, increased MC grants companies greater pricing power, enhances resource allocation efficiency through economies of scale, and generates positive spillover effects on NQPFs [54]. Therefore, the following hypothesis is proposed:

H3: 

DI empowers the development of NQPFs by increasing MC.

By engaging in DI, companies can provide more transparent and complete operational data and innovation outcomes, enabling the government to accurately assess their innovation capabilities, R&D performance, and potential economic contributions, thereby improving their chances of obtaining government innovation subsidies (GISs) [55]. DI facilitates companies’ access to GISs, providing critical support for the sustained development of NQPFs. Through financial support, innovation subsidy policies encourage companies to increase their investment in digital technology R&D, which not only accelerates the development of digital technologies but also enhances companies’ competitive advantages and growth potential in the digital era, thereby expediting the formation of NQPFs [2]. Therefore, the following hypothesis is proposed:

H4: 

DI promotes the development of NQPFs by facilitating the government to increase innovation subsidies.

3. Research Design

3.1. Sample and Data Collection

This paper selects Chinese A-share listed companies from 2011 to 2022 as the initial sample for the study. To enhance data comprehensiveness and result accuracy, data from multiple sources are utilized. The data mainly fall into the following four categories: (1) basic information, financial data, and company governance structure information, obtained from the China Stock Market & Accounting Research Database (CSMAR); (2) textual data from the annual reports of the listed companies; (3) textual data from the patents of the listed companies, sourced from the State Intellectual Property Office of China (SIPO); and (4) data on industrial robots, retrieved from the International Federation of Robotics (IFR).

Subsequently, the initial sample data are processed as follows: (1) Samples from ST and *ST companies are excluded. ST denotes “Special Treatment”, referring to companies that are under intensified regulatory scrutiny due to financial or other irregularities; *ST signifies companies that have suffered consecutive losses for more than three years and may face the imposition of suspension from listing or delisting warnings; (2) samples from companies in the financial and real estate industries are removed; (3) samples with missing values are excluded. To mitigate the impact of outliers, continuous variables are subjected to a 1% winsorization. The final sample consists of 2740 listed companies with a total of 13,269 observations.

3.2. Measures

3.2.1. New Quality Productive Forces (NQPFs)

This study is grounded in the conceptualization and structural composition of NQPFs, building upon the research frameworks proposed by Han et al. (2024), Zhang et al. (2024), and Xiao et al. (2024) [56,57,58]. This study employs the TOPSIS–entropy method to establish a comprehensive evaluation model, facilitating a systematic measurement of NQPFs from the two perspectives of tangible and penetrative elements (see

Table 1. The indicator system for measuring new quality productivity forces.

Variable	Factor	Sub-Factor	Indicator	Measurement Method
Tangible Elements	New-Quality Laborers	Employee Quality	High-Quality Employees	Proportion of employees with postgraduate education and above
			R&D Personnel Proportion	Proportion of R&D personnel to total employees
		Managerial Quality	New-Quality Laborers	Management teams with a digital background
			CEO Functional Experience Diversity	Count of CEO functional experiences
	New-Quality Labor Objects	Ecological Environment	Environmental Performance	The environmental score from the Huazheng ESG rating system
		Future Development	Fixed Asset Ratio	Fixed assets/Total assets
			Robot Penetration Rate	Company-level robot penetration rate
	New-Quality Labor Materials	Technological Labor Materials	Company Innovation Level	Ln(Number of patents applied for by the company + 1)
		Green Labor Materials	Number of Green Invention Patents	Ln(Number of green invention patents applied for in the year + 1)
			Number of Green Utility Model Patents	Ln(Number of green utility model patents applied for in the year + 1)
		Digital Labor Materials	Intelligent Level	Ln(Frequency of intelligent-related terms + 1)
			Digital Asset Ratio	Digital-related assets/Total intangible assets
Penetrative Elements	Digital Environment	Digital–Intelligent Integration	Digital Transformation Level	Ln(Frequency of digitalization-related terms + 1)
			Digital-Real Industry Integration Level	Company-level integration of digital and real industry technologies
		Data Elements	Enterprise Data Elements	Ln(Frequency of data asset-related terms + 1)
	New Technology R&D	Indirect Costs	Depreciation and Amortization Ratio in R&D Expenses	Depreciation and amortization in R&D expenses/revenue
			R&D Leasing Costs Ratio	Leasing costs in R&D expenses/revenue
		Direct Costs	Direct R&D Input Ratio	Direct input in R&D expenses/revenue

). The subsequent section delineates the indicators for each constituent element, in accordance with the classification of NQPF dimensions.

Firstly, tangible elements represent the material foundation of NQPFs, encompassing three critical dimensions: new-quality laborers, new-quality labor objects, and new-quality labor materials. The indicators are selected according to the following rationale:

• New-Quality Laborers.

• The defining features of new-quality laborers include their ability to innovate and their advanced skill set, allowing them to adapt effectively to new technologies and tools [56]. Drawing on the framework by Zhang et al. (2024), this study constructs indicators based on two levels: employee quality and managerial quality [57].
  • Employee Quality is evaluated using the “proportion of R&D personnel” and the “proportion of individuals with postgraduate degrees or higher”. The first reflects the intensity of investment in innovative human capital, while the second illustrates the extent of upgrade in the knowledge structure of laborers. Although traditional studies categorize labor quality by educational attainment or years of education, new-quality laborers emphasize the capacity to innovate and adapt to the application of new technologies. Therefore, this study focuses on the structural proportion of R&D and highly educated personnel, reflecting an innovation-driven approach.
  • Managerial Quality is measured through two indicators: “the digital background of the management team” and “CEO functional experience diversity”. The former, following Wang et al. (2023) [59], constructs dummy variables based on the presence of keywords related to information, intelligence, and software in executives’ professional backgrounds. The latter, following the functional experience categorization by Duan et al. (2023) [60], counts the number of cross-functional roles held by the CEO. These indicators are critical in evaluating whether the management team has the strategic leadership required for the successful implementation of NQPFs, particularly in the domains of digital transformation and diversified innovation.

2.

New-Quality Labor Objects.

• New-quality labor objects are characterized by the integration of new energy, new materials, and intelligent equipment in the production process [56]. The selection of such elements must consider both ecological efficiency and the potential for future development. This study combines the approaches by Xiao et al. (2024) [58] and Zhang et al. (2024) [57] to measure the following elements.
  • Ecological Environment: The “HuaZheng ESG Environmental Score” is employed to assess enterprise environmental performance. This indicator offers a comprehensive reflection of an enterprise’s actions in areas such as pollution control, resource utilization, and environmental sustainability, which is consistent with the green transformation requirements of NQPFs.
  • Future Development: Indicators such as the “Proportion of Fixed Assets” and the “Robot Penetration Rate at the Enterprise Level” are utilized. The former measures an enterprise’s investment in long-term production capacity through the ratio of fixed assets to total assets [57]. The latter follows the methodology by Wang and Dong (2023) [61] as well as Acemoglu and Restrepo (2020) [62], utilizing IFR robot stock data disaggregated to the enterprise level. The robot penetration rate directly measures the degree of automation and intelligence in production, acting as a pivotal marker of the advancement of new-quality labor objects.

3.

New-Quality Labor Materials.

• New-quality labor materials include technological, green, and digital labor tools. The selection of indicators must reflect three key characteristics: technological advancements, sustainability, and digitalization.
  • Technological Labor Materials: This dimension is quantified by the “number of patents filed by the enterprise” (in natural logarithms). Patents serve as a direct reflection of the enterprise’s technological research and development output and signify its investment in technological innovation, which contributes to the advancement of NQPFs through technological progress.
  • Green Labor Materials: The “number of green invention patents” and the “number of green utility model patents” (both in natural logarithms) are employed as key indicators of green innovation. Green patents serve as a direct indicator of the enterprise’s efforts in low-carbon technology, supporting NQPF’s objective of achieving a green transformation in production processes.
  • Digital Labor Materials: These are measured through “intelligence level” and the “proportion of digital assets”. The former, based on the work by Yue and Gu (2023) [63], assesses the intelligence level of enterprises by extracting and analyzing the frequency of keywords related to intelligent transformation and intelligent technologies in their annual reports. The latter draws from Zhang et al. (2024) [64], categorizing intangible assets such as “software”, “network”, “client”, “management systems”, and “intelligent platforms” as “digital assets”, measured by their ratio to total intangible assets. The intelligence level reflects the degree of digitalization in labor materials, while the proportion of digital assets indicates the enterprise’s investment in and allocation of digital resources. These two indicators collectively represent the penetration and application of digital technologies in labor materials, laying the foundation for assessing innovation in labor materials and improvements in productivity.

Secondly, penetrative elements enhance productivity by optimizing the efficiency of the combination of tangible elements. This study primarily focuses on two key components: the digital environment and new technology R&D:

4.

Digital Environment.

• The digital environment serves as an empowering foundation for NQPF, with its measurement involving the flow of digital–intelligent integration and the data elements.
  • Digital–Intelligent Integration: The “digital transformation level” and the “digital-physical industry integration level” are employed as key indicators of digital–intelligent integration. The first follows the methodology by Wu et al. (2021) [65], utilizing word frequency analysis to scrape keywords related to “digital transformation” (such as “cloud computing” and “blockchain”) from enterprise annual reports to quantify the intensity of public disclosure on the enterprise’s digital transformation efforts. The second method, based on Huang and Gao (2023) [66], identifies patents in non-digital industries that cite digital industry patents, measuring the level of integration by aggregating the number of integration events. These two indicators provide a comprehensive view of the extent to which enterprises penetrate and integrate digital technologies within traditional industries, shedding light on the direct impact of digital technologies on productivity improvement.
  • Data Elements: Following the approach of Yuan et al. (2022) [67], this study counts the frequency of “data asset” keywords (such as “big data” and “data mining”) in enterprise annual reports to measure the level of enterprise data elements. Enterprises that frequently disclose data asset information tend to prioritize the collection and application of data elements, and this indicator provides an effective measure of the enterprise’s data-driven capabilities in the formation of NQPFs.

5.

New Technology R&D.

• New technology R&D represents the core driving force behind penetrative elements. The design of its indicators must encompass both the intensity of R&D investment and the structure of associated costs.
  • Direct Costs: This is quantified by the ratio of direct investments in R&D to operating income. This indicator reflects the concentration of financial resources on core technological advancements, providing insights into whether enterprises prioritize resource allocation towards breakthrough technologies, a critical factor for advancing NQPF development.
  • Indirect Costs: This is measured through the “proportion of R&D depreciation and amortization” and the “proportion of R&D leasing costs”, which calculate the ratio of depreciation and leasing expenses in R&D to operating income. The allocation of indirect costs, particularly in fixed assets and external services, highlights the enterprise’s strategic deployment of innovation resources, signaling its long-term commitment and planning for innovation in driving the development of NQPFs.

3.2.2. Digital Innovation (DI)

Following the studies of Wang et al. (2023) and Liu et al. (2023), the number of digital-related patents is adopted as the measure of DI [68,69]. Digital patents refer to patents involving digital economy technologies and owned by companies. To identify digital patents, this study conducts a textual analysis of the abstracts, specifications, and claims of all invention and utility model patent applications of listed companies, focusing on “digital” keywords to determine whether a patent qualifies as a digital patent. The “digital” keywords are selected with reference to official documents such as the “2020 Digital Transformation Trend Report”, the “Special Action Program for Digital Empowerment of SMEs”, and the “White Paper on Enterprise Digital Transformation (2021 Edition)”. Building on this, the criteria for determining DI are set following the research by Ni et al. (2023). Specifically, if a company obtains a digital patent for the first time in a given year, that year is considered the starting point of its DI, and the DI variable is assigned a value of 1 from that year onward, as defined by the following formula [70]:

$$D{I}_{\mathrm{i},t}=\left\{\begin{array}{l}0,t<t_0,\\ 1,t\ge t_0.\end{array}\right.$$

(1)

3.2.3. Controls

This study, in order to better account for potential confounding factors, incorporates several firm-level control variables (Controls), as identified in the research by Huang Xianhai et al. (2023) and Ni Xuanming et al. (2023) [70,71]. These variables are classified into four categories: firm characteristics, financial condition, operational efficiency, and market performance. Specifically, (1) firm characteristics include firm size (Size), the number of Employees (Employee), and the shareholding ratio of the top 10 shareholders (Top10); (2) the financial condition encompasses the leverage ratio (Lev), equity multiplier (EM), the debt-to-equity ratio (DER), the debt to long-term capital ratio (DLCR), return on assets (ROAs), gross profit margin (GrossProfit), the current ratio (Liquid), the comprehensive tax rate (CTR), capital intensity (CAP), the rate of capital accumulation (RCA), and operating leverage (OL); (3)operational efficiency is captured by the accounts receivable ratio (REC), the proportion of fixed assets (FIXED), the proportion of intangible assets (Intangible), the proportion of tangible assets (Tangible), and the revenue growth rate (Growth); market performance is measured by earnings per share (EPS), annual return on equity (ROE), the price-to-book ratio (PB), Tobin’s Q ratio (TobinQ), the book-to-market ratio (BM), the and investment-to-cash ratio (Invest). Furthermore, to mitigate potential biases arising from unobserved firm-specific attributes and time-related effects, the study controls for both firm-specific fixed effects and time fixed effects. A comprehensive list of the control variables is presented in

Table 2. Control variables’ definitions.

Variable	Definition	Variable	Definition
Size	Natural logarithm of total assets	OL	Fixed costs/(operating revenue—variable costs)
Employee	Natural logarithm of the number of employees	REC	Accounts receivable/current assets
Top10	Total shares held by the top ten shareholders divided by total shares outstanding	FIXED	Net fixed assets/total assets
Lev	Total liabilities/total assets	Intangible	Intangible assets/total assets
EM	Total assets/shareholders’ equity	Tangible	Tangible assets/total assets
DER	Total liabilities/shareholders’ equity	Growth	(Current period operating revenue/previous period operating revenue)—1
DLCR	Non-current liabilities/long-term capital	EPS	Net profit divided by the number of common shares outstanding
ROAs	Net profit/average total assets
GrossProfit	Gross profit/operating revenue	ROE	Net profit divided by the average shareholders’ equity
Liquid	Current assets/current liabilities	PB	Market value of stocks/net assets
CTR	Income tax expense/total profit	TobinQ	Market value of company/replacement cost
CAP	Net fixed assets/operating revenue	BM	Book value/market value
RCA	Shareholders’ equity at year-end/shareholders’ equity at beginning of year	Invest	Expenditures on the construction of fixed assets, intangible assets, and other long-lived assets of an enterprise

Control variables are unitless ratios, except for firm size (natural log of year-end total assets, no unit).

.

3.3. Estimation Methods

To investigate the mechanism through which DI empowers the development of NQPFs, this paper employs a double machine learning (DML) model for evaluation. Unlike traditional models, the double machine learning model proposed by Chernozhukov et al. (2018) leverages machine learning algorithms to automatically control key variables and eliminate human bias, significantly enhancing estimation accuracy and robustness [72]. When analyzing how DI empowers the development of NQPFs, the double machine learning model accurately captures the actual effects of DI, controls for potential confounding factors, and ensures the validity and reliability of causal inference. In addition, the double machine learning model is capable of addressing the complexity of multi-dimensional features, making it a powerful tool for analyzing the mechanism through which DI empowers the development of NQPFs.

The model is based on the framework of partial linear regression:

$$Y={\theta }_{0}DI+g(X)+U ,\hspace{1em}E(U|X,DI)=0$$

(2)

where Equation (1) represents the main equation; $Y$ is the dependent variable, representing new quality productive forces; $DI$ is the treatment variable, indicating whether digital innovations are implemented; ${\theta }_0$ is the treatment effect coefficient; $X$ is the set of control variables, which are covariates affecting the dependent variable through the function $g\left(X\right)$; and $U$ is the error term, with a conditional mean of zero.

To ensure unbiased and valid estimates, bias correction was performed using orthogonalization. First, the residual term was obtained by regressing the treatment variable $DI$ in the main equation:

$$V=DI-m(X) ,\hspace{1em}E(V|X)=0$$

(3)

Subsequently, the parameter ${\theta }_0$ is estimated using the residual term $V$ as the instrumental variable.

$${\stackrel{⌣}{\theta }}_0={\left({\displaystyle \frac{1}{n}}\sum _{i\in I}\widehat{V_{\mathrm{i}}}D{I}_i\right)}^{-1}{\displaystyle \frac{1}{n}}\sum _{i\in I}\widehat{V_{\mathrm{i}}}\left[Y_i-\widehat{g}(X_i)\right]$$

(4)

This estimation process involves splitting the samples into subsets through cross-validation methods. To avoid overfitting, the auxiliary samples are first used to estimate $m\left(X\right)$ and $g\left(X\right)$, and the main samples are subsequently used to estimate ${\theta }_0$.

In addition, to further analyze the robustness of the impact of DI, this paper constructs a more general interaction model to provide comprehensive causal inference results:

$$Y=g(DI,X)+U,\hspace{1em}E(U|X,DI)=0$$

(5)

$$V=DI-m\left(X\right),\hspace{1em}E\left(V|X\right)=0$$

(6)

4. Results

4.1. Descriptive Statistics

This paper provides descriptive statistical analysis for all variables, and the results are detailed in

Table 3. Descriptive statistics.

Variable	Mean	Std. Dev.	Min	Max
DI	0.523	0.500	0.000	1.000
NQPF	0.102	0.089	0.004	0.462
Size	22.375	1.342	18.370	28.194
Lev	0.432	0.204	0.008	0.998
EM	2.325	5.632	0.357	417.253
DER	1.325	5.632	0.008	416.253
DLCR	0.152	0.170	−0.196	0.996
ROA	0.040	0.072	−1.130	1.285
GrossProfit	0.296	0.184	−2.978	0.975
Liquid	2.460	3.176	0.028	144.000
REC	0.128	0.107	0.000	0.813
FIXED	0.201	0.158	0.000	0.971
Intangible	0.049	0.067	0.000	0.926
Tangible	0.906	0.104	0.148	1.000
Growth	7.182	964.095	−2.734	135,000.000
CTR	0.032	0.062	−0.351	4.316
CAP	2.741	4.247	0.088	289.885
RCA	0.095	19.113	−2580.000	624.375
OL	1.397	6.161	−302.760	243.914
EPS	0.404	1.094	−7.671	49.930
ROE	0.115	0.560	−0.821	15.211
Invest	0.063	0.463	0.000	60.969
Top10	57.236	15.274	8.265	101.160
PB	3.649	6.731	0.117	354.175
Employee	7.746	1.261	2.398	13.254
TobinQ	2.073	1.623	0.641	56.664
BM	0.624	0.257	0.018	1.559

. The mean of DI is 0.523, which reflects a relatively even distribution of DI within the sample. The mean of NQPF is 0.102, reflecting that most firms exhibit low levels of NQPF. The standard deviation of 0.089 suggests that the NQPF values are tightly clustered around the mean, showing a degree of consistency among firms. Although the maximum value is 0.462, which signifies the superior performance of a few companies in NQPFs, the overall performance remains relatively low in this dimension.

Regarding the mean, standard deviation, and distribution of the extreme values of the control variables, the majority of variables show relatively balanced distributions, indicating minimal differences between firms in these dimensions, with financial and operational performance remaining relatively stable. Additionally, the standard deviation of Top10 (proportion of shares held by the top 10 shareholders) is 15.274, suggesting substantial variation in shareholder concentration, while the mean of ROAs (Return on Assets) is 0.040 with a small standard deviation, implying that most firms have low profitability with little fluctuation. Although variables such as Top10 present some data fluctuations, the adaptive capacity of the double machine learning model enables it to effectively process these fluctuations and extract meaningful relationships. Consequently, the overall sample is still suitable for application in this type of analysis.

4.2. Baseline Regression Analysis

This paper employs a double machine learning (DML) model to estimate the enabling effect of DI on NQPFs. The double machine learning method is based on the “cross-fitting algorithm”, which performs training and validation across multiple samples to reduce biases in high-dimensional data and improve the predictive accuracy of the model. Chernozhukov et al. (2018) [72] provided empirical evidence through iterative rotations of auxiliary and primary samples; partitioning the samples into five groups leads to a more robust performance than partitioning them into only two groups. Accordingly, this study implements a 1:4 sample split ratio, utilizing 80% of the samples for training and reserving the remaining 20% for validation. Within the double machine learning framework, commonly used regression algorithms include the random forest (RF), gradient boosting (GB), and the support vector machine (SVM). The RF is an ensemble learning technique that aggregates multiple decision trees for prediction, automatically selecting the most relevant features and demonstrating strong adaptability to high-dimensional data. GB is an iterative method ideal for regression problems, enhancing model performance by progressively reducing errors. The SVM is applicable to both classification and regression tasks, identifying the optimal hyperplane in high-dimensional data. Considering the complexity of data features and variables, this study adopts the RF as the baseline algorithm to process high-dimensional data, mitigate overfitting, and improve stability and accuracy. To further test the robustness of the model, GB and the SVM are also applied for verification. The results from the baseline model are shown in

Table 4. Baseline regression results.

	(1)	(2)	(3)	(4)	(5)	(6)
DI	0.030 ***	0.030 ***	0.022 ***	0.022 ***	0.035 ***	0.035 ***
	(18.899)	(19.055)	(13.731)	(13.973)	(21.558)	(21.774)
Linear terms of the control variables	YES	YES	YES	YES	YES	YES
Quadratic terms of the control variables	NO	YES	NO	YES	NO	YES
Time fixed effect	YES	YES	YES	YES	YES	YES
Individual fixed effect	YES	YES	NO	NO	NO	NO
Industry fixed effect	NO	NO	YES	YES	NO	NO
Province fixed effect	NO	NO	NO	NO	YES	YES

Robust standard errors are shown in parentheses; different asterisks indicate significance: *** (1%), ** (5%), and * (10%). The entries from linear terms to province fixed effects denote dichotomous assessments of whether corresponding control variables or fixed effects are incorporated into the model.

.

To investigate the enabling effect of DI on NQPFs, six distinct models are employed for testing.

Model (1) functions as the major regression model, predicated on the hypothesis that DI significantly contributes to the development of NQPFs. To ensure the precision of the regression estimates, the model incorporates both firm-specific fixed effects and time fixed effects. The introduction of firm-specific fixed effects is designed to control for the potential confounding effects of firm characteristics on the DI-NQPF relationship, while time fixed effects account for macroeconomic forces that impact all firms, thus mitigating the influence of common external factors on the DI effect. By integrating these two fixed effects, the model ensures an accurate and unbiased estimate of DI’s impact on NQPFs. The regression results demonstrate a significantly positive relationship between DI and NQPFs, supporting the proposed hypothesis.

In order to further assess the impact of DI in different contexts, models (3) and (5) replace the firm-specific fixed effects in model (1) with industry and provincial fixed effects, respectively. This modification allows for an examination of whether the effect of DI on NQPFs is consistent across different industries and regions. By controlling for heterogeneity at the industry and provincial levels, the model extends the applicability of DI’s enabling effect. The results indicate that, whether controlling for individual-level differences, industry-specific characteristics, or inter-provincial economic policies, the enabling effect of DI remains consistently positive, confirming its effectiveness at both industry and provincial levels.

To further examine the robustness of the results, models (2), (4), and (6) introduce squared terms of the control variables to models (1), (3), and (5) to account for potential nonlinear relationships. The findings show that, despite minor changes in the estimated coefficients after the inclusion of quadratic terms, the effect of DI on NQPFs remains significantly positive across all model. This suggests that the enabling effect of DI is stable and not significantly influenced by potential non-linearities.

4.3. Robustness Test

To ensure the robustness of the assessment results, this paper conducts robustness tests by incorporating interaction fixed effects, resetting the machine learning model, and replacing the DI variables.

4.3.1. Incorporation of Interaction Fixed Effects

To confirm the robustness of the foundational results, this study investigates the effects of spatial and industrial heterogeneity on the estimation outcomes. Specifically, recognizing the dynamic evolution of disparities in economic development, policy frameworks, and innovation ecosystems across provinces, which may exert multifaceted influences on NQPFs, we integrate year–province interaction fixed effects into the benchmark model to control for unobserved time-varying factors at the provincial level. The findings in Column (1) of

Table 5. Robustness test results considering interaction fixed effects and varying sample split proportions.

	(1)	(2)	(3)	(4)
	Year–Province	Year–Industry	1:2	1:6
DI	0.030 ***	0.029 ***	0.030 ***	0.029 ***
	(18.942)	(18.425)	(18.955)	(18.512)
Linear terms of the control variables	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES
Individual fixed effect	YES	YES	YES	YES
N	13,269	13,269	13,269	13,269

Robust standard errors are shown in parentheses; different asterisks indicate significance: *** (1%), ** (5%), and * (10%).

reveal that, after controlling for time-varying provincial attributes, the facilitative effect of DI on NQPFs remains significantly positive, underscoring the robustness of the core findings to inter-provincial heterogeneity.

Concurrently, considering the temporal variations in technological attributes, market configurations, and policy orientations across industries, Column (2) of Table 5 incorporates year–industry interaction fixed effects into the benchmark model. The results demonstrate that the coefficient of DI remains significantly positive at the 1% significance level, indicating that the DI-driven enhancement effect persists even after accounting for time-varying industry-specific characteristics. These examinations corroborate the reliability of the benchmark results from spatial and industrial dimensions.

4.3.2. Resetting Machine Learning Models

To prevent potential bias caused by model misspecification in the double machine learning framework from affecting the conclusions, this paper adopts the following measures: First, the sample split ratio is adjusted from the original 1:4 to 1:2 and 1:6, aiming to analyze the potential impact of different sample split ratios on the research findings. Specifically, a 1:2 ratio balances the sample sizes of the training and testing sets, thereby enhancing the model’s generalization capability and mitigating the risk of overfitting. In contrast, a 1:6 ratio increases the weight of the training samples, facilitating the model’s ability to capture subtle features and patterns within the data, albeit potentially facing challenges due to insufficient testing samples. By comparing model performance across different partition ratios, we gain a comprehensive understanding of the impact of sample segmentation on the results. As indicated in Columns (3) and (4) of Table 5, the influence of DI on NQPFs remains significant across varying sample partition ratios, with only minor differences in the effect size. This finding suggests that variations in sample partition ratios do not alter the baseline relationship between DI and NQPFs, thereby confirming the robustness of the research conclusions against sample division methods.

Second, the random forest (RF) algorithm is replaced with the gradient boosting (GB) and support vector machine (SVM) algorithms to evaluate the effects of different algorithms on the research results. The GB algorithm is renowned for its robust predictive capabilities and its proficiency in handling complex data, effectively capturing nonlinear relationships within the data. Meanwhile, SVMs excel in high-dimensional spaces, making them suitable for scenarios with elevated feature dimensions. Results presented in Columns (1) and (2) of

Table 6. Robustness test results for re-modeling and the change threshold.

	(1)	(2)	(3)	(4)
	GB	SVM	Interaction Model	Change Threshold
DI	0.042 ***	0.055 ***	0.040 ***	0.042 ***
	(27.346)	(40.168)	(15.843)	(23.860)
Linear terms of the control variables	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES
Individual fixed effect	YES	YES	YES	YES
N	13,269	13,269	13,269	13,269

Robust standard errors are shown in parentheses; different asterisks indicate significance: *** (1%), ** (5%), and * (10%).

indicate that regardless of the machine learning algorithm employed, the impact of DI on NQPFs remains significant, with coefficients differing only in magnitude. This finding underscores that the research conclusions are not influenced by the choice of specific algorithms, thereby enhancing the credibility and generalizability of the results, while also indicating that the relationship between DI and NQPFs is sufficiently stable to be consistently captured by various machine learning methods.

Finally, a more general interaction model is constructed based on the double machine learning framework to examine the effects of changes in model specifications on the research findings. Compared to the baseline regression analysis, the interaction model represented by Equations (4) and (5) relaxes the assumption of linear relationships among variables, allowing for more complex nonlinear interactions between the digital innovation variable and control variables. This configuration enables a more flexible characterization of the mechanisms through which digital innovation influences new quality productivity, aiding in the capture of potential nonlinear effects. Results in Column (3) of Table 6 indicate that even under a more flexible model specification, digital innovation continues to exert a significant positive impact on new quality productivity, further strengthening the persuasiveness of the research conclusions across different model forms.

4.3.3. Substitution of DI Variables

The DI variable is initially defined as the year in which a company first received a DI patent, serving as the starting point for DI. However, the definition of DI companies includes those that obtained only a single digital patent throughout the entire period, which risks overestimating their level of DI activities and potentially impacting the reliability of the study’s conclusions. To ensure the robustness of the conclusions, this paper adjusts the criterion for determining the starting point of DI to the year in which a firm first obtained at least five digital patents, thus defining the scope of DI firms more rigorously. This adjustment not only enhances the accuracy of identifying digital innovation enterprises but also renders the research findings more representative, thereby more authentically reflecting the actual investments and achievements of enterprises in the realm of digital innovation.

As shown in Column (4) of Table 6, the coefficients remain significant at the 1% level after adjusting the DI determination threshold, further validating the robustness of DI in empowering the development of NQPFs. This indicates that the research conclusions regarding the measurement standards of DI exhibit robustness; even when employing stricter criteria to define DI enterprises, the facilitative effect of DI on NQPFs remains significantly evident.

4.4. Heterogeneity Analysis

Considering that city-level differences may lead to heterogeneity in the impact of DI on NQPFs, this paper analyzes the heterogeneous effects of DI from three perspectives: city class, geographic location, and economic structure. The results are presented in

Table 7. Results of heterogeneity analysis.

	(1)				(2)			(3)
	Megacity Behemoths	Megacities	Large Cities	SMCs	Northwest	Hu Line	Southeast	OIBCs	NOIBCs
DI	0.0339 ***	0.0356 ***	0.0244 ***	0.0217 ***	0.0270 **	0.0363 ***	0.0298 ***	0.0189 ***	0.0305 ***
	(11.550)	(8.317)	(10.489)	(5.892)	(2.499)	(7.205)	(17.597)	(5.493)	(17.683)
Linear term	YES	YES	YES	YES	YES	YES	YES	YES	YES
Time	YES	YES	YES	YES	YES	YES	YES	YES	YES
Individual	YES	YES	YES	YES	YES	YES	YES	YES	YES
N	6670	3094	5745	2031	373	2161	16,122	1993	16,663

Robust standard errors are shown in parentheses; different asterisks indicate significance: *** (1%), ** (5%), and * (10%).

.

4.4.1. Heterogeneity of Urban Hierarchies

Given the size differences across cities and following the study of Jiang et al. (2023), prefecture-level cities are classified into four categories: megacity behemoths, megacities, large cities, and small and medium-sized cities (SMCs), based on the number of permanent residents. Regressions are then performed for cities in each category [73]. Column (1) of Table 7 demonstrates significant city-tier heterogeneity in the enabling effect of digital innovation on new quality productivity: regression coefficients for megacity behemoths and megacities reach 0.0339 and 0.0356, respectively (significant at 1% level), surpassing those of large cities (0.0244) and small and medium-sized cities (SMCs) (0.0217). This hierarchical effect emanates from several mechanisms: First, megacity behemoths and megacities possess more sophisticated digital infrastructure and advanced innovation ecosystems, facilitating DI implementation. Second, these cities leverage their superior talent attraction and economies of scale to catalyze the agglomeration of innovation resources. Finally, the diversity and complexity of market demands in large urban areas expand the application scenarios for DI, thereby amplifying its stimulative effect on NQPFs.

4.4.2. Heterogeneity of the Urban Geographic Location

Considering regional economic differences within China and drawing on the study by Sheng et al. (2020), the country is divided into three major regions: the northwest region, along the Hu Line region, and the southeast region, with the “Hu Line” serving as the dividing boundary [74]. Column (2) of Table 7 unveils significant geographical heterogeneity: DI exhibits stronger effects in the southeast region (coefficient 0.0298) and along the Hu Line region (coefficient 0.0363), both significant at the 1% level, while the coefficient for the northwest region is 0.0270, significant at the 5% level. This spatial variation reflects China’s unbalanced DI development: The southeast and along the Hu Line regions capitalize on their diversified industrial structures, abundant innovation elements, and substantial R&D investments to maximize DI benefits. In contrast, the northwest region confronts constraints from industrial monotony, digital talent scarcity, and limited innovation resources, diminishing its innovation conversion efficiency. These findings underscore the imperative to bridge regional digital divides.

4.4.3. Heterogeneity of the Urban Economic Structure

Considering differences in historical development, resource allocation, and industrial structures among cities, this paper draws on the study by Liu et al. (2023) and adopts the criteria outlined in the “National Adjustment and Reconstruction Plan for Old Industrial Bases (2013–2022)” to categorize the sample into two groups: old industrial base cities (OIBCs) and non-old industrial base cities (NOIBCs), which are analyzed separately [75]. Column (3) of Table 7 reveals that while DI significantly enhances NQPFs in both old industrial bases and non-old industrial bases (at 1% level), non-old industrial bases demonstrate a higher coefficient (0.0305) compared to old industrial bases (0.0189). This structural divergence stems from multiple factors: First, non-old industrial bases exhibit more flexible industrial structures and lower transformation costs, enabling swifter adaptation to digital transformation. Second, their industrial systems demonstrate higher compatibility with digital technologies, facilitating deeper integration between digital innovation and the real economy. Third, old industrial bases, burdened by their high proportion of traditional industries, encounter greater resistance to digital technology adoption, constraining the contribution of DI to NQPF development. These insights emphasize the necessity for city-specific, precision-oriented approaches to DI.

4.5. Mechanism Analysis

From the baseline test results, it is evident that DI significantly enhances NQPFs. However, the enabling pathways warrant further exploration to maximize the positive role of DI in this process. Based on the literature, DI may empower the growth of NQPFs through three primary mechanisms: promoting IURC, increasing MC, and facilitating greater government subsidies for innovation. Drawing on the study by Farbmacher et al. (2022), this paper employs a double machine learning approach to analyze causal mediation effects. By using logistic regression and linear support vector machine (SVM) regression, this study investigates the mechanism pathways of DI in enhancing NQPFs. The results are presented in

Table 8. Causal mediating effects.

Variables	Total Effect	Treatment Group	Control Group	Treatment Group	Control Group
		Direct Effect	Direct Effect	Indirect Effect	Indirect Effect
IURC	0.419 ***	0.042 ***	−0.228 ***	0.377 ***	−0.107 ***
HHI	0.194 ***	−0.111 ***	−0.478 ***	0.305 ***	0.062 ***
GIS	0.051 ***	0.062 ***	−0.174 ***	−0.011 ***	0.247 ***

Robust standard errors are shown in parentheses; different asterisks indicate significance: *** (1%), ** (5%), and * (10%).

[76].

In this study, these mechanisms are operationalized as three analytical dimensions—IURC, MC, and GISs—for which corresponding evaluation methods and metrics are established.

Referring to the study by Liu et al. (2023), the number of joint patent applications is used to assess IURC. A value of one is assigned if a company has cooperative patents in a given year; otherwise, it is assigned zero [77]. Following the methodology proposed by Liu et al. (2011), the Herfindahl–Hirschman Index (HHI) is employed to measure MC [78]. The measurement of government innovation subsidies (GISs) is based on the methodology outlined by Guo (2018), which identifies innovation subsidy projects by searching for project names in government subsidy details using relevant keywords. The total annual innovation subsidy for each company is then calculated by summing these identified projects [79].

4.5.1. Industry–University–Research Cooperation

As shown in Table 8, the total effect of DI in promoting NQPFs through IURC is significantly positive at the 1% level. In the control group, both the direct and indirect effects are significantly negative, while in the treatment group, both direct and indirect effects are significantly positive at the 1% level, indicating that IURC plays a critical role in enhancing NQPFs through DI. Specifically, for company implementing DI, DI not only directly enhances NQPFs but also amplifies this effect through IURC, reflecting the critical role of such cooperation in this process. Conversely, firms that do not implement DI experience negative effects, both directly and indirectly through IURC, suggesting that the absence of a foundation for DI may weaken the effectiveness of such cooperation. This finding further highlights the synergistic relationship between DI and IURC, which jointly drive the enhancement of NQPFs.

This mechanism mainly works through the following channels: First, the information sharing platform built by DI significantly reduces collaboration costs and improves communication efficiency among industry, academia, and research institutions. The real-time interaction and resource sharing supported by digital tools enable the rapid integration and optimal allocation of innovation elements. Second, the synergy between digital technology and R&D capabilities accelerates the transformation of innovative achievements. The digital infrastructure of enterprises provides strong support for IURC and promotes knowledge spillover effects. Third, IURC amplifies the positive effects of digital innovation by integrating multi-party advantageous resources. The complementary advantages of all parties create collaborative innovation forces, improving innovation efficiency and transformation effectiveness.

4.5.2. Market Concentration

To examine whether DI empowers the development of NQPFs by enhancing MC, this paper uses the Herfindahl–Hirschman Index (HHI) to measure the degree of MC.

As shown in Table 8, the total effect of DI on empowering the development of NQPFs by enhancing MC is significantly positive at the 1% level. In both the treatment group and the control group, the direct effect is significantly negative, while the indirect effect is significantly positive, indicating that MC plays a mediating role in the relationship between DI and the development of NQPFs. Upon decomposition, the total effect remains significant, confirming that DI significantly promotes the development of NQPFs by increasing MC. The negative direct effect of stripping MC suggests that in the short term, DI may suppress the development of NQPFs due to firm adaptation costs and organizational restructuring. However, the positive value of the indirect effect reveals that DI, through MC, enables companies to acquire more market resources and scale advantages, ultimately driving the enhancement of NQPF. Therefore, MC serves as a critical pathway through which DI promotes the development of NQPFs. Companies should leverage the MC effect brought about by DI to achieve a dual enhancement of economies of scale and innovation efficiency.

Specifically, digital innovation enhances MC through the following channels: First, the accumulation of data elements brings scale and network effects, enabling enterprises with DI capabilities to gain larger market shares. Second, the integration effect of digital platforms promotes the effective allocation of upstream and downstream resources in the industrial chain, improving market efficiency. Third, digital technology reduces transaction costs and optimizes the market structure. Higher MC further promotes the concentration of innovation resources toward advantageous enterprises, forming a virtuous cycle.

4.5.3. Government Innovation Subsidies

To investigate the mediating effect of GISs in the linkage between DI and the advancement of NQPFs, this study quantifies GISs based on the total annual amount of innovation subsidies awarded to companies.

As shown in Table 8, the total effect of DI in promoting the development of NQPFs by facilitating GIS is significantly positive at the 1% level. In the treatment group, the direct effect is positive, while the indirect effect is negative; in contrast, in the control group, the direct effect is negative, while the indirect effect is positive. This reveals a significant heterogeneity in the mediating role of GISs. A deeper analysis shows that for companies actively engaged in DI, the intervention of GISs may weaken the positive effects of DI. In contrast, for companies with a weaker foundation in DI, GISs can effectively compensate for their lack of innovation capabilities and promote the enhancement of NQPFs. Therefore, there exists a complex interactive relationship between GISs and DI, with the effect highly dependent on the levels of DI.

In conclusion, digital innovation enables the development of new quality productivity through three main channels: promoting IURC, increasing MC, and facilitating GISs. This multi-dimensional transmission mechanism reveals the systematic impact of DI on NQPFs and provides important references for policy-making. Future efforts should further strengthen the coordination between DI and various transmission mechanisms to fully leverage the positive role of DI in promoting NQPF development.

5. Discussion and Recommendations

5.1. Discussion

Utilizing the dual machine learning method, this study elucidates the impact of DI on the development of the NQPF and its underlying mechanisms. The results substantiate the positive relationship between DI and NQPFs and also identify crucial mechanism pathways while revealing significant regional heterogeneity effects.

The research establishes a robust positive correlation between DI and NQPFs, thereby advancing the current literature on the relationship between digital technologies and productivity. Previous studies have predominantly focused on the effects of DI on conventional productivity metrics. For instance, Wang et al. (2023) [68] demonstrated a positive relationship between DI and total factor productivity, and Huang et al. (2023) [80] observed that DI enhances labor productivity. However, these studies primarily measure quantitative increases in productivity rather than qualitative shifts. In contrast, this study develops NQPF indicators, which encompass dimensions such as high-quality development, innovation-driven growth, and digital transformation. These indicators illustrate how DI induces a fundamental leap in productivity, beyond simple efficiency improvements. This finding aligns with the research by Zhang (2024) [81], which emphasizes that digital NQPFs, driven by DI, can facilitate a leap in productivity levels. Our empirical results corroborate this assertion, providing evidence that DI drives a systemic transformation in production methods rather than mere incremental improvements.

The transformation of productivity driven by DI is fundamentally distinct from traditional technological innovation paradigms. While revolutions in technology, such as mechanization and electrification, have historically promoted productivity growth by enhancing physical production efficiency, DI induces a profound transformation in productivity by enabling comprehensive process optimization, restructuring value networks, and refining decision-making mechanisms. This argument transcends the confines of traditional technological determinism, highlighting DI as a pivotal factor driving the qualitative leap in productivity. Such a transformation is not only relevant in the context of China but is also evident in the global expansion of the digital economy.

Building on this, we identify three critical pathways through which DI influences NQPFs. The first pathway involves the IURC mechanism, which illustrates the fundamental enhancement of innovation network efficiency facilitated by DI. Unlike conventional IURC models, digital technologies enable the overcoming of spatial-temporal limitations and organizational boundaries in the production and flow of knowledge. Jie (2010) [82] examined how firms interact with universities and research institutions to establish technological and knowledge chains, but this interaction is often hindered by high communication costs and information asymmetry in traditional settings. In contrast, DI significantly reduces collaboration barriers through virtual R&D platforms, data-sharing mechanisms, and intelligent matching algorithms, accelerating knowledge integration and shifting the collaboration model from a linear point-to-point structure to a comprehensive, networked system. Wu (2012) [83] suggested that university research activities stimulate companies to increase their investment in innovation, thereby effectively utilizing external knowledge spillovers. Our study further highlights how DI enhances knowledge production and sharing, expanding the reach and depth of this complementary effect.

MC functions as the second key intermediary variable, reflecting the structural implications of DI on industry organization. Existing research in industrial economics often identifies an inverse relationship between market concentration and innovation. Empirical studies by Yan and Feng (2005) [84] and Tang and Tang (2004) [85] suggest that elevated market concentration may dampen corporate innovation, leading to a decline in industry-wide innovation. Nevertheless, our study demonstrates that DI, by increasing market concentration, plays a pivotal role in advancing NQPF development. This finding requires interpretation through the lens of the unique operational logic inherent to the digital economy. Liu et al. (2019) [86] observed that innovation activities in the digital economy tend to foster a “winner-takes-all” market environment, and our study further elucidates how this concentration translates into productivity gains. The concentration instigated by DI differs fundamentally from traditional monopolistic concentration; it is propelled by network effects and data-scale effects, which channel resources towards high-efficiency firms, giving rise to an “efficiency-driven concentration” rather than a “rent-seeking-driven concentration”. This concentration model optimizes resource allocation, thereby contributing to the overall enhancement of productivity. This phenomenon, prevalent in the global digital platform economy, is exemplified by platforms like Amazon and Alibaba, which leverage innovation-driven market concentration to improve the efficiency of the entire value network.

The third significant intermediary mechanism identified in our study is government innovation subsidies. We find that DI significantly enhances firms’ capacity to access government innovation support, thereby accelerating the development of NQPFs. This insight adds depth to our understanding of the dynamic interplay between innovation policies and corporate behavior. An et al. (2009) [87] examined how firms strategically engage in innovation activities to acquire government subsidies, focusing primarily on their proactive adaptation. Our research, however, extends this by showing that DI enhances firms’ technological leadership, transparency in information, and project implementation capacity, positioning them more advantageously in government resource allocation. Yang and Rui (2020) [88] found that high-tech firms receive more government subsidies, and DI boosts the likelihood of firms obtaining such certifications. Bai et al. (2022) [89] emphasized that the positive externalities of innovation underpin the theoretical rationale for government subsidies, while our research further demonstrates how DI amplifies these externalities, reinforcing the legitimacy and targeting of government support.

This study identifies significant regional heterogeneity effects, demonstrating that the impact of DI on NQPFs is characterized by pronounced spatial variation. In contrast to the homogeneous assumptions of traditional technology diffusion models, the effects of DI are deeply contingent upon regional innovation environments. Huang and Zhuang (2007) [90] analyzed the disparities in innovation diffusion from a network structure perspective, while our research reveals that the effectiveness of DI hinges on the collaborative interaction of several factors, such as regional digital infrastructure, innovation capacity, and the institutional environment. Wang et al. (2023) [91] empirically confirmed that DI has a more pronounced effect on enhancing investment efficiency in regions with advanced digital infrastructure, a finding that corroborates our regional heterogeneity results and underscores the path dependence of DI.

Taken together, the findings of this study present a comprehensive framework for understanding the relationship between innovation and productivity in the digital era. DI plays a central role in the development of NQPFs by restructuring innovation networks, optimizing industrial structures, and invigorating institutional environments. While this process is significantly influenced by regional conditions, it also provides potential pathways for reducing regional disparities. Although this study is based on data from China, DI, as a global technological phenomenon, is likely to exhibit shared characteristics across various economies. Understanding the mechanisms through which DI influences NQPFs offers critical insights for countries to foster innovation and achieve productivity transformation.

5.2. Policy Recommendations

Based on these conclusions, the following policy recommendations are proposed:

First, enhance support for DI to drive the rapid development of NQPFs. DI plays a critical role in advancing NQPFs, but many companies lack sufficient investment and long-term innovation planning, hindering its potential benefits. Therefore, it is crucial to intensify support for DI to enable a transformation in productivity and ensure sustainable development. Firstly, efforts should focus on strengthening digital infrastructure, enhancing technical support such as 5G, cloud computing, and big data platforms to enable companies to more effectively participate in DI and drive the development of NQPFs. Secondly, regional DI pilot zones should be established based on regional resource endowments and characteristics. Localities should tailor DI strategies according to differences in natural resources, industrial structures, and market demands in order to promote the enhancement of NQPFs and coordinated regional economic development. Finally, the government should focus on long-term talent development and the recruitment of high-end digital technology professionals, providing continuous talent support for DI and ensuring the sustainable and healthy development of NQPFs. In advancing digital technology innovation, it is crucial to remain vigilant against the systemic risks that may arise from the overuse of digital technologies. As digital technologies become increasingly integrated into both production and daily life, they pose risks such as heightened cybersecurity threats, increased system vulnerabilities, and excessive dependency on digital infrastructures. Therefore, alongside bolstering support for digital innovation, it is imperative to establish a corresponding risk prevention and control system. Specifically, a risk assessment and early warning mechanism for digital technology should be implemented to comprehensively assess the social, economic, and security impacts of DI and develop tiered risk management measures to ensure the coordinated development of DI and security, avoiding an imbalance between innovation and risk.

The second recommendation is to deepen the cooperation mechanism for IURC in the digital era to facilitate the synergistic development of NQPFs. Thus, the government should harness the characteristics of DI, enhance the integration of digital and traditional industries, and further refine the mechanisms for IURC. Firstly, deepen industry–academia–research cooperation under digital transformation by establishing a dedicated cooperation fund, providing guidance on digital technologies, and allocating digital resources. This would help companies actively engage in DI and enhance their NQPFs. Secondly, universities, research institutes, and companies should collaborate to create innovation alliances and new research organizations, fostering cross-sectoral and cross-disciplinary innovation. By optimizing the allocation of DI resources and improving the efficiency of technology conversion, we can optimize resource allocation, reduce consumption, and achieve sustainable and green development goals, thereby driving the overall enhancement of NQPFs. Finally, the government should revise the Intellectual Property Law and improve the protection of data property rights to ensure long-term legal protection for innovations, thereby promoting their continuous transformation and efficient use and supporting the ongoing healthy development of NQPFs. However, the issue of data security in IURC has become increasingly critical. Issues such as data misuse, leakage, and unclear ownership can significantly hinder the healthy development of the digital innovation ecosystem. To address this, data classification and protection standards should be established, clearly defining the security responsibilities of each party in the data-sharing process and creating a robust data security governance framework for IURC. A multi-stakeholder data ethics committee should be established to oversee the ethical use of sensitive data, ensuring a balance between fostering innovation and protecting privacy and ensuring that the use of data promotes innovation without infringing on individual rights or public interests, thereby facilitating the coordinated development of data security and DI.

The third recommendation is to clarify the dynamic relationship between government guidance and market leadership to facilitate the coordinated development of NQPFs. Research demonstrates that MC and GISs are crucial mechanisms through which DI drives the development of NQPFs. Government subsidies, in particular, play a more prominent role in supporting companies with weaker DI foundations. Accordingly, the following policy recommendations are proposed: First, establish a hierarchical DI support mechanism to prioritize support for enterprises with weaker DI capabilities while gradually advancing the market-oriented development of enterprises with stronger DI foundations. Additionally, a regularized policy effect evaluation system should be implemented to ensure effectiveness. Second, improve policies on enterprise mergers, acquisitions, and restructuring to support innovative enterprises in increasing their market share through resource integration. Establish a mechanism for sharing DI elements to fully leverage economies of scale and enhance NQPFs. Finally, build a DI ecosystem that integrates government guidance and market mechanisms. Enhance platforms for sharing DI resources and establish a risk-sharing mechanism to jointly promote the coordinated development of NQPFs. Nevertheless, it is important to recognize that excessive market concentration can result in an uneven allocation of innovation resources, restricting the growth opportunities for small and medium-sized enterprises. A few large technology firms, leveraging their advantages in data, algorithms, and platforms, may create monopolistic market structures, stifling fair competition, impeding innovation, and widening the digital divide, which ultimately undermines the sustainable development of the digital economy. Therefore, alongside clarifying the roles of government guidance and market dominance, it is critical to strengthen antitrust regulations in the digital market. Establishing a fair competition monitoring mechanism within the digital economy is essential for identifying and correcting market inequities, preserving market diversity and dynamism while fostering digital innovation, and ensuring a balanced relationship between competition and innovation.

The fourth recommendation is to promote the balanced development of NQPFs across regions by tailoring policies to local conditions. Research indicates that the positive enabling effect of DI on NQPFs varies significantly across regions. Therefore, policymakers should develop region-specific policies based on the natural resources, economic conditions, and social development needs of each area to facilitate the balanced growth of NQPFs. First, for megacity behemoths and megacities, efforts should focus on further improving high-end science and technology parks and innovation incubation bases, fostering cross-industry collaboration to establish innovation ecosystems, and implementing talent incentive policies to attract top global scientific and technological talent, thereby enhancing the international competitiveness of innovation teams. Second, for large cities and SMCs, it is important to focus on enhancing infrastructure investments, supporting the development of innovation-oriented industrial parks, advancing energy efficiency, promoting environmental technologies, and creating regionally distinctive industrial clusters that improve enterprise innovation capacity and drive the advancement of NQPFs. Third, in regions along the Hu Line and the southeast, efforts should emphasize guiding and supporting high-tech industries and focusing on creating “green innovation highlands”. In contrast, for the northwest region, specialized policy pilot programs should be established to support the digital and intelligent transformation of local enterprises, thereby stimulating their innovation capacity. Finally, non-old industrial base cities should accelerate industrial transformation and upgrading, particularly by guiding traditional manufacturing industries toward smart and green manufacturing. Meanwhile, old industrial base cities should introduce “digital factory” projects to facilitate the in-depth integration of traditional manufacturing industries with digital technologies, ultimately realizing the transformation and upgrading of productive forces into NQPFs. Additionally, while promoting the balanced development of NQPFs across regions through digital innovation, attention must be paid to the problem of excessive energy consumption associated with the development of digital technologies. Without proper regulation, this could conflict with national carbon peaks and carbon neutrality goals, exacerbating regional imbalances in energy distribution. Therefore, it is necessary to set energy consumption standards for digital infrastructure, promote the development of green algorithms and low-power computing technologies, and encourage the integration of renewable energy with digital infrastructure. By combining technological innovation with institutional guidance, a virtuous cycle between the digital economy and green development can be achieved, ensuring that digital technologies contribute to NQPF development without becoming a drain on energy resources, but rather fostering green, low-carbon development.

5.3. Limitations and Directions for Future Research

First, due to the availability of research sample data, this paper selects data from A-share listed companies from 2011 to 2022, which primarily reflects the DI characteristics of large and medium-sized companies. However, the relationship between DI and NQPFs in small and micro enterprises and non-listed companies remains unclear. Future research could expand the sample scope by including data from more types of enterprises to enhance the generalizability of the findings.

Second, this paper incorporates IURC and MC from the inter-company dimension, as well as innovation subsidies from the government level, to systematically analyze their mediating roles between DI and NQPFs. However, it lacks an investigation into the intra-company mechanisms. Future research could integrate intra-company factors into the framework to explore how they interact with inter-company and government-level variables, thereby offering a more comprehensive understanding of the mechanisms that enhance NQPFs.