Digital Transformation and Firm Innovation: A Dual-Path Analysis of R&D Investment and Governance Mechanisms

Abstract

With the digital economy advancing at a fast pace, digital transformation plays a pivotal role in reinforcing firms’ innovation capability and promoting high-quality development. This study analyzes Chinese non-financial publicly listed firms on the A-share market over the period 2009–2023. Based on text mining of annual reports, this study constructs an index capturing digital transformation and empirically evaluate its impact on innovation output with firm and year fixed effects. The estimates suggest that digital transformation meaningfully increases firms’ innovation output; the inference is unchanged when applying instrumental-variable approaches and conducting extensive robustness checks. Mechanism analysis reveals two parallel channels: (1) the R&D investment mechanism, characterized by improvements in R&D intensity, capitalization rate, per capita efficiency, and investment growth; (2) the governance environment mechanism, reflected in enhanced internal control, improved information disclosure quality, and strengthened audit supervision. Once firms are stratified by characteristics, the estimated positive effect of digital transformation is most pronounced for firms with low financial constraints, large size, eastern locations, and state ownership. This study identifies both direct and indirect mechanisms linking digital transformation to innovation and highlights how firm- and region-specific features condition the magnitude of this effect, thereby offering empirical implications for corporate digitalization strategies and policy design.

1. Introduction

As the digital economy accelerates, digital transformation has become increasingly important for firms seeking innovation-driven and high-quality development (Jiang and Li, 2024 [1]; Duan and Zhang, 2025 [2]). Digital technologies such as artificial intelligence, big data analytics, cloud computing, and blockchain are no longer merely technical tools; rather, they are increasingly embedded in firms’ production systems, management processes, and business models (ElMassah and Mohieldin, 2020 [3]; Su and Wu, 2024 [4]). In China, digital transformation has also been promoted as a strategic priority through policies encouraging the deeper integration of digital technologies with the real economy (Liu et al., 2022 [5]; Li et al., 2022 [6]). However, although digital transformation has attracted growing scholarly attention, how it is associated with firm innovation, through which internal mechanisms this relationship operates, and under what conditions the relationship becomes stronger remain insufficiently clarified.

From a theoretical perspective, innovation is a core driver of firms’ sustained competitive advantage. Innovation output is often shaped by sustained R&D spending, effective resource allocation, and a sound governance environment (Fabrizio and Tsolmon, 2014 [7]; Massey and Johnston-Miller, 2016 [8]; Yin et al., 2023 [9]). Digital transformation influences these critical factors through multiple channels. On one hand, it helps break down information barriers, reduce asymmetries and financing constraints, thereby enabling increased R&D investment and efficiency (Aben et al., 2021 [10]; Antonio and Sorin, 2023 [11]; Zhang et al., 2025 [12]). On the other hand, it facilitates improvements in internal controls, enhances information transparency, and strengthens external supervision—thus fostering a more stable institutional environment conducive to innovation (Keller et al., 2021 [13]; Cordery et al., 2023 [14]). However, the literature remains divided regarding the precise pathways and boundary conditions through which digital transformation affects innovation. While some studies affirm its significant role in improving R&D productivity and overall innovative performance (Li et al., 2023 [15]), others argue that such effects may vary depending on firm size, financing conditions, and regional development levels (Yang and Han, 2023 [16]). This divergence calls for a more systematic empirical investigation grounded in a coherent theoretical framework and supported by large-scale micro-level data.

From a practical perspective, Chinese firms display considerable variation in their levels of digital transformation. On one hand, some large enterprises and firms located in eastern regions have leveraged superior resources and policy support to achieve relatively advanced levels of digitalization and intelligent operations, accompanied by strong R&D capabilities and innovation outputs. On the other hand, constraints in financing, human capital, and access to technology leave SMEs and firms in central and western regions at a clear disadvantage, leaving their digital transformation relatively behind. Such an imbalance can weaken firms’ innovation capability and, at the same time, impede the formation of a balanced and resilient cross-industry innovation ecosystem (Hu et al., 2023 [17]; Li et al., 2023 [18]; Wu et al., 2024 [19]). Therefore, a systematic investigation of how digital transformation influences innovation output, including its underlying mechanisms, will not only contribute to a better understanding of firm-level innovation dynamics but also inform regional development and industrial upgrading policies.

Despite the growing literature on digital transformation and corporate innovation, several important questions remain insufficiently clarified. This study is therefore guided by three research questions. First, is corporate digital transformation positively associated with firm innovation output? Second, do R&D investment and the governance environment provide mechanism-consistent channels through which digital transformation relates to innovation? Third, under what firm-level and regional conditions is the digital transformation–innovation relationship more pronounced? By addressing these questions, this study aims to provide a more structured understanding of how digital transformation is linked to innovation output and why this relationship may vary across firms.

We study Chinese A-share non-financial listed firms from 2009 to 2023, focusing on how digital transformation relates to innovation output and on the mechanism pathways behind this relationship. Methodologically, we develop a firm-level index of digital transformation by applying text-mining methods to the MD&A narratives in annual reports. Firm and year effects are included in the baseline model; endogeneity is examined via instrumental variables; robustness is further assessed through multiple checks. Furthermore, we identify two parallel mechanisms: (1) the R&D investment mechanism, whereby digital transformation improves R&D intensity, capitalization, efficiency, and investment growth; and (2) the governance environment mechanism, through which digital transformation enhances internal control, improves information disclosure, and strengthens audit supervision. Heterogeneity is explored by stratifying firms by key attributes and external conditions and comparing the estimated innovation effects of digital transformation.

Although prior studies have provided important insights into the relationship between digital transformation and corporate innovation, the existing literature remains limited in three respects. First, many studies emphasize the resource-allocation or efficiency-enhancing role of digital transformation, but pay less attention to how digital transformation reshapes the governance and information environment in which innovation decisions are made. Second, studies that examine governance or information effects often treat them separately from firms’ R&D resource allocation, leaving the interaction between internal innovation-input mechanisms and external governance-support mechanisms insufficiently theorized. Third, existing research has not fully explained why the innovation implications of digital transformation differ across firms with different resource endowments, financing conditions, ownership structures, and regional environments. Therefore, this study seeks to provide an integrated dual-path framework that links digital transformation to firm innovation through both R&D resource allocation and governance environment improvement.

The principal contribution of this study lies in developing and empirically examining a dual-path framework that links digital transformation to firm innovation through R&D resource allocation and the governance/information environment. The text-based measurement, Chinese listed-firm panel evidence, and heterogeneity analysis are used to support and contextualize this theoretical framework rather than being treated as separate core contributions.

This study makes three main contributions. First, it develops a dual-path theoretical framework that explains how digital transformation is associated with firm innovation through both R&D resource allocation and the governance environment. This framework moves beyond a single resource-efficiency explanation and highlights the coexistence of internal innovation-input mechanisms and external governance-support mechanisms.

Second, this study deepens the theoretical understanding of digital transformation by treating it not merely as a process of technological adoption but as an organizational transformation process that reshapes knowledge processing, resource allocation, and information transparency. In doing so, the paper connects digital transformation more explicitly with the organizational foundations of innovation.

Third, this study clarifies the boundary conditions of the digital transformation-innovation relationship by examining how financing constraints, firm size, regional location, and ownership structure condition the strength of this association. This contributes to a more differentiated understanding of why the innovation implications of digital transformation vary across firms and institutional environments.

2. Theoretical Analysis and Research Hypothesis

2.1. Core Theoretical Foundation: Resource-Based View and Complementary Perspectives

This study adopts the Resource-Based View as its central theoretical foundation. From this perspective, digital transformation can be understood as a process through which firms convert data resources, digital technologies, organizational routines, and knowledge-processing capabilities into strategic resources that support innovation. This theoretical lens is appropriate because the innovation implications of digital transformation depend not only on the adoption of digital tools but also on whether firms can transform these tools into valuable, rare, and difficult-to-imitate organizational capabilities. Digital transformation may therefore enhance firms’ innovation capacity by improving the availability, recombination, and utilization of strategic resources.

Dynamic capability theory and information asymmetry theory are used as complementary perspectives. Dynamic capability theory helps explain how firms reconfigure digital resources and organizational processes in response to technological and market changes. Information asymmetry theory and agency theory help explain why digital transformation may improve the governance and information environment surrounding innovation activities. In this sense, the theoretical framework of this study is centered on the Resource-Based View, while the complementary theories are used to explain the two specific channels emphasized in this paper: R&D resource allocation and governance environment improvement.

In the era of the digital economy, corporate innovation models and sources of competitive advantage are profoundly shaped by digital transformation (DT). Unlike the traditional perception of digitalization as a mere adoption of information technology tools, DT refers to a systemic transformation that integrates frontier digital tools, such as AI, data analytics, cloud infrastructure, and blockchain solutions, into firms’ production, management, and decision-making. This transformation restructures resources, reengineers workflows, and optimizes governance systems, thereby exerting significant influence on innovation (Miklosik and Evans, 2020 [20]; Massaro, 2023 [21]; Ma and Chang, 2024 [22]). Its impact can be interpreted from three interrelated perspectives.

First, the resource and capability perspective. RBV argues that an enduring competitive advantage arises when firms control resources that are scarce and not easily imitated or replaced. DT enables the datafication of resources and the sharing of information, which facilitates the rapid integration of knowledge, technologies, and market intelligence. This process reduces information processing and communication costs in R&D and enhances the efficiency of knowledge recombination and application (Li et al., 2021 [23]). In terms of dynamic capability theory, DT enhances organizational capabilities in opportunity sensing, knowledge absorption, and agile responses to uncertainty. Consequently, firms can better balance exploratory and exploitative innovation, thereby maintaining innovation vitality in dynamic environments (Ceipek et al., 2021 [24]; Kowalski et al., 2024 [25]).

Second, an R&D-based view suggests that innovation outcomes hinge on how much firms invest in R&D, how those inputs are allocated, and how efficiently they are converted into outputs. R&D decisions are inherently linked to financing availability, cost efficiency, and strategic orientation—all of which are significantly influenced by digitalization. DT lowers transaction and organizational costs in R&D activities by enabling intelligent resource allocation and optimized information flows (Liu et al., 2023 [26]; Cai et al., 2024 [27]). It also reinforces data-driven strategic judgments, encouraging firms to engage in long-term, capitalized R&D projects that generate higher-value innovation outputs (Appio et al., 2021 [28]). In this sense, DT promotes innovation indirectly through the “cost reduction—efficiency improvement–structural optimization” pathway, which simultaneously enhances the intensity and quality of R&D activities.

Third, from a governance and information-transparency angle, innovation inherently involves substantial risk and uncertainty, which can intensify agency conflicts and deter investors. From the lens of principal–agent theory and information asymmetry theory, DT can improve governance by enhancing internal control, information disclosure, and audit mechanisms (Manita et al., 2020 [29]; Li et al., 2024 [30]). On one hand, digital tools strengthen the authenticity and traceability of internal information, thereby mitigating managerial opportunism (Yang et al., 2021 [31]). On the other hand, improved disclosure quality and more effective audit supervision enhance investor confidence in firms’ R&D strategies and innovation projects, increasing both capital availability and market recognition (Zhang et al., 2025 [32]). Thus, improvements in governance and information transparency not only lower agency costs but also amplify the marginal effects of DT on innovation outcomes.

Taken together, DT contributes to corporate innovation through both direct and indirect channels. Directly, it enhances knowledge recombination and innovation efficiency. Indirectly, it increases R&D investment and strengthens governance mechanisms, thereby creating a more supportive institutional environment for innovation.

2.2. Mathematical Model Derivation

Before presenting the hypotheses, we provide a simplified analytical model to formalize the theoretical logic discussed above. The purpose of this model is not to establish an independent structural estimation framework or to provide causal identification. Rather, it serves as a supplementary theoretical device that clarifies why digital transformation may be linked to innovation through R&D investment and the governance environment. Therefore, the model should be read as an analytical extension of the theoretical discussion rather than as a substitute for the empirical research design.

Corporate innovation output reflects both knowledge accumulation and allocation efficiency in resource use. Digital transformation (DT) enhances innovative output by raising the efficiency of firms’ data processing and knowledge recombination. Let the innovation output of firm i at time t be denoted as Innov, which is a function of R&D investment (RD), governance quality (Gov), and the degree of digital transformation (DIGI). A simplified representation can be expressed as:

$$Innov=A·\varphi (DIGI)·{RD}^{\alpha }·g{(Gov)}^{\beta }$$

(1)

Here, $\varphi (DIGI)$ reflects the digital-transformation–induced gain in the efficiency of knowledge transformation, and $\varphi ’(DIGI)>0$ and $g(Gov)$ captures the extent to which governance quality amplifies innovation output. This relationship indicates that digital transformation not only can directly increase the knowledge productivity of enterprises, but also operates through R&D activities and the governance environment to shape innovation indirectly.

From the perspective of R&D decision-making, firms weigh the marginal benefits against the marginal costs of R&D investment. While R&D costs typically increase with investment scale, DT reduces effective costs by mitigating financing constraints and minimizing information frictions. The R&D cost function can be expressed as:

$$C(RD)={\displaystyle \frac{\chi }{2}}{\displaystyle \frac{{RD}^2}{1+\zeta DIGI}}$$

(2)

Among them, $\mathsf{\zeta }\mathrm{D}\mathrm{I}\mathrm{G}\mathrm{I}$ reflects the effect of the alleviation of financing constraints brought about by digital transformation. The denominator is increasing in $DIGI$, which means that digitalization reduces the marginal cost of one additional unit of R&D.

The expected return from R&D depends on both the scale of investment and knowledge transformation efficiency. The return function can be defined as:

$$B(RD)=\kappa ·\varphi (DIGI)·{RD}^{\theta }, 0<\theta <1$$

(3)

Among them, $\varphi (DIGI)$ represents the extent of digital transformation–driven improvements in R&D output efficiency. Optimal R&D occurs at the level where the marginal benefit of R&D just matches its marginal cost:

$${\displaystyle \frac{\partial B}{\partial RD}}={\displaystyle \frac{\partial C}{\partial RD}}$$

(4)

Substituting (2) and (3) yields:

$$\kappa \theta \varphi (DIGI){RD}^{\theta -1}=\chi ·{\displaystyle \frac{RD}{1+\zeta DIGI}}$$

(5)

Solving this condition indicates that:

$${\displaystyle \frac{\partial {RD}^{*}}{\partial DIGI}}>0$$

(6)

This result implies that higher levels of DT reduce the effective marginal cost of R&D while simultaneously enhancing R&D output efficiency, thereby increasing optimal R&D investment and indirectly promoting innovation output. This provides theoretical support for the R&D mechanism proposed in this study.

Beyond the R&D channel, governance mechanisms also constitute a critical pathway through which DT influences innovation. Given the high risk and uncertainty of innovation, external supervision and governance quality become indispensable. Innovation output can thus be further expressed as:

$$Innov=\lambda ·DIGI·p(Gov)$$

(7)

Here, $p(Gov)$ represents the degree of support provided by the external governance environment for innovation, and $p’(Gov)$ is greater than 0. This functional form shows that better governance strengthens the innovation gains associated with digital transformation at the margin.

Here, improvements in the governance environment strengthen DT’s innovation effect at the margin. Decomposing governance quality, we assume it is jointly determined by internal control (IC), disclosure quality (KV), and audit quality (AQ):

$$p(Gov)={\psi }_0+{\psi }_{1}IC+{\psi }_{2}KV+{\psi }_{3}AQ$$

(8)

Substituting (8) into (7) gives:

$${\displaystyle \frac{\partial Innov}{\partial DIGI}}=\lambda ·p(Gov), {\displaystyle \frac{\partial Innov}{\partial Gov}}=\lambda ·DIGI·p’(Gov)>0$$

(9)

Equation (9) illustrates the complementarity between governance and digital transformation: when internal controls are stronger, disclosures are more transparent, and audits more effective, the innovation-enhancing effect of DT is amplified.

2.3. Research Hypotheses

Building on the preceding analysis, we argue that digital transformation affects innovation output both directly and indirectly, with R&D investment and governance serving as the key transmission mechanisms. On this basis, we formulate three hypotheses as follows.

Digital transformation, as a systematic change, can directly enhance the innovation efficiency of enterprises by improving the rate of knowledge conversion, optimizing the methods of information processing and resource allocation. In Equation (1), the positive effect of $\varphi (DIGI)$ implies that higher digitalization speeds up knowledge creation and application, thereby raising innovation output. Therefore, Hypothesis 1 is proposed:

H1. 

Digital transformation exerts a significantly positive impact on enterprises’ innovation output.

R&D activities are a central engine of innovation output, and the amount invested in R&D as well as the efficiency of that investment jointly determine the quantity and quality of innovation outcomes. Models (2)–(6) imply that digital transformation raises the optimal R&D investment level by easing financing constraints and lowering marginal costs, while also improving R&D output efficiency. Put differently, digital transformation improves innovation efficiency directly and also boosts innovation output through higher R&D intensity and a more optimized R&D structure. The enhancement should be reflected across multiple R&D metrics, including intensity, capitalization, growth, per capita input, and the expense ratio. Therefore, Hypothesis 2 is proposed:

H2. 

Digital transformation is positively associated with R&D investment and R&D efficiency, which provide a plausible mechanism-consistent channel linking digital transformation to firm innovation.

Innovative activities typically involve substantial risk and long horizons, which can intensify agency conflicts and undermine external investors’ trust. Models (7)–(9) suggest complementarity between digital transformation and governance quality, such that stronger governance amplifies the innovation-enhancing effect of digital transformation. Specifically, digital tools can strengthen external monitoring and resource support by enhancing the effectiveness of internal controls, increasing disclosure transparency, and reinforcing audit oversight. These changes mitigate information frictions and agency costs during innovation activities, thereby amplifying DT’s marginal innovation payoff. Thus, Hypothesis 3 is proposed:

H3. 

Digital transformation is positively associated with improvements in the governance and information environment, which provide a plausible mechanism-consistent channel linking digital transformation to firm innovation.

3. Study Design

3.1. Samples and Data

The dataset covers non-financial firms that are publicly listed as A-shares in Shanghai and Shenzhen, observed from 2009 through 2023. To secure a reliable and continuous panel dataset, we conducted the following screening process. First, we eliminated firms classified as ST or *ST and those with major operational irregularities to reduce potential distortions attributable to abnormal operating conditions. Second, to address data-quality concerns, we filter the sample by omitting cases with incomplete key variables and unusually extreme values, which helps curb estimation bias.

The resulting sample contains 20,254 firm-year observations. This balanced and comprehensive sample underpins our assessment of the innovation effects of digital transformation and aligns with the econometric specifications used in the subsequent analyses.

3.2. Definition of Variables

3.2.1. Innovation Output

To measure innovation output, we use the number of invention patents, which cover new technical solutions applied to products, production processes, or their improvements. Following prior studies (He et al., 2018 [33]), this study measures firms’ innovation output using the natural logarithm of the total number of invention patent applications plus one. The use of patent applications rather than granted patents is justified because application years more accurately reflect the timing of innovation activities, whereas the granting process typically involves considerable delays. This approach thus captures the contemporaneous effects of corporate strategies and resources on innovation outcomes.

3.2.2. Digital Transformation

To construct the DIGI index, we first built a digital-transformation keyword dictionary based on prior studies and the terminology commonly used in Chinese annual reports. The dictionary covers several categories of digital technologies and applications, including artificial intelligence, big data, cloud computing, blockchain, digital platforms, intelligent manufacturing, industrial internet, data management, information systems, and digital business models (Brown et al., 2024 [34]; Yang et al., 2025 [35]). To reduce fragmentation caused by different expressions of similar digital concepts, related synonyms and semantically similar terms were grouped into broader categories before standardization.

After extracting the frequency of digital-transformation-related terms from the MD&A section, we standardized the term-frequency variables and applied an entropy-based weighting method. The entropy method assigns greater weights to terms with higher cross-firm variation, thereby reducing the influence of terms that appear frequently but provide limited discriminatory information. The resulting DIGI index is used as a text-based proxy for firms’ digital transformation orientation and intensity.

We acknowledge that this measure may capture both substantive digital transformation and disclosure intensity. Some firms may strategically emphasize digitalization in annual reports without making equivalent real investments. However, comparable firm-level digital investment data are not consistently disclosed in financial statements over the full 2009–2023 sample period. Digital-related expenditures may be reported under software, intangible assets, information systems, R&D equipment, or management expenses, making it difficult to construct a consistent and reliable indicator across firms and years. Therefore, this study retains the text-based DIGI index while interpreting it cautiously as a proxy for digital transformation orientation and intensity.

3.2.3. Control Variables

Omitted-variable bias is mitigated by adding controls that capture firm attributes, performance and valuation, governance arrangements, external audit oversight, and ownership structure, which also enhances the robustness of the results.

Firm characteristics. The controls comprise firm size (Size), the natural logarithm of total assets; firm age (Age), the log of years since establishment (observation year minus establishment year); leverage (Lev), total liabilities scaled by total assets; tangibility (Tang), tangible assets scaled by total assets; and capital intensity (CapInt), total assets divided by operating revenue.

Operating performance and market value. We additionally control for return on assets (Roa), measured as net profit divided by total assets; Tobin’s Q (TobinQ), measured as market value over total assets; and revenue growth (Growth), defined as the year-on-year change in operating revenue relative to the previous year.

Corporate governance. We control for corporate governance using Top1 (largest shareholder ownership), Board (log directors), ManHold (managerial ownership), and DUAL, an indicator equal to 1 when the CEO concurrently holds the chair position and 0 otherwise.

External auditing and ownership structure. We also control for a Big Four auditor dummy (AUDIT), set to 1 when the firm’s financial statements are audited by a Big Four accounting firm and 0 otherwise, and institutional ownership (InsHold), measured as institutional investors’ shareholdings divided by total shares outstanding.

Table 1. Variable definitions.

Variables	Explanation	Definition
Innov	Innovation output	The logarithm of one plus the total number of invention patent applications
DIGI	Digital transformation	Standardizing the frequency data related to digital transformation
Size	Firm Size	ln(total assets)
Roa	Return on Assets	Net profit/total assets
Age	Firm Age	ln(observation year—establishment year)
Lev	Leverage Ratio	Total liabilities/total assets
Tang	Tangible Asset Ratio	Total tangible assets/total assets
TobinQ	Tobin’s Q	Market value/total assets
Growth	Revenue Growth Rate	The annual growth rate of operating revenue, computed relative to the previous year
CapInt	Capital Intensity	Total assets/operating revenue
Top1	Largest Shareholder Ownership	Shareholding ratio of the largest shareholder
Board	Board Size	ln(the total number of directors)
ManHold	Management Shareholding Ratio	Number of shares held by the management divided by total shares outstanding
DUAL	CEO-Chairman Duality	A dummy variable set to 1 when the CEO concurrently holds the chair position, and 0 otherwise
AUDIT	Big Four Auditor	Dummy variable coded as 1 for Big Four audit engagement and 0 otherwise
InsHold	Institutional Shareholding Ratio	Institutional investors’ shareholding as a percentage of total shares outstanding

Notes: Table 1 shows the variables definition in this paper.

provides detailed definitions for all variables.

3.3. Model Setting and Descriptive Statistics

3.3.1. Model Setting

To identify the relationship of interest, we include both firm and year effects in the empirical specification:

$${Innov}_{i,t}=\alpha +\beta \times {DIGI}_{i,t}+\gamma \times {Controls}_{i,t}+{\delta }_i+{\theta }_t+{\epsilon }_{i,t}$$

(10)

where ${Innov}_{i,t}$ is firm i’s innovation output in t year, proxied by the logarithm of invention patent applications. ${DIGI}_{i,t}$ is the key explanatory variable, capturing the extent of digital transformation. ${Controls}_{i,t}$ is specified as a covariate vector spanning firm characteristics, operating performance, governance arrangements, and external auditing conditions. We include firm effects (${\delta }_i$) to absorb unobserved, time-invariant firm heterogeneity and year effects (${\theta }_t$) to capture common macro shocks and policy changes over time. This specification helps mitigate biases arising from unobserved firm characteristics and common time trends.

To address location- and sector-specific unobservables, we augment the baseline model with region and industry indicators. This helps isolate the innovation effect of digital transformation from regional gaps in economic development and industry-specific factors.

3.3.2. Descriptive Statistics

Table 2. Descriptive statistics.

	N	Mean	Sd	Min	P50	Max
Innov	20,254	2.0512	1.6158	0.0000	1.9459	7.0622
DIGI	20,254	5.9834	13.7806	0.0000	1.0000	115.0000
Size	20,254	22.5108	1.2561	19.8980	22.3305	26.5794
Roa	20,254	0.0419	0.0518	−0.2152	0.0377	0.2227
Age	20,254	2.9624	0.3215	1.3863	2.9957	3.6889
Lev	20,254	0.4343	0.1839	0.0518	0.4333	0.8771
Tang	20,254	0.9298	0.0777	0.4593	0.9545	1.0000
TobinQ	20,254	1.9305	1.1410	0.7635	1.5674	9.5402
Growth	20,254	0.1349	0.3001	−0.6677	0.0937	2.2761
CapInt	20,254	2.1955	1.7304	0.3212	1.7296	13.1931
Top1	20,254	34.8404	14.9274	8.1100	32.8800	76.4400
Board	20,254	2.2237	0.2318	1.6094	2.1972	2.8904
ManHold	20,254	0.1074	0.1768	0.0000	0.0027	0.6867
DUAL	20,254	0.2447	0.4299	0.0000	0.0000	1.0000
AUDIT	20,254	0.0744	0.2624	0.0000	0.0000	1.0000
InsHold	20,254	46.4139	24.2877	0.0485	48.7467	93.0852

Notes: Table 2 contains the summary statistics for the full set of variables used in the analysis, including dependent, explanatory, and control measures.

summarizes the main variables. Regarding innovation output (Innov), the mean value is 2.0512 with a median of 1.9459, a maximum of 7.0622, and a standard deviation of 1.6158. These figures indicate substantial heterogeneity across firms in terms of patent applications. While a small subset of firms demonstrates high levels of patenting activity, the majority of firms cluster around lower ranges, producing a distribution that is moderately skewed. This pattern is consistent with the reality that innovation investment and output vary significantly across industries and regions in China.

The digital transformation index (DIGI) has an average of 5.9834 (s.d. = 13.7806), with values ranging up to 115, while the median is only 1. This distribution underscores the considerable divergence among firms in advancing digitalization strategies. Some enterprises have achieved relatively high levels of digital application and integration, whereas many others remain in the early stages of digital adoption. Given the wide dispersion, the regressions are well identified, and the innovation gains from digital transformation are likely not uniform across firms. In addition, the results of the correlation analysis related to the main variables are displayed in

Table A1. Correlation analysis.

	Innov	DIGI	Size	Roa	Age	Lev	Tang	TobinQ	Growth	CapInt	Top1	Board	ManHold	DUAL	AUDIT	InsHold
Innov	1
DIGI	0.16	1
Size	0.41	0.03	1
Roa	0.06	−0.05	0.02	1
Age	0.07	0.09	0.16	−0.11	1
Lev	0.14	−0.05	0.46	−0.36	0.06	1
Tang	0.02	−0.07	−0.03	0.03	−0.02	0.05	1
TobinQ	−0.06	0.07	−0.36	0.29	−0.10	−0.32	0.02	1
Growth	0.04	−0.01	0.04	0.25	−0.09	0.06	−0.10	0.07	1
CapInt	−0.13	0.01	0.03	−0.18	0.05	−0.11	−0.11	−0.05	−0.13	1
Top1	−0.01	−0.11	0.18	0.12	−0.18	0.02	0.08	−0.04	0.02	−0.02	1
Board	0.03	−0.06	0.22	−0.02	0.04	0.12	−0.03	−0.10	0.00	0.04	0.02	1
ManHold	−0.01	0.08	−0.28	0.09	−0.10	−0.22	−0.04	0.06	0.04	−0.06	−0.10	−0.21	1
DUAL	0.01	0.09	−0.12	0.04	−0.02	−0.10	−0.03	0.08	0.02	−0.02	−0.06	−0.18	0.23	1
AUDIT	0.16	−0.01	0.37	0.04	0.00	0.11	−0.05	−0.08	0.00	−0.03	0.15	0.08	−0.12	−0.04	1
InsHold	0.09	−0.12	0.42	0.16	−0.06	0.17	0.02	0.00	0.06	0.00	0.53	0.21	−0.65	−0.18	0.25	1

Notes: Within the correlation matrix, the lower triangle is the Pearson correlation coefficient.

of Appendix A.

4. Empirical Results

4.1. Benchmark Regression

Table 3. Benchmark regression.

Variables	(1)	(2)	(3)	(4)	(5)
	Innov	Innov	Innov	Innov	Innov
DIGI	0.019 ***	0.005 ***	0.003 ***	0.003 ***	0.003 ***
	(23.69)	(7.17)	(3.61)	(3.76)	(3.98)
Size			0.456 ***	0.469 ***	0.451 ***
			(25.16)	(25.55)	(23.56)
Roa			0.204	0.020	−0.033
			(1.19)	(0.11)	(−0.19)
Age			0.107	0.161	0.202 *
			(1.01)	(1.48)	(1.86)
Lev			−0.247 ***	−0.263 ***	−0.240 ***
			(−3.35)	(−3.56)	(−3.24)
Tang			−0.476 ***	−0.483 ***	−0.464 ***
			(−3.52)	(−3.57)	(−3.43)
TobinQ			0.011	0.013	0.006
			(1.36)	(1.53)	(0.71)
Growth			−0.022	−0.037 *	−0.043 *
			(−1.01)	(−1.65)	(−1.93)
CapInt				−0.025 ***	−0.025 ***
				(−3.44)	(−3.48)
Top1				0.001	−0.001
				(0.52)	(−0.96)
Board				0.056	0.051
				(1.54)	(1.40)
ManHold				0.387 ***	0.514 ***
				(4.12)	(5.17)
DUAL					0.012
					(0.58)
AUDIT					−0.119 **
					(−2.25)
InsHold					0.003 ***
					(4.05)
Constant	1.936 ***	2.018 ***	−8.025 ***	−8.593 ***	−8.416 ***
	(158.55)	(276.67)	(−15.23)	(−15.78)	(−15.34)
Firm FE	No	Yes	Yes	Yes	Yes
Year FE	No	Yes	Yes	Yes	Yes
Observations	20,254	20,254	20,254	20,254	20,254
R-squared	0.03	0.78	0.78	0.78	0.79

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

reports the baseline regression estimates linking digital transformation to corporate innovation output. The specification in Column (1) excludes both control variables and fixed effects. Even without any controls or fixed effects (Column 1), the estimated DIGI coefficient remains positive and precisely estimated (0.019, p < 0.01), indicating that digitally transformed firms file more patent applications in this simplest specification.

Moving from Column (2) to Column (5), the model is enriched by absorbing firm- and year-level effects and by adding a broader set of controls. Although the estimated coefficient on DIGI becomes smaller, the coefficient remains positive with strong statistical support (p < 0.01) and is approximately 0.003 in magnitude. After absorbing unobserved firm heterogeneity and year effects and adding comprehensive controls, the estimated DT effect on innovation output remains positive, pointing to robust results.

Taken together, the baseline estimates are consistent with H1, consistent with higher innovation output among firms with stronger digital transformation, with the effect estimated precisely. Similar estimates across alternative specifications strengthen confidence in the results. Specifically, digital transformation not only improves information processing and resource allocation efficiency, but also exerts a stable and generalizable impact on innovation activities, even in complex business and governance environments. These results lay a solid empirical foundation for the subsequent mechanism and heterogeneity analyses.

4.2. Endogeneity Test

Endogeneity is an important concern when examining the relationship between digital transformation and firm innovation. Firms with stronger innovation capabilities may be more willing and better able to implement digital transformation, leading to potential reverse causality. In addition, unobserved factors, such as managerial ability, organizational culture, industry-level technological opportunities, and firms’ long-term strategic orientation, may simultaneously affect both digital transformation and innovation output. To further alleviate these concerns, this study employs an instrumental-variable approach based on U.S. industry-level industrial robot penetration.

The instrumental variable is constructed from U.S. industry-level industrial robot data and matched to Chinese listed firms according to industry classification. The rationale for using this variable is that industrial robot penetration in the United States reflects frontier automation, intelligent manufacturing, and digital-technology diffusion at the global industry level. As the United States is one of the major economies leading global technological change, robot adoption in U.S. industries can capture external technological trends and automation pressure faced by comparable industries worldwide. When robot penetration increases in a corresponding U.S. industry, Chinese firms in similar industries are more likely to be exposed to international technology diffusion, supply-chain upgrading pressure, competitive imitation, and digital transformation incentives. Therefore, U.S. industry-level industrial robot penetration is expected to be positively associated with the digital transformation of Chinese firms, satisfying the relevance condition of the instrumental variable.

Regarding the exclusion restriction, the instrument is based on industry-level robot penetration in the United States rather than on the digital transformation behavior or innovation activities of individual Chinese firms. Therefore, it is unlikely to be directly determined by the innovation output of Chinese listed companies. Moreover, the instrument varies at the external industry level, which helps reduce concerns that it is driven by firm-specific unobserved characteristics, such as managerial ability, internal organizational culture, or persistent innovation orientation. Since the regressions include firm fixed effects and year fixed effects, time-invariant firm heterogeneity and common macroeconomic shocks are further controlled for. The inclusion of firm-level control variables also helps mitigate the influence of observable differences in firm size, profitability, leverage, growth, ownership structure, and governance characteristics.

In conceptual terms, the identifying logic is that U.S. industry-level industrial robot penetration affects the digital transformation of Chinese firms through external technology diffusion and industry-level digitalization pressure. However, it is less likely to directly affect the innovation output of a particular Chinese listed firm except through its influence on the firm’s digital transformation. This setting helps alleviate concerns about reverse causality and omitted firm-level determinants of innovation. Nevertheless, as industry-level global technology trends may also be related to broader innovation opportunities, the IV results should still be interpreted cautiously as evidence that helps reduce endogeneity concerns rather than as definitive causal proof.

Table 4. Instrumental variables test.

Variables	(1)	(2)
	First Stage	Second Stage
	DIGI	Innov
IV	0.024 **
	(0.011)
DIGI		0.423 **
		(0.212)
Size	2.076 ***	−0.339
	(0.261)	(0.456)
Roa	−3.822 *	1.480
	(2.277)	(1.296)
Age	−2.952 *	1.293
	(1.658)	(0.971)
Lev	−0.528	−0.013
	(0.973)	(0.440)
Tang	−1.322	0.004
	(1.843)	(0.851)
TobinQ	0.133	−0.067
	(0.115)	(0.059)
Growth	−0.477	0.171
	(0.312)	(0.169)
CapInt	−0.044	−0.010
	(0.121)	(0.053)
Top1	−0.015	0.008
	(0.016)	(0.008)
Board	0.908 **	−0.289
	(0.457)	(0.278)
ManHold	−2.037	1.292 *
	(1.315)	(0.721)
DUAL	0.462 *	−0.125
	(0.258)	(0.150)
AUDIT	0.642	−0.253
	(0.668)	(0.324)
InsHold	−0.051 ***	0.025 **
	(0.011)	(0.012)
Firm FE	Yes	Yes
Year FE	Yes	Yes
Observations	7969	7969
R-squared	0.632	0.594

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

reports the instrumental-variable estimation results. Column (1) presents the first-stage regression results. The coefficient of the instrumental variable is 0.024 and statistically significant at the 5% level, indicating that U.S. industry-level industrial robot penetration is positively associated with firm-level digital transformation in China. This result supports the relevance condition of the instrument. Column (2) reports the second-stage regression results. The estimated coefficient of DIGI is 0.423 and statistically significant at the 5% level, suggesting that digital transformation remains positively associated with firm innovation output after using the external industry-level instrument.

4.3. Robustness Test

The robustness tests are designed to examine whether the baseline findings are sensitive to alternative specifications, variable measurements, sample restrictions, and external shocks. Each test addresses a specific empirical concern. Alternative model specifications are used to examine whether the results are driven by omitted regional or industry-level factors. Alternative dependent variables are used to assess whether the findings depend on the measurement of innovation output. Additional regional controls are included to account for macroeconomic and financial conditions at the city level. Excluding major event years helps reduce the influence of large external shocks. Excluding municipalities addresses the concern that firms located in administratively special and economically advanced cities may disproportionately influence the results. Finally, the balanced-panel test examines whether the findings are affected by sample entry and exit over time.

Because some robustness specifications include different combinations of fixed effects and controls, coefficient magnitudes are not expected to be directly identical across all columns. Therefore, the robustness analysis focuses primarily on whether the sign, statistical significance, and substantive interpretation of the DIGI coefficient remain stable across alternative specifications.

4.3.1. Alternative Model Specifications

This test addresses the concern that the baseline results may be driven by unobserved city-level or industry-level characteristics. By adding city and industry fixed effects, we examine whether the digital transformation–innovation association remains stable after accounting for regional and sectoral heterogeneity

To evaluate robustness, we vary the specification and re-estimate the baseline models accordingly. In particular, we sequentially absorb city and industry effects to capture unobserved differences across locations and sectors. By controlling for regional and industry influences, we mitigate concerns that the results merely reflect regional development differences or sectoral traits rather than digital transformation itself.

The estimates reported in

Table 5. Alternative Model Specifications.

Variables	(1)	(2)	(3)
	Innov	Innov	Innov
DIGI	0.003 ***	0.003 ***	0.007 ***
	(3.60)	(3.82)	(9.66)
Size	0.448 ***	0.460 ***	0.662 ***
	(23.10)	(23.09)	(62.49)
Roa	−0.037	−0.023	0.473 **
	(−0.21)	(−0.13)	(2.33)
Age	0.180	0.149	−0.067 **
	(1.64)	(1.35)	(−1.99)
Lev	−0.248 ***	−0.251 ***	−0.196 ***
	(−3.34)	(−3.35)	(−3.10)
Tang	−0.576 ***	−0.686 ***	0.156
	(−4.24)	(−5.01)	(1.35)
TobinQ	0.010	0.010	0.055 ***
	(1.13)	(1.20)	(6.06)
Growth	−0.043 *	−0.054 **	−0.091 ***
	(−1.92)	(−2.45)	(−3.17)
CapInt	−0.027 ***	−0.028 ***	−0.061 ***
	(−3.73)	(−3.83)	(−10.43)
Top1	−0.001	−0.000	−0.002 **
	(−0.50)	(−0.18)	(−2.06)
Board	0.054	0.057	0.113 ***
	(1.47)	(1.57)	(3.00)
ManHold	0.521 ***	0.483 ***	0.048
	(5.21)	(4.82)	(0.67)
DUAL	0.008	0.011	−0.037 *
	(0.38)	(0.54)	(−1.84)
AUDIT	−0.118 **	−0.134 **	0.020
	(−2.23)	(−2.53)	(0.57)
InsHold	0.003 ***	0.003 ***	0.001 **
	(3.31)	(3.58)	(2.31)
Constant	−8.161 ***	−8.250 ***	−13.004 ***
	(−14.73)	(−14.64)	(−46.90)
Firm FE	Yes	Yes	No
Year FE	Yes	Yes	Yes
City FE	Yes	Yes	Yes
Industry FE	No	Yes	Yes
Observations	20,254	20,254	20,254
R-squared	0.79	0.79	0.55

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

show that the coefficient on DIGI remains positive with strong statistical support (p < 0.01) under all specifications, ranging from 0.003 to 0.007. The estimates are broadly comparable to the benchmark regression, supporting the stability of the positive digital transformation effect on innovation output under different specifications.

4.3.2. Alternative Dependent Variables

This test addresses the concern that the findings depend on the use of invention patent applications as the innovation measure. We therefore use invention patent grants as an alternative dependent variable, which captures more realized and quality-screened innovation outcomes.

To test whether the results depend on how innovation output is measured, we adopt an alternative proxy. Beyond invention patent applications, we also use the log-transformed count of invention patent grants (Innov2) as the dependent variable. As opposed to patent applications, granted patents involve stricter examination and higher thresholds, thereby better reflecting the quality and practical implementation of innovation outcomes. If the results remain consistent under this alternative specification, concerns about inflated application counts or strategic patent filings are alleviated, thus reinforcing the credibility of our conclusions.

Regardless of the specification,

Table 6. Alternative Dependent Variables.

Variables	(1)	(2)	(3)	(4)
	Innov2	Innov2	Innov2	Innov2
DIGI	0.005 ***	0.005 ***	0.006 ***	0.009 ***
	(7.47)	(7.28)	(8.06)	(12.92)
Size	0.405 ***	0.395 ***	0.418 ***	0.563 ***
	(23.64)	(22.77)	(23.49)	(60.33)
Roa	−0.699 ***	−0.685 ***	−0.689 ***	−0.416 **
	(−4.41)	(−4.32)	(−4.36)	(−2.33)
Age	0.326 ***	0.362 ***	0.303 ***	−0.031
	(3.34)	(3.70)	(3.07)	(−1.03)
Lev	−0.263 ***	−0.260 ***	−0.267 ***	−0.344 ***
	(−3.96)	(−3.91)	(−3.98)	(−6.17)
Tang	0.010	−0.071	−0.123	0.280 ***
	(0.09)	(−0.58)	(−1.01)	(2.74)
TobinQ	0.002	0.003	0.005	0.053 ***
	(0.28)	(0.38)	(0.61)	(6.57)
Growth	−0.058 ***	−0.057 ***	−0.069 ***	−0.103 ***
	(−2.94)	(−2.90)	(−3.50)	(−4.07)
CapInt	−0.034 ***	−0.036 ***	−0.039 ***	−0.053 ***
	(−5.29)	(−5.52)	(−5.99)	(−10.31)
Top1	−0.001	−0.001	−0.000	−0.001
	(−0.92)	(−0.57)	(−0.20)	(−1.11)
Board	0.035	0.033	0.042	0.093 ***
	(1.06)	(1.02)	(1.30)	(2.79)
ManHold	0.237 ***	0.304 ***	0.282 ***	−0.157 **
	(2.66)	(3.39)	(3.15)	(−2.46)
DUAL	−0.013	−0.021	−0.012	−0.034 *
	(−0.71)	(−1.13)	(−0.62)	(−1.92)
AUDIT	−0.130 ***	−0.138 ***	−0.162 ***	0.092 ***
	(−2.73)	(−2.92)	(−3.42)	(2.98)
InsHold	0.001 **	0.001	0.001 *	0.000
	(2.02)	(1.39)	(1.85)	(0.19)
Constant	−8.636 ***	−8.429 ***	−8.762 ***	−11.477 ***
	(−17.59)	(−17.00)	(−17.39)	(−46.98)
Firm FE	Yes	Yes	Yes	No
Year FE	Yes	Yes	Yes	Yes
City FE	No	Yes	Yes	Yes
Industry FE	No	No	Yes	Yes
Observations	20,254	20,254	20,254	20,254
R-squared	0.76	0.76	0.77	0.51

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

yields a positive and precisely estimated DIGI coefficient (p < 0.01). The coefficient remains of a similar size to the benchmark results, with estimates ranging from 0.005 to 0.009. Taken together, digital transformation boosts both “potential” innovation output (i.e., patent applications) and “realized” innovation outcomes (i.e., patent grants).

In other words, digital transformation not only stimulates firms to increase R&D investment and pursue more innovation attempts but also improves the quality and conversion efficiency of innovation activities, enabling a higher proportion of R&D projects to translate into granted patents. Innovation gains from digital transformation are two-dimensional, affecting not only quantity but also quality.

4.3.3. Adding Additional Control Variables

This test addresses the concern that regional macroeconomic and financial conditions may simultaneously affect digital transformation and innovation output. We therefore add city-level economic growth, industrial structure, per capita GDP, and financial loan variables.

To account for regional macro conditions, we enrich the baseline specification by including macroeconomic variables at the region level, which also serves as a robustness exercise. The added regional controls comprise city-level GDP growth (GDP), the GDP shares of the secondary (I2) and tertiary (I3) sectors, the log of per capita GDP (PerGDP), and the loan balance of financial institutions (Loan). The inclusion of these variables allows us to control for influences on innovation output arising from firm-level resources and governance, in addition to effects stemming from the broader economy and regional development.

Table 7. Adding Additional Control Variables.

Variables	(1)	(2)	(3)	(4)
	Innov	Innov	Innov	Innov
DIGI	0.003 ***	0.002 ***	0.002 ***	0.007 ***
	(3.20)	(3.07)	(3.02)	(9.21)
Size	0.481 ***	0.488 ***	0.492 ***	0.674 ***
	(21.96)	(22.06)	(21.64)	(58.27)
Roa	−0.215	−0.194	−0.181	0.241
	(−1.07)	(−0.97)	(−0.91)	(1.06)
Age	0.191	0.162	0.146	−0.043
	(1.60)	(1.36)	(1.22)	(−1.16)
Lev	−0.375 ***	−0.357 ***	−0.323 ***	−0.228 ***
	(−4.47)	(−4.27)	(−3.84)	(−3.25)
Tang	−0.442 ***	−0.443 ***	−0.564 ***	0.217 *
	(−2.88)	(−2.88)	(−3.64)	(1.69)
TobinQ	0.007	0.009	0.010	0.053 ***
	(0.69)	(0.95)	(1.02)	(5.38)
Growth	−0.048 *	−0.046 *	−0.055 **	−0.092 ***
	(−1.94)	(−1.85)	(−2.21)	(−2.84)
CapInt	−0.036 ***	−0.036 ***	−0.036 ***	−0.063 ***
	(−4.61)	(−4.55)	(−4.52)	(−9.94)
Top1	−0.002 *	−0.002	−0.001	−0.002 ***
	(−1.86)	(−1.44)	(−0.89)	(−2.86)
Board	0.042	0.048	0.047	0.069 *
	(1.02)	(1.17)	(1.14)	(1.65)
ManHold	0.529 ***	0.480 ***	0.433 ***	0.094
	(4.58)	(4.14)	(3.72)	(1.15)
DUAL	−0.010	−0.010	−0.009	−0.036
	(−0.42)	(−0.42)	(−0.38)	(−1.60)
AUDIT	−0.154 ***	−0.162 ***	−0.166 ***	0.006
	(−2.66)	(−2.80)	(−2.87)	(0.17)
InsHold	0.004 ***	0.004 ***	0.004 ***	0.002 ***
	(4.66)	(4.30)	(4.28)	(3.00)
GDP	−0.229	−0.177	−0.136	−0.315
	(−1.26)	(−0.97)	(−0.75)	(−1.25)
I2	0.008	0.019 **	0.015	0.009
	(0.95)	(1.98)	(1.63)	(0.74)
I3	0.015 *	0.027 ***	0.022 **	0.021
	(1.95)	(2.75)	(2.26)	(1.63)
PerGDP	−0.001	−0.001	−0.000	0.005
	(−0.09)	(−0.11)	(−0.04)	(0.34)
Loan	−0.000 *	−0.000	−0.000	−0.000
	(−1.69)	(−0.50)	(−0.72)	(−0.56)
Constant	−9.950 ***	−11.201 ***	−10.774 ***	−14.804 ***
	(−10.62)	(−10.34)	(−9.81)	(−11.77)
Firm FE	Yes	Yes	Yes	No
Year FE	Yes	Yes	Yes	Yes
City FE	No	Yes	Yes	Yes
Industry FE	No	No	Yes	Yes
Observations	16,259	16,258	16,258	16,295
R-squared	0.80	0.80	0.80	0.55

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

indicates that the DIGI coefficient is consistently positive and highly significant (p < 0.01) across specifications, with estimates between 0.002 and 0.007. After controlling for regional economic conditions and industrial composition, the estimated link between digital transformation and innovation output remains positive, indicating that these factors do not drive the findings.

4.3.4. Excluding Major Events

This test addresses the concern that major external shocks may confound the estimated relationship between digital transformation and innovation. We therefore exclude years affected by representative macroeconomic or global disruptions.

To address the concern that external shocks may confound the relationship between digital transformation and innovation output, we re-estimate the baseline model after excluding observations from three representative years characterized by major external disruptions: 2015, when China’s stock market experienced severe turbulence that undermined firms’ financing environment and investor confidence; 2018, when the escalation of the U.S.–China trade conflict created uncertainty for export-oriented and high-tech enterprises; and 2020, when the COVID-19 outbreak interrupted global supply chains and precipitated a sharp economic downturn. These events could systematically affect both digital transformation strategies and innovation activities, thereby biasing the estimated relationship.

As shown in

Table 8. Excluding Major Events.

Variables	(1)	(2)	(3)	(4)
	Innov	Innov	Innov	Innov
DIGI	0.004 ***	0.003 ***	0.004 ***	0.007 ***
	(4.41)	(4.02)	(4.51)	(9.09)
Size	0.446 ***	0.442 ***	0.453 ***	0.643 ***
	(20.59)	(20.19)	(20.15)	(53.25)
Roa	−0.114	−0.088	−0.043	0.474 **
	(−0.53)	(−0.40)	(−0.20)	(1.99)
Age	0.221 *	0.197 *	0.164	−0.089 **
	(1.87)	(1.66)	(1.37)	(−2.33)
Lev	−0.278 ***	−0.286 ***	−0.286 ***	−0.196 ***
	(−3.28)	(−3.37)	(−3.35)	(−2.73)
Tang	−0.472 ***	−0.589 ***	−0.707 ***	0.104
	(−2.96)	(−3.68)	(−4.38)	(0.77)
TobinQ	−0.003	0.001	0.002	0.051 ***
	(−0.25)	(0.12)	(0.16)	(4.56)
Growth	−0.055 **	−0.060 **	−0.071 ***	−0.088 ***
	(−2.09)	(−2.32)	(−2.74)	(−2.70)
CapInt	−0.030 ***	−0.034 ***	−0.034 ***	−0.057 ***
	(−3.69)	(−4.04)	(−4.06)	(−8.51)
Top1	−0.001	−0.001	−0.000	−0.002 *
	(−1.00)	(−0.57)	(−0.28)	(−1.86)
Board	0.052	0.055	0.060	0.087 **
	(1.22)	(1.29)	(1.41)	(2.04)
ManHold	0.584 ***	0.579 ***	0.521 ***	0.015
	(5.13)	(5.03)	(4.53)	(0.19)
DUAL	0.021	0.014	0.018	−0.037
	(0.84)	(0.58)	(0.72)	(−1.63)
AUDIT	−0.093	−0.096	−0.112 *	0.039
	(−1.53)	(−1.57)	(−1.84)	(0.97)
InsHold	0.003 ***	0.002 ***	0.003 ***	0.002 **
	(3.47)	(2.70)	(2.90)	(2.28)
Constant	−8.336 ***	−8.065 ***	−8.124 ***	−12.466 ***
	(−13.59)	(−13.01)	(−12.91)	(−39.32)
Firm FE	Yes	Yes	Yes	No
Year FE	Yes	Yes	Yes	Yes
City FE	No	Yes	Yes	Yes
Industry FE	No	No	Yes	Yes
Observations	15,880	15,878	15,878	15,879
R-squared	0.78	0.78	0.79	0.54

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

, the DIGI coefficient remains positive and highly significant (p < 0.01) in every specification, with estimates spanning 0.003 to 0.007. After excluding years affected by major shocks, the estimated digital transformation effect stays positive and broadly consistent with the benchmark regression.

4.3.5. Excluding Four Municipalities

To assess robustness, we re-estimate the models after excluding firms located in Beijing, Shanghai, Tianjin, and Chongqing, the four directly administered municipalities. These cities occupy a unique position in China’s economic landscape, characterized by advanced levels of digital economy development, strong innovation capacity, and pronounced agglomeration effects. In addition, they benefit disproportionately from policy support, capital availability, and market advantages compared to other regions. Including these municipalities without distinction may bias the estimates due to their extreme values and exceptional development trajectories.

Table 9. Excluding Four Municipalities.

Variables	(1)	(2)	(3)	(4)
	Innov	Innov	Innov	Innov
DIGI	0.002 ***	0.002 **	0.002 **	0.008 ***
	(2.64)	(2.31)	(2.36)	(8.75)
Size	0.459 ***	0.449 ***	0.467 ***	0.608 ***
	(21.46)	(20.61)	(20.81)	(50.22)
Roa	0.123	0.127	0.124	1.015 ***
	(0.62)	(0.65)	(0.63)	(4.60)
Age	−0.018	−0.025	−0.030	0.072 *
	(−0.15)	(−0.20)	(−0.24)	(1.91)
Lev	−0.183 **	−0.206 **	−0.218 ***	−0.068
	(−2.22)	(−2.49)	(−2.60)	(−0.96)
Tang	−0.394 ***	−0.519 ***	−0.652 ***	0.170
	(−2.64)	(−3.46)	(−4.29)	(1.33)
TobinQ	0.008	0.010	0.010	0.048 ***
	(0.82)	(0.98)	(1.03)	(4.82)
Growth	−0.041	−0.040	−0.059 **	−0.083 ***
	(−1.63)	(−1.63)	(−2.36)	(−2.67)
CapInt	−0.017 **	−0.019 **	−0.021 **	−0.045 ***
	(−2.05)	(−2.25)	(−2.46)	(−6.67)
Top1	−0.001	−0.001	−0.001	−0.004 ***
	(−1.07)	(−1.04)	(−0.56)	(−4.83)
Board	0.049	0.047	0.052	0.133 ***
	(1.20)	(1.14)	(1.26)	(3.17)
ManHold	0.502 ***	0.525 ***	0.485 ***	0.190 **
	(4.61)	(4.78)	(4.41)	(2.40)
DUAL	0.018	0.010	0.013	−0.008
	(0.78)	(0.42)	(0.57)	(−0.35)
AUDIT	−0.138 **	−0.134 **	−0.159 **	−0.152 ***
	(−2.10)	(−2.03)	(−2.42)	(−3.42)
InsHold	0.002 ***	0.002 **	0.003 ***	0.003 ***
	(2.81)	(2.51)	(2.96)	(4.07)
Constant	−8.004 ***	−7.596 ***	−7.911 ***	−12.324 ***
	(−13.05)	(−12.22)	(−12.46)	(−40.02)
Firm FE	Yes	Yes	Yes	No
Year FE	Yes	Yes	Yes	Yes
City FE	No	Yes	Yes	Yes
Industry FE	No	No	Yes	Yes
Observations	16,450	16,450	16,450	16,453
R-squared	0.77	0.77	0.78	0.54

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

indicates that the DIGI coefficient is consistently positive and highly significant (p < 0.01) across specifications, with estimates between 0.002 and 0.008. Both the coefficient signs and their statistical significance align with the baseline estimates, s which alleviates concerns that the estimated effect is concentrated in only a few highly developed municipalities. Rather, the effect is broadly applicable to firms across different regions.

4.3.6. Balanced Panel Estimation

As an additional robustness check, we re-run the regressions on a balanced panel that includes only firms observed in every year of the sample period. Compared with an unbalanced panel, the use of a balanced panel reduces biases arising from missing data and improves the comparability of results across firms and over time. However, this approach also entails a substantial reduction in sample size, which may lower statistical precision. If the results remain consistent under the balanced sample, the robustness of our conclusions will be further reinforced.

As reported in

Table 10. Balanced Panel Estimation.

Variables	(1)	(2)	(3)	(4)
	Innov	Innov	Innov	Innov
DIGI	0.003 **	0.003 **	0.003 **	0.007 ***
	(2.26)	(2.28)	(2.18)	(4.38)
Size	0.493 ***	0.466 ***	0.440 ***	0.668 ***
	(12.55)	(11.70)	(10.68)	(28.46)
Roa	−0.219	−0.181	−0.132	0.227
	(−0.58)	(−0.48)	(−0.35)	(0.53)
Age	−0.352	−0.323	−0.202	−0.495 ***
	(−1.45)	(−1.33)	(−0.81)	(−5.14)
Lev	0.043	−0.006	0.007	−0.067
	(0.28)	(−0.04)	(0.04)	(−0.50)
Tang	−0.821 **	−0.849 ***	−1.007 ***	−1.380 ***
	(−2.57)	(−2.67)	(−3.06)	(−4.44)
TobinQ	0.019	0.015	0.013	0.089 ***
	(1.08)	(0.85)	(0.70)	(4.65)
Growth	−0.053	−0.041	−0.057	−0.024
	(−1.12)	(−0.87)	(−1.19)	(−0.44)
CapInt	−0.059 ***	−0.056 ***	−0.058 ***	−0.025 *
	(−3.63)	(−3.48)	(−3.47)	(−1.87)
Top1	−0.005 **	−0.006 **	−0.003	−0.004 **
	(−2.02)	(−2.47)	(−1.46)	(−2.32)
Board	−0.002	−0.002	0.040	0.178 **
	(−0.03)	(−0.02)	(0.54)	(2.30)
ManHold	0.913 ***	0.968 ***	0.770 **	0.344
	(2.64)	(2.80)	(2.10)	(1.44)
DUAL	0.043	0.043	0.046	−0.052
	(0.97)	(0.97)	(1.04)	(−1.24)
AUDIT	0.012	−0.006	−0.008	0.073
	(0.12)	(−0.06)	(−0.08)	(1.02)
InsHold	0.002	0.002	0.003 *	0.003 **
	(1.26)	(1.41)	(1.87)	(2.01)
Constant	−7.030 ***	−6.421 ***	−6.256 ***	−10.668 ***
	(−5.85)	(−5.32)	(−5.05)	(−15.77)
Firm FE	Yes	Yes	Yes	No
Year FE	Yes	Yes	Yes	Yes
City FE	No	Yes	Yes	Yes
Industry FE	No	No	Yes	Yes
Observations	4635	4633	4632	4632
R-squared	0.81	0.81	0.81	0.70

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

, DIGI is consistently positive and statistically significant, and its estimated coefficient ranges from 0.003 to 0.007. Although the number of observations decreases markedly from over 20,000 firm-year samples to approximately 4600, the results are qualitatively unchanged compared with the baseline findings.

5. Further Analysis

5.1. Mechanism-Consistent Evidence

This section provides mechanism-consistent evidence for the proposed dual-path framework. It should be noted that the purpose of this analysis is not to establish a complete causal mediation chain. Rather, we examine whether digital transformation is systematically associated with a set of R&D-related and governance-related variables that are theoretically linked to innovation. Therefore, the results should be interpreted as evidence consistent with the proposed mechanisms, rather than as definitive causal mediation evidence.

5.1.1. R&D Investment Mechanism

R&D investment represents the most direct and critical pathway linking DT to innovation output. As suggested by the theoretical derivation, DT can improve resource allocation, reduce information frictions, and ease financing constraints, thereby enhancing both the scale and efficiency of R&D activities. In

Table 11. R&D Investment Mechanism.

Variables	(1)	(2)	(3)	(4)	(5)
	RDint	RDcap	RDgrowth	RDpc	RDexp
DIGI	0.019 ***	0.030 ***	0.001 *	0.004 *	0.004 *
	(9.02)	(2.62)	(1.81)	(1.91)	(1.91)
Size	0.191 ***	2.377 ***	0.136 ***	1.461 ***	−0.408 ***
	(3.67)	(6.04)	(3.68)	(15.90)	(−4.72)
Roa	−6.569 ***	−8.176 ***	−0.064	1.743 **	−8.138 ***
	(−13.65)	(−3.11)	(−0.21)	(2.22)	(−16.83)
Age	−0.296	8.073 **	0.505 **	1.414 ***	2.203 **
	(−1.00)	(2.56)	(2.34)	(2.71)	(2.57)
Lev	−2.634 ***	−0.948	0.088	−3.476 ***	−2.534 ***
	(−13.07)	(−0.70)	(0.63)	(−9.95)	(−8.81)
Tang	−0.118	−4.733 **	0.128	1.199 *	−0.367
	(−0.32)	(−2.03)	(0.52)	(1.93)	(−0.68)
TobinQ	0.081 ***	0.030	0.025	0.074 *	0.010
	(3.46)	(0.23)	(1.64)	(1.93)	(0.36)
Growth	−0.282 ***	−0.185	0.785 ***	−0.225 **	−0.196 ***
	(−4.68)	(−0.52)	(18.87)	(−2.17)	(−2.60)
CapInt	0.418 ***	0.473 ***	0.018	−0.272 ***	0.651 ***
	(21.41)	(3.34)	(1.20)	(−7.13)	(22.59)
Top1	−0.012 ***	−0.029	−0.001	−0.007	0.005
	(−3.77)	(−1.28)	(−0.37)	(−1.31)	(1.10)
Board	0.167 *	0.070	0.078	0.206	−0.011
	(1.68)	(0.12)	(1.21)	(1.26)	(−0.10)
ManHold	0.108	−0.051	−0.099	1.006 **	0.693 **
	(0.40)	(−0.03)	(−0.58)	(2.30)	(2.01)
DUAL	−0.066	0.383	0.061	0.218 **	0.046
	(−1.14)	(1.13)	(1.64)	(2.33)	(0.69)
AUDIT	0.129	−2.581 ***	−0.026	−0.182	0.249
	(0.89)	(−2.68)	(−0.25)	(−0.72)	(1.33)
InsHold	−0.000	0.000	0.000	−0.015 ***	0.005 *
	(−0.12)	(0.02)	(0.23)	(−4.09)	(1.67)
Constant	0.558	−66.329 ***	−4.807 ***	−31.292 ***	6.389 **
	(0.37)	(−5.08)	(−4.50)	(−11.89)	(1.97)
Firm FE	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
Observations	20,254	11,156	16,121	17,011	10,113
R-squared	0.80	0.75	0.17	0.75	0.90

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

, we re-estimate the regressions by redefining R&D investment with several proxies, including R&D intensity (RDint), R&D capitalization ratio (RDcap), R&D investment growth rate (RDgrowth), per capita R&D input (RDpc), and R&D expenditure ratio (RDexp).

The estimated DIGI coefficient is positive across all models and reaches statistical significance in most specifications. For R&D intensity, the coefficient equals 0.019 (p < 0.01), indicating a higher R&D share scaled by total assets or operating revenue as digital transformation deepens. For the R&D capitalization ratio, the estimated coefficient is 0.030 (p < 0.01), suggesting that deeper digital transformation is accompanied by a larger capitalized share of R&D expenditures. This reflects a shift toward long-term, strategic R&D projects, enhancing the sustainability and strategic value of innovation outcomes.

Similarly, both R&D investment growth and per capita R&D input rise significantly with DT. These results imply that digitalized firms exhibit stronger dynamic growth in R&D activities and allocate human capital more efficiently, resulting in higher innovation productivity per employee. Finally, revenue-based R&D spending increases with digital transformation, as reflected in the significantly positive R&D expenditure ratio. Taken together, these results provide evidence consistent with Hypothesis 2, suggesting that R&D investment and R&D efficiency constitute a plausible channel through which digital transformation is associated with innovation output.

5.1.2. Governance and Information Mechanism

In addition to R&D investment, the governance and information environment represents another important mechanism-consistent channel. This study uses internal control quality, disclosure quality, and audit quality as related indicators of this broader environment. These indicators should not be interpreted as strictly independent sub-dimensions. In particular, because internal control evaluation may include information communication elements, internal control and disclosure quality may partially overlap. Therefore, the following analysis is intended to provide complementary evidence on the governance and information environment rather than to establish separate causal mechanisms for each indicator.

In addition to R&D investment, improvements in corporate governance and the information environment represent another crucial channel through which digital transformation (DT) promotes innovation. Innovation activities are inherently high risk and highly uncertain, which often exacerbate agency conflicts and investor concerns. By strengthening governance transparency and supervisory efficiency, DT can create a more favorable environment for innovation.

Table 12. Governance and Information Mechanism.

Variables	(1)	(2)	(3)
	IC	KV	AQ
DIGI	0.001 *	−0.037 **	0.006 ***
	(1.71)	(−2.47)	(2.97)
Size	0.094 ***	8.050 ***	0.012 ***
	(5.46)	(20.89)	(5.11)
Roa	4.564 ***	35.737 ***	0.545 ***
	(28.82)	(10.03)	(26.34)
Age	−0.111	3.751 *	0.036 ***
	(−1.13)	(1.72)	(2.80)
Lev	0.017	−8.380 ***	−0.068 ***
	(0.26)	(−5.63)	(−7.77)
Tang	0.104	7.195 ***	−0.018
	(0.86)	(2.65)	(−1.16)
TobinQ	0.004	3.211 ***	−0.002 **
	(0.58)	(18.47)	(−2.41)
Growth	0.371 ***	0.622	0.008 ***
	(18.69)	(1.39)	(2.97)
CapInt	−0.025 ***	−0.540 ***	0.002 *
	(−3.96)	(−3.70)	(1.93)
Top1	−0.001	−0.088 ***	0.000
	(−0.86)	(−3.88)	(1.35)
Board	−0.095 ***	−0.690	−0.006
	(−2.92)	(−0.94)	(−1.50)
ManHold	0.184 **	15.794 ***	−0.005
	(2.07)	(7.92)	(−0.42)
DUAL	−0.013	−0.011	−0.006 **
	(−0.66)	(−0.03)	(−2.32)
AUDIT	0.082 *	2.624 **	0.009
	(1.73)	(2.46)	(1.37)
InsHold	0.003 ***	0.146 ***	−0.000
	(4.37)	(9.22)	(−0.73)
Constant	4.601 ***	−154.671 ***	−0.358 ***
	(9.38)	(−14.03)	(−5.56)
Firm FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Observations	20,244	20,133	17,383
R-squared	0.42	0.46	0.33

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

reports regression estimates for three dependent variables: internal control quality (IC), disclosure quality (KV), and audit quality (AQ).

First, we rely on the CSMAR Dibo Internal Control Index as an indicator of firms’ internal control quality. DIGI is positive and marginally significant (p < 0.10), indicating that stronger digital transformation is associated with improved internal control systems. Enhanced authenticity and traceability of internal information reduce managerial opportunism and foster a more reliable governance environment, which in turn provides stronger institutional support for innovation activities.

Second, disclosure quality is proxied by the KV index (Kim and Verrecchia, 2001 [36]), where smaller values indicate lower information asymmetry. DIGI is estimated at −0.037 and is statistically significant (p < 0.05), indicating that digital transformation materially improves disclosure quality and transparency. Greater transparency strengthens investor trust in firms’ long-term strategies, thereby increasing the availability of external financing and reducing the uncertainty surrounding R&D investment.

Finally, following Gul et al. (2013 [37]), audit quality is defined as the negative of the absolute deviation between the realized audit opinion and the predicted likelihood of receiving a standard unqualified opinion. DIGI is estimated at 0.006 and is highly significant (p < 0.01), indicating that stronger digital transformation is associated with higher audit quality. Improved audit quality not only bolsters investor confidence in financial reporting and innovation strategies but also pressures firms to adopt stricter internal governance practices, indirectly supporting innovation activities.

Overall, the results provide evidence consistent with Hypothesis 3. They suggest that digital transformation is associated with a more supportive governance and information environment, which may help explain the positive relationship between digital transformation and innovation output.

5.2. Heterogeneity Analysis

The heterogeneity analysis is intended to explore whether the association between digital transformation and innovation differs across firms with different resource endowments and institutional conditions. These subgroup results should be interpreted as conditional patterns rather than definitive causal differences, because heterogeneity analyses may be sensitive to sample partition rules and unobserved group-specific factors. Nevertheless, they provide useful evidence on the boundary conditions under which digital transformation is more strongly associated with innovation output. Building on the evidence regarding DT’s aggregate impact on innovation output and its mechanisms, we further conduct heterogeneity analyses to examine whether the effect varies across different firm characteristics and external environments.

Table 13. Heterogeneity analysis.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Innov	Innov	Innov	Innov	Innov	Innov	Innov	Innov
	High FC	Low FC	Large Size	Small Size	West	East	SOEs	Non-SOEs
DIGI	0.003	0.003 *	0.003 **	0.001	0.003	0.003 ***	0.006 ***	0.002 **
	(1.46)	(1.70)	(2.42)	(1.23)	(1.63)	(3.77)	(3.95)	(2.12)
Size	0.419 ***	0.393 ***	0.479 ***	0.473 ***	0.434 ***	0.455 ***	0.475 ***	0.458 ***
	(6.16)	(6.25)	(12.61)	(13.85)	(12.90)	(19.10)	(16.23)	(17.24)
Roa	0.343	−0.407	−0.065	0.340	0.418	−0.192	−0.175	0.030
	(1.03)	(−1.39)	(−0.20)	(1.62)	(1.24)	(−0.92)	(−0.57)	(0.14)
Age	−0.695 **	−0.837	−0.076	0.468 ***	0.254	0.176	0.529 ***	−0.078
	(−2.21)	(−0.79)	(−0.42)	(2.85)	(1.16)	(1.41)	(3.36)	(−0.51)
Lev	−0.204	−0.292 *	−0.180	−0.126	0.115	−0.379 ***	−0.085	−0.223 **
	(−1.15)	(−1.65)	(−1.22)	(−1.37)	(0.87)	(−4.22)	(−0.74)	(−2.23)
Tang	−0.266	−0.852 **	−2.125 ***	−0.401 **	−1.263 ***	−0.163	−0.690 ***	−0.209
	(−0.91)	(−2.47)	(−7.42)	(−2.33)	(−4.98)	(−1.02)	(−2.80)	(−1.26)
TobinQ	−0.001	0.023	−0.021	−0.005	−0.018	0.017 *	0.005	0.015
	(−0.08)	(1.46)	(−1.15)	(−0.52)	(−1.11)	(1.70)	(0.32)	(1.36)
Growth	−0.015	−0.049	−0.018	−0.064 **	−0.070 *	−0.038	−0.037	−0.061 **
	(−0.38)	(−1.34)	(−0.53)	(−2.18)	(−1.82)	(−1.41)	(−1.13)	(−2.05)
CapInt	0.007	−0.047 ***	−0.011	−0.020 **	−0.038 ***	−0.020 **	−0.051 ***	−0.001
	(0.29)	(−3.34)	(−0.89)	(−2.10)	(−3.11)	(−2.27)	(−4.91)	(−0.14)
Top1	−0.002	0.003	−0.002	−0.001	0.003	−0.004 ***	−0.003 *	−0.002
	(−0.54)	(0.78)	(−1.04)	(−0.75)	(1.45)	(−2.62)	(−1.95)	(−0.99)
Board	0.018	0.109 **	0.084	0.028	0.192 ***	−0.008	0.030	0.059
	(0.27)	(1.97)	(1.48)	(0.60)	(2.96)	(−0.18)	(0.57)	(1.16)
ManHold	0.394 *	0.481 *	1.010 ***	0.310 ***	0.629 ***	0.449 ***	0.430	0.155
	(1.70)	(1.83)	(3.83)	(2.70)	(2.74)	(4.03)	(0.59)	(1.39)
DUAL	−0.046	0.042	0.091 **	−0.013	−0.040	0.021	−0.002	0.019
	(−0.99)	(1.03)	(2.43)	(−0.49)	(−0.99)	(0.85)	(−0.07)	(0.71)
AUDIT	−0.046	−0.161	−0.041	0.246 **	0.054	−0.180 ***	−0.152 **	−0.018
	(−0.27)	(−0.88)	(−0.60)	(2.36)	(0.52)	(−2.91)	(−2.18)	(−0.21)
InsHold	0.006 ***	−0.000	0.005 ***	0.000	0.004 ***	0.003 ***	0.005 ***	0.001
	(2.96)	(−0.05)	(3.63)	(0.24)	(2.85)	(2.86)	(4.16)	(1.27)
Constant	−5.470 ***	−3.517	−6.864 ***	−9.593 ***	−8.160 ***	−8.353 ***	−9.850 ***	−7.827 ***
	(−3.10)	(−0.98)	(−6.32)	(−10.98)	(−8.09)	(−12.54)	(−11.80)	(−10.27)
Firm FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	10,014	10,240	8942	11,312	6326	13,928	8444	11,810
R-squared	0.84	0.81	0.83	0.75	0.77	0.79	0.84	0.75

Notes: Statistical significance is indicated by *** (p < 0.01), ** (p < 0.05), and * (p < 0.10). T-statistics are shown in parentheses below each coefficient estimate.

reports regression results by grouping firms according to financing constraints, firm size, regional location, and ownership structure.

5.2.1. Financing Constraints

Firms with lower financing constraints are more likely to have sufficient resources to convert digital transformation into innovation-related activities. Digital transformation often requires complementary investments in digital infrastructure, human capital, organizational restructuring, and R&D projects. When financing constraints are severe, firms may adopt digital technologies rhetorically or partially but lack the resources required to transform digital capabilities into sustained innovation output. Therefore, the digital transformation–innovation association is expected to be more pronounced among firms with lower financing constraints.

With respect to financing constraints, digital transformation significantly enhances innovation output in the low-constraint group, while the estimate becomes statistically indistinguishable from zero in the high-constraint group. This suggests that firms with better access to capital are more capable of leveraging digital tools to allocate resources effectively and support R&D, thereby releasing stronger innovation potential. In contrast, firms under tight financial constraints remain limited in their ability to translate digital transformation into innovation due to funding bottlenecks.

5.2.2. Firm Size

Firm size may also condition the innovation implications of digital transformation. Larger firms usually have stronger resource bases, more diversified knowledge assets, more stable R&D teams, and better-developed data systems. These conditions enable them to absorb digital technologies more effectively and integrate them into R&D and governance processes. By contrast, smaller firms may face greater difficulties in building complementary capabilities. Therefore, the positive association between digital transformation and innovation may be more pronounced among larger firms.

After dividing firms by size, a strong positive innovation effect emerges for large firms, while the corresponding estimate for small firms is comparatively modest. This finding reflects the fact that large firms are better positioned to exploit economies of scale and synergy effects during digital transformation, allowing them to translate digital capabilities into innovation outcomes more efficiently. By contrast, small firms, constrained by limited resources, may find the innovation-enhancing effect of digital transformation attenuated.

5.2.3. Regional Location

Regional institutional and infrastructural conditions may further shape the role of digital transformation. Firms located in eastern regions generally benefit from better digital infrastructure, stronger market-supporting institutions, more developed financial systems, and richer innovation ecosystems. These external conditions may help firms translate digital transformation into innovation output more effectively. Therefore, the digital transformation–innovation association is expected to be more pronounced among firms located in eastern regions.

Regional heterogeneity is evident: the digital transformation effect is statistically significant for firms in eastern China, but it is not significant for those in western regions. This discrepancy is consistent with regional disparities in digital infrastructure, R&D ecosystems, and policy support. Firms in the more developed eastern regions operate within stronger innovation ecosystems and possess greater digital readiness, which strengthens the innovation impact of digital transformation. Conversely, in western regions with weaker infrastructure and limited innovation resources, the marginal effect of DT is less evident.

5.2.4. Firm Ownership

Ownership structure may influence firms’ capacity and incentives to implement digital transformation. State-owned enterprises may have stronger access to policy support, financing resources, and long-term investment opportunities, which can facilitate the integration of digital transformation with innovation activities. However, this interpretation should be made cautiously, as ownership-related differences may also reflect broader institutional and resource conditions. Therefore, the stronger association observed among state-owned enterprises is interpreted as a conditional pattern rather than definitive causal evidence.

When firms are grouped by ownership, the innovation gains associated with digital transformation are markedly stronger among SOEs than among non-SOEs. This may be attributed to the greater financial support, policy guidance, and industrial positioning enjoyed by SOEs, which allow them to leverage digital technologies more effectively in fostering innovation. In contrast, while non-SOEs tend to be more flexible in market competition, their weaker institutional support and limited access to external resources reduce the extent to which digital transformation translates into innovation outcomes.

6. Discussion

6.1. Theoretical Implications

The findings of this study offer several theoretical implications for research on digital transformation and corporate innovation. First, this study extends the literature on the relationship between digital transformation and innovation by proposing a dual-path framework. Rather than treating digital transformation solely as a technological input or an efficiency-enhancing tool, this study conceptualizes it as a broader organizational transformation process that influences innovation through both R&D resource allocation and the governance environment. This perspective helps explain why digital transformation may promote innovation not only by increasing firms’ innovation inputs, but also by improving the institutional, informational, and managerial conditions under which innovation activities take place.

Second, this study contributes to the Resource-Based View by clarifying how digital resources can evolve into innovation-related strategic resources. Digital transformation does not automatically translate into innovation output. Its innovation value depends on whether firms can transform digital technologies, data resources, and organizational routines into capabilities that support knowledge recombination, R&D decision-making, and innovation governance. This interpretation also helps explain why the innovation effects of digital transformation may vary across firms with different resource endowments, ownership structures, financing conditions, and regional environments.

6.2. Interpretation of Mechanism-Consistent Evidence

The mechanism analysis should be interpreted as evidence consistent with the proposed channels rather than as definitive causal mediation evidence. The empirical results show that digital transformation is positively associated with several R&D-related and governance-related indicators. These findings are in line with the theoretical argument that digital transformation may promote corporate innovation by strengthening R&D investment and improving the governance environment. However, because this study is based on observational panel data, the results should not be interpreted as fully proving a causal mediation chain. Instead, they provide empirical support for the plausibility of the proposed dual-path framework.

This cautious interpretation is necessary because innovation is a cumulative, dynamic, and uncertain process. R&D investment, governance quality, and innovation output may influence one another over time. Therefore, the mechanism results are best understood as showing that the observed empirical patterns are consistent with the theoretical logic of this study. Future research may further examine the causal mediation mechanisms through more refined data, policy shocks, quasi-natural experiments, or other identification strategies.

6.3. Managerial and Policy Implications

The findings provide meaningful implications for both firms and policymakers. For firms, digital transformation should not be understood merely as the adoption of digital tools. Instead, firms need to embed digital technologies into R&D management, resource allocation, internal control, and information disclosure systems. Only when digital transformation is integrated with organizational capability building can it be more effectively converted into innovation output.

For policymakers, the heterogeneous results suggest that digital transformation policies should avoid a one-size-fits-all approach. Firms with high financing constraints, smaller scale, non-state ownership, or weaker regional digital infrastructure may face greater difficulty in transforming digital initiatives into innovation outcomes. Therefore, policy support should focus not only on encouraging the adoption of digital technologies, but also on improving financing access, digital infrastructure, governance standards, and innovation-supporting services for firms with relatively weak resource endowments. Such differentiated policy design can help narrow the innovation gap among firms and enhance the inclusive effects of digital transformation.

6.4. Limitations and Future Research

This study has several limitations. First, the text-based DIGI index constructed from annual reports may capture both substantive digital transformation and disclosure intensity. Although annual-report text provides useful information about firms’ digital orientation, it may also reflect managerial rhetoric or symbolic disclosure. Future research may combine textual measures with indicators such as digital investment, software assets, IT capital, digital patents, or other measures of actual digital capability.

Second, although this study includes firm and year fixed effects, instrumental-variable estimation, and several robustness checks, the observational panel-data design cannot fully eliminate endogeneity concerns. Therefore, the findings should be interpreted as robust associations rather than definitive causal estimates. Future studies may use quasi-natural experiments, policy shocks, or more rigorous causal identification strategies to further examine the relationship between digital transformation and innovation.

Third, this study focuses mainly on innovation output measured by invention patents and granted patents. Future research may further distinguish between different types of innovation, such as exploratory and exploitative innovation, radical and incremental innovation, or green and non-green innovation. Such extensions would provide a more nuanced understanding of how digital transformation shapes the direction, quality, and sustainability of corporate innovation.

7. Conclusions

This study investigates the innovation consequences of digital transformation among Chinese A-share listed firms from 2009 to 2023, with particular attention to the roles of R&D investment and governance improvement. Based on panel data analysis and a series of robustness checks, the study finds that digital transformation is positively associated with corporate innovation output. This conclusion remains consistent when innovation is measured by patent applications and patent grants, and it is further supported by alternative model specifications and instrumental-variable estimation.

The empirical findings lead to three main conclusions. First, firms with a higher degree of digital transformation tend to exhibit stronger innovation performance, suggesting that digitalization has become an important driver of corporate innovation. Second, the mechanism analysis shows that this relationship is consistent with two complementary channels: digital transformation is associated with stronger R&D inputs and a more supportive governance environment. Specifically, digital tools may improve resource allocation, enhance R&D efficiency, and support longer-term innovation investment, while also strengthening internal control, information disclosure, and external supervision. Third, the heterogeneity analysis indicates that the innovation effect of digital transformation is more pronounced among firms with lower financing constraints, larger enterprises, firms located in eastern regions, and state-owned enterprises. This suggests that firm characteristics and external conditions play an important role in shaping the innovation outcomes of digital transformation.

Overall, this study shows that digital transformation is not only a technological process, but also an organizational and institutional process that supports corporate innovation. The findings highlight the need for firms to integrate digital strategies with R&D management and governance improvement, while also suggesting that policymakers should create more inclusive conditions for digital innovation by improving infrastructure, financing access, and institutional support for firms with weaker resource endowments. Future research could extend this study by incorporating micro-level behavioral data, distinguishing between different types of innovation, or conducting cross-country comparative analysis. These extensions would deepen our understanding of how digital transformation interacts with corporate innovation and provide more precise insights for policy design and managerial practice.