The Impact of Open Public Data on Corporate Low-Carbon Technological Innovation: Evidence from China

Abstract

Open public data is a vital institutional arrangement for overcoming data constraints in corporate low-carbon technological innovation. Using a panel dataset of China’s Shanghai and Shenzhen A-share listed firms over the 2007–2023 period, this study employs a difference-in-differences (DID) approach to examine the impact of open public data on corporate low-carbon technological innovation. The results show that open public data has a significant positive effect on corporate low-carbon technological innovation, and the results remain robust across multiple validation tests. Mechanism tests point out that government transparency negatively moderates the promotional effect of public data openness on corporate low-carbon technological innovation, while barriers to factor mobility positively moderate this effect. The heterogeneity analysis indicates that the positive impact of open public data is more pronounced among firms characterized by higher R&D investment, lower financial constraints, and greater digitalization. Further analysis indicates that open public data also exhibits significant geographic and industry spillover effects, with the geographic spillover following an inverted U-shaped pattern of decay and the industry spillover driven by peer imitation. This study provides evidence on leveraging open public data to stimulate low-carbon innovation and facilitate green economic transformation, offering valuable insights for advancing data-driven sustainable development globally.

1. Introduction

The global climate system is entering a phase of accelerated warming. According to the 2025 Sustainable Development Goals Report, average global temperatures have risen to approximately 1.55 °C above pre-industrial levels, and greenhouse gas emissions reached a historical peak in 2023. This trend signals an urgent need for more decisive mitigation actions worldwide. As the largest consumer of energy and emitter of carbon dioxide, China’s pathway toward low-carbon transition holds crucial implications for global climate governance [1]. In line with this responsibility, China has pledged to achieve carbon peaking by 2030 and carbon neutrality by 2060, while promoting structural reforms and green technological advancement. Low-carbon technological innovation is central to achieving these dual carbon targets. It encompasses a broad range of technologies related to renewable energy, energy efficiency, carbon capture, and clean industrial processes [2]. Nevertheless, firms frequently face two major obstacles in developing low-carbon innovations: fragmented and costly access to relevant data and limited capability to process and integrate complex information [3]. These constraints impede firms’ ability to identify viable technological directions and reduce the efficiency of innovation investments, making it essential to explore institutional mechanisms that can alleviate data bottlenecks.

The emergence of open public data provides a new institutional avenue for reducing information frictions and supporting firms’ innovation activities. Data have increasingly become key production factors in the digital economy [4]. In 2024, China’s total data output reached 41.06 zettabytes, representing a 25% year-on-year increase (China Academy of Information and Communications Technology (CAICT) and China Cyberspace Research Institute, National Data Resources Survey Report (2024), April 2025). Public data —generated, collected, recorded, and preserved by public institutions in the lawful performance of their public functions—constitutes the largest share of data resources. Characterized by high credibility, wide coverage, and strong externalities, this data has substantial economic and social value [5].

However, for a long time, much of this data remained underutilized due to limited disclosure incentives, capacity constraints, and institutional barriers, resulting in what scholars have described as “dormant data” [6]. In recent years, China has made significant efforts to promote public data openness, with provincial governments gradually launching open data platforms. Public data openness not only safeguards the public’s right to governmental information and enhances transparency, but more crucially, it grants enterprises access to data as a factor of production [7]. This effectively alleviates corporate data shortages, thereby maximizing the resource allocation effects of data and ultimately driving low-carbon technological innovation.

The existing literature offers valuable insights but also reveals certain limitations. First, prior studies have affirmed the institutional role of open public data in facilitating general innovation activities [8,9]. However, such studies have seldom distinguished its impacts across different categories of innovation, particularly those pertaining to low-carbon or other environmentally oriented innovations. In addition, the extant literature on the determinants of low-carbon technological innovation has predominantly concentrated on firm-level resources—such as R&D investment, managerial competencies, and organizational adaptability [10,11,12]—as well as policy interventions related to carbon governance, including credit policies, carbon taxation, low-carbon city pilot programs, and emission trading schemes [10,13,14,15]. Nevertheless, limited attention has been paid to elucidating how public data openness, from the perspective of digital factors, may alleviate bottlenecks in corporate low-carbon technological innovation. Furthermore, whether such effects give rise to discernible spillover outcomes remains an open question requiring empirical substantiation.

Against this backdrop, this study employs data from Chinese A-share listed firms on the Shanghai and Shenzhen stock exchanges from 2007 to 2023 and constructs a multi-period difference-in-differences model to empirically test the impact of public data openness on corporate low-carbon technological innovation. The empirical analysis reveals that open public data significantly promotes corporate low-carbon technological innovation, and this conclusion passes a series of robustness tests. Furthermore, the study examines the moderating role of macro-level institutions in the relationship between public data openness and corporate low-carbon technological innovation, as well as the heterogeneity associated with firms’ internal capabilities. Further analysis reveals that the positive impact of open public data on corporate low-carbon technological innovation exhibits spillover effects at both the geographic and industry levels.

The primary contributions of this study are threefold. First, this study provides micro-level evidence on how public data openness influences corporate low-carbon technological innovation. It enriches the academic discourse on unlocking the value of public data for innovation-driven development and offers valuable policy insights for other economies pursuing green transformation through data openness. Second, by identifying the moderating roles of government transparency and barriers to factor mobility, the study uncovers the complex pathways through which policy environments and market foundations jointly shape the effects of open data on corporate low-carbon innovation. This deepens the understanding of the boundary conditions of data openness. Finally, from a Resource-Based View (RBV) perspective, this paper highlights the asymmetric effects of public data openness on low-carbon technological innovation across dimensions such as R&D capabilities, financing conditions, and digitalization levels. Additionally, it reveals the geographic and industry spillover effects of this policy. These findings provide a scientific foundation for the formulation of targeted, differentiated data openness strategies and the optimization of policy coverage.

The rest of the paper is structured as follows: Section 2 outlines the institutional background and theoretical hypotheses. Section 3 describes research methodology. Section 4 presents specific empirical results and analysis. Section 5 provides further discussion. Section 6 summarizes conclusions, policy implications, and limitations.

2. Institutional Background and Theoretical Hypotheses

2.1. Policy Background

The promotion of public data openness is a key initiative to unlock the value of data as a factor of production. Public data open platforms serve as the vehicles for the release and access of public data, acting as a bridge between data providers and users. In 2015, the Action Plan for Promoting Big Data Development elevated government data openness to a national strategy for the first time. The document emphasized that promoting public data openness and sharing by the government is an effective means to alleviate the shortage of data supply. By making government data public, it can stimulate the data collection and openness of market entities, thus transforming dormant data into valuable applications. Since then, China has gradually established a systematic framework for public data openness. Under the guidance of policies, governments have actively promoted the construction of local public data open platforms, driving data openness in an orderly manner. At the provincial level (including autonomous regions and municipalities) (the term “provincial level” mentioned in the text includes autonomous regions and municipalities directly under the central government), the Shanghai and Beijing governments were the first to launch public data open platforms in 2012, followed by Zhejiang, Guangdong, Guizhou, and other provincial governments. As of 2024, 28 provincial governments have launched public data open platforms, accounting for 90.32% (Lab for Digital & Mobile Governance of Fudan University (2024), China local government data opening report (2024)). Although there are variations in coverage and maturity across regions, these platforms collectively represent a systematic shift in China’s governance model towards a more transparent and open data resource framework.

2.2. Theoretical Hypotheses

Open public data can alleviate data constraints on enterprises, thereby fostering low-carbon technological innovation. This is because of the following: First, the public data released by the government is a “public good”. Enterprises can access, utilize, and analyze the knowledge, market and policy information contained within this data at little or no cost [16,17]. This allows enterprises to more effectively pinpoint the strategic direction of their low-carbon innovation efforts. Moreover, public data is authoritative and reliable. Enterprises that access these data resources can reduce unnecessary costs associated with data search, verification, and trial and error [18,19], thereby effectively reducing the foundational costs of low-carbon technological innovation. It can be seen that open public data improves the generalizability of data resources and lowers the threshold of access to them, enabling firms to better discern policy directions and reallocate resources toward production and R&D activities [20], thereby facilitating low-carbon technological innovation. Based on the above analysis, this study proposes the following hypothesis.

H1. 

Open public data promotes corporate low-carbon technological innovation.

However, the promotional effect of public data openness on corporate low-carbon technological innovation is not uniform across regions; instead, it shows strong dependence on institutional contexts, particularly reflected in differences in government transparency and in barriers to factor mobility.

First, government transparency affects both the ease with which firms access policy information and the stability of their policy expectations. In regions where transparency is high, firms can obtain policy information through existing channels firms [21], and additional public data merely refines or reinforces existing expectations. By contrast, regions characterized by limited transparency are marked by pronounced informational frictions, which make it difficult for firms to discern policy priorities. Because low-carbon innovation typically requires long-term planning and involves uncertain returns [22], a blurred policy environment may lead managers to prioritize short-term objectives at the expense of forward-looking, sustainability-oriented innovation [23]. Public data openness, however, provides a stable and authoritative source of information that partially offsets these shortcomings, reducing ambiguity in the institutional environment [24,25] and thereby exerting stronger incremental value in promoting low-carbon technological efforts. Building on this theoretical reasoning, this paper proposes the following hypothesis.

H2a. 

Government transparency negatively moderates the relationship between public data openness and corporate low-carbon technological innovation.

Second, barriers to factor mobility shape the efficiency with which firms acquire data, technology, and other innovation-relevant resources. In regions where these barriers are low, market entry barriers are weak and resource misallocation is limited. In such environments, enterprises can effectively acquire the information and resources needed for innovation via market channels, including external patents and industry technology trends. In contrast, high barriers to factor mobility coupled with pronounced resource distortions weaken market competition, diminishing leading firms’ incentives to pursue innovation-driven advantages [26,27]. Small and medium-sized enterprises (SMEs) are further constrained by limited access to innovation information and technological resources, thereby impeding their green innovation efforts [28]. Under these conditions, public data openness can mitigate unequal access to data resources and enhance the flow of knowledge and information [7], particularly benefiting resource-constrained firms. Accordingly, the marginal impact of public data openness on low-carbon technological innovation is more pronounced in regions with high factor mobility barriers. Building on this theoretical reasoning, this paper proposes the following hypothesis.

H2b. 

Factor mobility barriers positively moderate the relationship between public data openness and corporate low-carbon technological innovation.

3. Research Design

3.1. Sample Selection and Data Sources

This study examines A-share listed firms on China’s Shanghai and Shenzhen stock exchanges from 2007 to 2023 as the research sample. To enhance the representativeness of the sample, the study employed the following approach: exclusion of (1) ST, *ST, and PT firms; (2) financial and insurance firms; and (3) observations with significant missing data. Finally, we obtained 37,705 firm-year observations. To mitigate the impact of outliers on regression results, all continuous variables were trimmed at the 1% and 99% tails. Regarding data sources, the firm-level data were sourced from the CSMAR database, and patent data were retrieved from the China National Intellectual Property Administration and the Global Patent Database. The data on provincial open public data platforms were derived from the China Local Government Data Openness Report, an annual publication issued by the Digital and Mobile Governance Lab of Fudan University.

3.2. Variable Construction

3.2.1. Dependent Variable

Corporate low-carbon technological innovation($\mathrm{L}\mathrm{c}\mathrm{t}$). While patent data is widely adopted in existing research as a quantitative metric, this approach suffers from significant limitations. Green patents are broader in scope, covering various aspects of environmental protection and sustainable development. In contrast, low-carbon patents concentrate on innovations explicitly designed to support low-carbon technological advancement. To enhance measurement precision, this study draws on Zhu et al. [29] and Chen et al. [10], selecting patents from two primary categories closely related to low-carbon technologies (i.e., alternative energy production and energy-saving technology patents) based on the IPC Green List classification standards. These patents were aggregated, log-transformed, and incorporated into the model.

3.2.2. Core Independent Variable

Open public data($\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$). Open public data is a dummy variable indicating whether the province where a firm is registered has launched a public data platform. It takes the value 1 in the year the platform is launched and in all subsequent years and 0 otherwise.

3.2.3. Control Variables

Referring to Dong ang Wang [30], this paper incorporates a set of firm-level characteristics as control variables: the gearing ratio ($\mathrm{L}\mathrm{e}\mathrm{v}$), return on assets ($\mathrm{R}\mathrm{o}\mathrm{a}$), firm size ($\mathrm{S}\mathrm{i}\mathrm{z}\mathrm{e}$), firm age ($\mathrm{A}\mathrm{g}\mathrm{e}$), social wealth creativity ($\mathrm{T}\mathrm{o}\mathrm{b}\mathrm{i}\mathrm{n}\mathrm{q}$), board size ($\mathrm{B}\mathrm{o}\mathrm{a}\mathrm{r}\mathrm{d}$), the ratio of management shareholding ($\mathrm{M}\mathrm{h}\mathrm{o}\mathrm{l}\mathrm{d}$), and independent directors ($\mathrm{I}\mathrm{n}\mathrm{d}$).

Table 1. Descriptive statistics.

Variables	Definition	Obs.	Mean	SD
Lct	The logarithm of the number of low-carbon patent applications	37,705	0.194	0.553
Open	A value of 1 is assigned when the firm’s province has implemented an open public data platform; otherwise, the value is 0	37,705	0.575	0.494
Lev	The proportion of total debts to total assets	37,705	0.398	0.197
Roa	Net profit after tax to total assets	37,705	0.044	0.054
Size	The natural logarithm of total assets	37,705	22.155	1.271
Age	The natural logarithm of the current year minus the year of listing plus one	37,705	1.913	0.940
Tobinq	The ratio of firm market value to replacement capital	37,705	1.963	1.118
Board	The number of directors on the board	37,705	2.238	0.177
Mhold	The ratio of management shareholding	37,705	0.153	0.205
Idr	The ratio of independent directors	37,705	0.376	0.053

reports summary statistics for all variables.

3.3. Estimation Model

Given the phased implementation of open public data, this study designates provinces that have launched public data platforms as the treatment group, with the remainder serving as the control group. We employ a staggered DID method to evaluate the policy’s impact on corporate low-carbon technological innovation. The baseline regression specification is constructed as follows:

$${Lct}_{it}={\beta }_0+{\beta }_1{Open}_{it}+{\beta }_2{Control}_{it}+{\lambda }_i+{\mu }_t+{\epsilon }_{it}$$

(1)

where $i$ and t denote firm $i$ and year $t$, respectively. ${Lct}_{it}$ represents the level of low-carbon technological innovation of firm $i$ in year $t$. ${Open}_{it}$ indicates whether the province of a firm’s registration has launched a public data platform. ${Control}_{it}$ denotes a set of control variables. ${\lambda }_i$ and ${\mu }_t$ are individual (firm) and year fixed effects, respectively. ${\epsilon }_{it}$ is the random error term, with robust standard errors clustered at the firm level.

4. Results and Analysis

4.1. Benchmark Regression

Table 2. Benchmark regression results.

Variables	Lct
	(1)	(2)
Open	0.028 ***	0.026 ***
	(0.009)	(0.009)
Lev		0.015
		(0.032)
Roa		0.047
		(0.059)
Size		0.060 ***
		(0.010)
Age		0.011
		(0.010)
Tobinq		0.006 **
		(0.003)
Board		−0.004
		(0.035)
Mhold		0.137 ***
		(0.034)
Idr		−0.020
		(0.087)
Firm FE	YES	YES
Year FE	YES	YES
F-value	10.675 ***	8.304 ***
Observations	37,705	37,705
R-squared	0.636	0.638
Adj. R-squared	0.590	0.592

Notes:; ** p < 0.05; *** p < 0.01; standard errors in parentheses are clustered at the firm level.

reports the results of the benchmark regression, examining the impact of open public data on corporate low-carbon technological innovation. Column (1) displays baseline results with no control variables, whereas Column (2) presents the results incorporating the full set of covariates. It is discernible that the estimated coefficients for $\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ in Columns (1) and (2) are both positive at the 1% significance level, confirming that open public data significantly promotes corporate low-carbon technological innovation. From an economic perspective, an increase of one standard deviation (0.494) in public data openness leads to an average of a 2.322% increase in corporate low-carbon technological innovation (economic significance is calculated as the coefficient of the independent variable multiplied by its standard deviation, divided by the standard deviation of the dependent variable). Based on this, Hypothesis 1 is validated.

4.2. Robustness Checks

4.2.1. Parallel Trend Test

Since the DID framework depends on the parallel trend assumption between treated and untreated firms, we followed Beck et al. [31] and disaggregated the event timeline around the policy implementation to examine the time-varying impact of public data openness on corporate low-carbon innovation. Additionally, to mitigate sample imbalance, following by Fung et al. [32], observations from periods earlier than $t$ = −5 were reset to −5, and those later than $t$ = 4 were reset to 4, with $t$ = −1 designated as the baseline (omitted) period. The dynamic DID model is specified as follows:

$${Lct}_{it}={\beta }_0+\sum _{k=-5,k\ne -1}^{k=4}{\beta }_k{Open}_{it}^k+{\beta }_2{Control}_{it}+{\lambda }_i+{\mu }_t+{\epsilon }_{it}$$

(2)

In the equation, ${Open}_{it}^k$ is a dummy variable indicating whether firm $i$ is in the kth period of the pre-treatment phase ($k<0$) or the post-treatment phase ($k\ge 0$) during time period $t$. The meanings of other variables remain consistent with baseline regression model (1).

The results of the parallel trends test are presented in Figure 1. Under the condition of a 95% confidence interval, the coefficients ${\beta }_k$($k<0$) fluctuate around 0 and are not significant before the establishment of the public data platform. This indicates that, prior to the platform launch, there is no significant difference in low-carbon technological innovation between treatment and control groups, satisfying the parallel trend assumption. Following the release of public data, the coefficients ${\beta }_k$($k\ge 0$) become significantly positive and show an upward trend, indicating that the promotional effect of open public data on corporate low-carbon technological innovation is both sustained and increasing. This suggests that the policy’s effectiveness strengthens as the breadth and depth of data openness increase.

4.2.2. Tests for Heterogeneous Treatment Effects

The effects of a policy may vary across individuals in terms of treatment timing and treatment duration. In the presence of such multidimensional heterogeneity, traditional two-way fixed-effects (TWFE) models may lead to biased estimates. To address this issue, we adopted the CSDID method proposed by Callaway and Sant’Anna [33] to assess the robustness of the multi-period DID estimates. This method, based on a double-robust approach, helped mitigate potential biases in the DID estimates. The results are presented in

Table 3. Heterogeneous treatment effects.

Variables	Simple Weighting	Calendar Time	Group	Dynamic
	(1)	(2)	(3)	(4)
Simple ATT	0.057 ***
	(0.022)
GAverage		0.036 *
		(0.020)
GAverage			0.057 ***
			(0.021)
Pre_avg				−0.085
				(0.560)
Post_avg				−0.071 **
				(0.031)

Notes: * p < 0.1; ** p < 0.05; *** p < 0.01; standard errors in parentheses are clustered at the firm level.

. Our findings indicate that the average treatment effects across four different specifications consistently show that public data openness significantly promotes corporate low-carbon technological innovation, thereby validating the robustness of our baseline results.

4.2.3. Placebo Test

To further eliminate the potential influence of unobserved random variables on corporate low-carbon technological innovation, this study employed a placebo test by randomly assigning both the timing and location of public data releases. Specifically, we performed 500 random sampling iterations. In each iteration, 26 provinces were randomly selected and assigned fictitious launch dates to form a virtual treatment group, while the remaining 5 provinces constituted the virtual control group. The benchmark regression model was then re-estimated.

Figure 2 presents the kernel density distributions of the estimated coefficients and p-values for the explanatory variables in the placebo test. The kernel density of the estimated coefficients nears a normal distribution, centered at zero. Only a very small number of regression coefficients exceed the actual regression coefficients listed in Column (2) of Table 2. Furthermore, the estimated coefficients are non-significant in most cases. This outcome reveals that the influence of open public data on low-carbon technological innovation is not coincidental or attributable to other unknown factors, thereby confirming the reliability of this study’s findings.

4.2.4. Eliminating Other Policies’ Interference

To test whether the benchmark results are affected by other relevant policies, this study controlled for several concurrent initiatives in Columns (1) to (4) of

Table 4. Test outcomes isolating influence from other policies.

Variables	Lct
	(1)	(2)	(3)	(4)	(5)
Open	0.023 ***	0.026 ***	0.029 ***	0.031 ***	0.028 ***
	(0.009)	(0.009)	(0.009)	(0.009)	(0.010)
Pinfor	−0.022				−0.028 *
	(0.015)				(0.016)
Ndata		−0.007			−0.009
		(0.016)			(0.017)
Wcity			−0.014		−0.007
			(0.017)		(0.017)
Lcarbon				−0.023 *	−0.024 *
				(0.013)	(0.014)
Control variables	YES	YES	YES	YES	YES
Firm FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
Observations	37,705	37,705	34,150	34,150	34,150
R-squared	0.638	0.638	0.642	0.642	0.642

Notes: * p < 0.1; *** p < 0.01; standard errors in parentheses are clustered at the firm level.

: the Public Information Resource Opening Pilot Policy ($\mathrm{P}\mathrm{i}\mathrm{n}\mathrm{f}\mathrm{o}\mathrm{r}$), the National Big Data Comprehensive Pilot Zone Policy ($\mathrm{N}\mathrm{d}\mathrm{a}\mathrm{t}\mathrm{a}$), the Smart City Pilot Policy ($\mathrm{W}\mathrm{c}\mathrm{i}\mathrm{t}\mathrm{y}$), and the Low-carbon City Pilot Policy ($\mathrm{L}\mathrm{c}\mathrm{a}\mathrm{r}\mathrm{b}\mathrm{o}\mathrm{n}$). Columns (1) to (4) in Table 4 display the regression results with each policy controlled individually, while Column (5) presents the results controlling for all policies simultaneously. The results show that the estimated coefficients for $\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ are all significantly positive, further validating the robustness of the benchmark regression results.

4.3. Other Robustness Tests

To further validate the robustness of the benchmark results, we conducted the following additional tests.

4.3.1. Two-Way Clustered Standard Errors

To account for potential intertemporal autocorrelation and intra-provincial correlation, we employed two-way clustered standard errors across firm and provincial dimensions in the benchmark regression. Columns (1) and (2) of

Table 5. Results of robustness tests.

Variables	Two-Way Clustered Standard Errors		Removing Samples Prior to 2012	PSM	Replacing Dependent Variable
	(1)	(2)	(3)	(4)	(5)
	Lct	Lct	Lct	Lct	Gup
Open	0.026 **	0.026 **	0.027 ***	0.027 ***	0.019 **
	(0.009)	(0.009)	(0.010)	(0.009)	(0.010)
Control variables	YES	YES	YES	YES	YES
Firm FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
Observations	37,705	37,705	32,019	36,336	33,488
R-squared	0.638	0.638	0.674	0.641	0.639

Notes:; ** p < 0.05; *** p < 0.01; standard errors in parentheses are clustered at the firm level.

shows that the coefficients of $\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ are statistically significant and positive at the 5% level, suggesting that variations in the clustering level of standard errors do not materially influence the conclusions of the baseline regression.

4.3.2. Adjusting the Sample Period

Since Shanghai launched China’s first provincial-level open public data platform in 2012, when very few platforms existed, this paper excluded samples prior to 2012 from the regression analysis. Column (3) of Table 5 presents the estimation results, indicating that the coefficient of $\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ is 0.027 and achieves significance at the 1% level, further confirming the robustness of the benchmark regression conclusions.

4.3.3. PSM

Given that sample characteristics may introduce selection bias due to group incompatibility, this study employed the PSM–DID method to mitigate this bias. In particular, we used control variables as matching covariates and implemented annual 1:1 nearest-neighbor matching. Subsequently, regression was performed on the matched observations that met these requirements. The results presented in Columns (4) of Table 5 are consistent with benchmark regression, reaffirming the robustness of the baseline results.

4.3.4. Replacing Dependent Variable

This study used the logarithm of green utility model patent applications ($\mathrm{G}\mathrm{u}\mathrm{p}$) to measure low-carbon technological innovation. The regression results are shown in Column (5) of Table 5, where the coefficient of $\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ is still positive and significant, confirming the robustness of the benchmark regression conclusions.

4.4. Endogeneity Test

The above analyses controlled for multiple factors that may influence the baseline model’s results across multiple dimensions, thereby alleviating endogeneity to some extent. However, potential biases arising from reverse causality may still exist. For example, leading low-carbon technology firms may have incentives to promote data openness through policy advocacy and other channels to strengthen their competitive advantage. Concurrently, governments may proactively release data to reinforce the success of their transformation efforts. This two-way dynamic creates a bidirectional interaction between corporate demand for low-carbon innovation and government data supply. To address this, we employed an instrumental variable approach, estimated using two-stage least squares (2SLS).

Following Lyu et al. [11], this study selected the density of long-distance optical cables ($\mathrm{L}\mathrm{a}\mathrm{b}\mathrm{l}\mathrm{e}$) in the province where a firm is located as an instrumental variable. The rationale for its use is as follows: First, the density of optical cables reflects the foundational level of regional digital construction and is a prerequisite for building, maintaining, and expanding public data platforms. The establishment of such platforms requires stable high-bandwidth communication networks to support data aggregation, API access, and real-time data delivery. The Outline of Action for Big Data Development and the 14th Five-Year Plan for Digital Economy Development explicitly require local governments to “strengthen the underlying digital infrastructure such as broadband and optical-fiber networks to support public data openness and interconnection”, thereby satisfying the relevance condition. Second, the construction of long-distance optical fiber cables in China is centrally planned and managed by the country’s four major network operators. Consequently, individual firms cannot alter or influence this process. In addition, optical cable density does not shape firms’ internal production technologies, carbon-related costs, or incentives to conduct R&D. Since no theoretical mechanism links optical cable density directly to low-carbon technological innovation at the firm level, the exogeneity assumption is reasonably satisfied.

Column (1) of

Table 6. Endogenous treatment estimation results.

Variables	First Stage	Second Stage
	(1)	(2)
	Open	Lct
Open		0.167 **
		(0.0679)
Iv	0.893 ***
	(0.0599)
Control variables	YES	YES
Firm FE	YES	YES
Year FE	YES	YES
Kleibergen–Paap rk LM statistics		122.605 ***
Cragg–Donald Wald F statistic		442.057
Observations	37,705	37,705

Notes:; ** p < 0.05; *** p < 0.01; standard errors in parentheses are clustered at the firm level.

presents the regression results for the first stage. At the 1% significance level, the instrumental variable shows a significant positive correlation with public data openness, consistent with theoretical expectations. Column (2) of Table 6 presents the regression results for the second stage. The coefficient of $\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ is significantly positive at the 5% level, consistent with the direction observed in the benchmark regression. Furthermore, the Kleibergen–Paap rk LM statistic is significantly positive at the 1% level, and the Cragg–Donald Wald F-statistic is 333.67, exceeding the threshold of 10. This eliminates concerns of under-identification and weak instrumental variables. These results indicate that, after addressing endogeneity, the main conclusions of this study remain valid.

4.5. Mechanism Analysis

4.5.1. Government Transparency

Referencing to Berliner [34], this study uses the Government Transparency Index published by the Chinese Academy of Social Sciences to measure government transparency($\mathrm{G}\mathrm{t}$), with lower values indicating weaker transparency. Specifically, the variable $\mathrm{G}\mathrm{t}$ and its interaction term $\mathrm{G}\mathrm{t}\times \mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ are introduced into model (1) for estimation. As shown in Column (1) of

Table 7. Mechanism test results.

Variables	Lct
	(1)	(2)
Open	0.039 ***	0.012
	(0.009)	(0.010)
Gt	0.022 **
	(0.009)
Open × Gt	−0.027 ***
	(0.010)
Fmb		−0.025 **
		(0.010)
Open × Fmb		0.031 ***
		(0.012)
Control variables	YES	YES
Firm FE	YES	YES
Year FE	YES	YES
Observations	29,629	36,715
R-squared	0.674	0.638

Notes:; ** p < 0.05; *** p < 0.01; standard errors in parentheses are clustered at the firm level.

, the coefficient of the interaction term is significantly negative, indicating that the positive effect of open public data on corporate low-carbon technological innovation is more pronounced when government transparency is lower. This finding partially supports Hypothesis 2a that government transparency negatively moderates the relationship between public data openness and corporate low-carbon technological innovation.

4.5.2. Barriers to Factor Mobility

Referencing to Gutierrez and Philippon [35], we adopt the Herfindahl–Hirschman Index ($\mathrm{H}\mathrm{H}\mathrm{I}$) as a proxy for barriers to factor mobility($\mathrm{F}\mathrm{m}\mathrm{b}$). A higher $\mathrm{H}\mathrm{H}\mathrm{I}$ value indicates greater market concentration and, consequently, stronger barriers to factor mobility. Specifically, both $\mathrm{F}\mathrm{m}\mathrm{b}$ and the interaction term $Fmb\times Open$ $\mathrm{F}\mathrm{m}\mathrm{b}\times \mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ are incorporated into model (1) for estimation. As shown in Column (2) of Table 7, the coefficient of the interaction term is significantly positive, implying that the promotional effect of open public data on corporate low-carbon technological innovation is stronger in regions with higher market concentration—that is, where barriers to factor mobility are greater. This finding partially supports Hypothesis 2b that factor mobility barriers positively moderate the relationship between public data openness and corporate low-carbon technological innovation.

4.6. Heterogeneity Analysis

Open public data is an enabling resource, providing additional informational support to help firms identify innovation opportunities. However, whether a firm can effectively transform these external resources into innovative outcomes depends on its internal resources and capabilities [36]. First, as low-carbon technological innovation involves long cycles, high risks, and considerable uncertainty [37], R&D activities demand substantial financial resources. Firms with greater R&D investment are more likely to engage in sustained innovation activities [38]. Second, the limited capacity of internal financing forces firms to rely on external capital to pursue high-quality innovation [39]. An improved ability to obtain financing not only lowers the cost of innovation funding but also enables firms to scale up their financing, thereby providing more robust support for high-quality innovation activities. Finally, the adoption of advanced digital technologies—such as blockchain and the IoT—substantially enhances automation, data security, and analytical efficiency [40]. This multi-layered digital infrastructure equips enterprises to collect, integrate, and process diverse data sources more effectively [41], thereby significantly strengthening their capacity to transform open public data into low-carbon innovation outcomes. Accordingly, this study employs three firm-level variables—R&D investment, financing constraints, and digitalization level—to capture the heterogeneous effects of firm capabilities. This analysis elucidates how firms integrate, build, and reconfigure internal and external resources to adapt to rapidly changing environments.

4.6.1. R&D Investment

The logarithmic value of R&D expenditure is used to measure firms’ R&D investment. Specifically, the sample is divided into high- and low-R&D investment groups based on whether a firm’s R&D investment exceeds the sample median. The results are presented in Columns (1) and (2) of

Table 8. Heterogeneity analysis results.

Variables	R&D Investment		Financing Constraints		Digitalization
	Low	High	Low	High	Low	High
	(1)	(2)	(3)	(4)	(5)	(6)
Open	0.022 **	0.042 **	0.037 ***	0.007	0.017	0.022 *
	(0.009)	(0.017)	(0.012)	(0.013)	(0.011)	(0.012)
Control variables	YES	YES	YES	YES	YES	YES
Firm FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	13,654	13,777	18,426	18,419	16,866	15,995
R-squared	0.572	0.709	0.666	0.671	0.610	0.728
p-value of Chow test	0.000		0.029		0.061

Notes: * p < 0.1; ** p < 0.05; *** p < 0.01; standard errors in parentheses are clustered at the firm level.

. The estimated coefficient of $\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ is significantly positive across both subsamples. However, the Chow test confirms a statistically significant difference between them. The results show that the effect of open public data is greater for firms with high R&D investment than it is for those with low R&D investment.

4.6.2. Financing Constraints

Following Kaplan and Zingales [42], we use the KZ index to measure firms’ financing constraints, with a lower value indicating weaker constraints. Specifically, firms are divided into high- and low-constraint groups based on whether their KZ index exceeds the sample median. The results are presented in Columns (3) and (4) of Table 8. The estimated coefficient of $\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ is significantly positive only for firms facing lower financing constraints. This suggests that open public data primarily promotes low-carbon technological innovation among those with weaker financing constraints relative to firms with higher financing constraints.

4.6.3. Digitalization Level

Following He et al. [43], this study develops a comprehensive index system to assess digital transformation by employing a text analysis approach. Specifically, the sample is divided into high- and low-digitalization groups based on whether a firm’s digital transformation index exceeds the sample median. Columns (5) and (6) of Table 8 indicate that the estimated coefficient of $\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ is significantly positive only for firms with higher levels of digitalization. This indicates that open public data primarily promotes low-carbon technological innovation among firms with higher levels of digitalization, relative to firms with lower digitalization levels, underscoring the amplifying role of digital capability.

5. Further Discussion: Non-Rival Spillover Effects of Public Data Openness

The non-rivalrous nature of public data allows its value to extend beyond direct beneficiaries, creating the potential for extensive spillover effects through knowledge diffusion. Accordingly, this section extends the analysis along two dimensions: geographic and industrial. First, it investigates whether open public data can transcend administrative boundaries to foster regional collaborative innovation. Second, it explores whether data openness induces technological imitation and competitive convergence within industries, thereby reshaping the broader innovation ecosystem.

5.1. Geographical Spillover Effect

Following Alder et al. [44], this study constructs the following model to analyze the spatial spillover effects of public data openness:

$${Lct}_{it}={\beta }_0+{\beta }_1{Open}_{it}+\sum _{s=200}^{1200}{\delta }_s\times N_{it}^s+{\beta }_2{Control}_{it}+{\lambda }_i+{\mu }_t+{\epsilon }_{it}$$

(3)

where $s$ denotes the geographical distance between provinces (in kilometers, $s\ge 200$), measured by the straight-line distance between any two provincial capitals. ${\delta }_s$ reflects the impact of public data openness on low-carbon technological innovation among firms in other regions. $N_{it}^s$ is the geographic spillover variable, taking a value of 1 if at least one provincial public data openness platform exists within $s$ kilometers of firm i’s location (excluding its own province) and 0 otherwise ($s$ = 200, 400, …, 1000). The remaining variables are defined consistent with model (1).

As shown in Figure 3, the geographic spillover effects of public data openness exhibit an inverted U-shaped pattern. First, there exists a localized promotion effect: within the 400–600 km range, the coefficient ${\delta }_s$ is 0.028 and statistically significant at the 1% level. This occurs because a medium distance achieves a balance between the knowledge diffusion effect and the competitive suppression effect—proximity may lead to resource siphoning, whereas excessive distance results in information distortion. Second, a boundary attenuation effect is observed. When the distance is below 400 km or beyond 600 km, the coefficients become insignificant, confirming the existence of a “critical spillover radius” in geoeconomic space. As distance increases, spillover effects display a nonlinear pattern of “insignificant → significant → insignificant,” rather than a monotonic decline.

5.2. Industry Spillover Effect

To examine the industrial spillover effects of public data openness, this study follows Li et al. [45] and constructs the following model:

$${Lct}_{it}={\beta }_0+{\beta }_1{Open}_{it}+{\beta }_2{Peer}_{it}+{\beta }_3{Control}_{it}+{\lambda }_i+{\mu }_{pt}+{\epsilon }_{it}$$

(4)

$${Peer}_{it}={\displaystyle \frac{{n\_treatfirms}_{mt}-{Open}_{it}}{{n\_firms}_{mt}-1}}$$

(5)

where $m$ denotes industry. ${Peer}_{it}$ represents the extent to which firms in the same industry as firm $i$ are affected by the launch of provincial public data platforms in year $t$, thereby capturing the industry peer effect of public data openness. Specifically, ${n\_firms}_{mt}$ denotes the total number of firms in industry $m$ during year $t$; ${n\_treatfirms}_{mt}$ represents the number of firms in industry $m$ during year $t$ that are affected by the launch of the provincial public data platform in their location; and $({n\_treatfirms}_{mt}-{Open}_{it})$ can be interpreted as the number of other firms—excluding firm $i$—in industry $m$ and year $t$ that are influenced by the launch of provincial public data platforms in their respective provinces. The remaining variables are defined consistently with model (1).

As reported in

Table 9. Industry spillover effect.

Variables	Lct
	(1)	(2)
Open	0.025 ***	0.025 ***
	(0.009)	(0.009)
Peer	0.065 **	0.054 *
	(0.031)	(0.031)
Control variables	YES	YES
Firm FE	YES	YES
Year FE	YES	YES
Industry FE	YES	YES
Observations	37,600	37,600
R-squared	0.639	0.641

Notes: * p < 0.1; ** p < 0.05; *** p < 0.01; standard errors in parentheses are clustered at the firm level.

, the regression results show that the coefficients for both $\mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}$ and $\mathrm{P}\mathrm{e}\mathrm{e}\mathrm{r}$ are significantly positive, demonstrating that the impact of public data platforms exhibits a significant within-industry peer effect. Specifically, as more peer firms within the same industry are affected by the establishment of provincial-level public data platforms, individual firms exhibit greater low-carbon technological innovation. Overall, these results suggest that the non-rival nature of public data reshapes industrial innovation ecosystems through mechanisms of technological imitation and knowledge diffusion.

6. Conclusions and Policy Implications

This study leverages China’s local government open public data platforms as a quasi-natural experiment to systematically examine the impact of public data openness on corporate low-carbon technological innovation. The study finds that public data openness significantly promotes corporate low-carbon technological innovation. Mechanism analysis reveals that government transparency negatively moderates the promotional effect of public data openness on corporate low-carbon technological innovation, while barriers to factor mobility positively moderate this effect. Heterogeneity analyses indicate that the promoting effect of public data openness on low-carbon technological innovation is more pronounced in firms with higher R&D investment, greater digitalization levels, and lower financing constraints. Further analysis indicates that open public data also exhibits significant geographic and industry spillover effects, with the geographic spillover following an inverted U-shaped pattern of decay and the industry spillover driven by peer imitation.

Based on the above conclusions, the following policy recommendations are proposed. First, open public data plays a crucial role in promoting corporate low-carbon technological innovation. While pilot policies are progressing, many provinces and cities have yet to participate in the development of open public data platforms. Therefore, governments at all levels should establish unified public data platforms to ensure equitable access to data across regions. These platforms should prioritize supporting data essential for low-carbon technologies, such as renewable energy and energy efficiency, to facilitate innovation in this sector.

Second, differentiated data support policies should be developed. As regions with lower government transparency and higher barriers to factor mobility stand to benefit more from open data, policies should focus on providing data resources and information to areas with weaker institutional environments, thereby addressing regional disparities. For instance, fostering stronger cooperation between the public sector and firms and promoting resource sharing and flow can help facilitate innovation.

Finally, policy measures should be implemented to alleviate enterprise capability constraints. Since open public data has a stronger impact on companies with higher R&D investment, greater digitalization, and fewer financing constraints, the government should incentivize firms with lower R&D investment to leverage open data for low-carbon innovation through fiscal subsidies, tax incentives, and similar tools. For firms facing significant financing challenges, green credit and green finance initiatives can help ease these constraints. For enterprises with lower levels of digitalization, the government can offer technical assistance, enabling them to better utilize open data for low-carbon technological innovation.

This study has several limitations. First, although the study verifies the positive effects of open public data on corporate low-carbon technological innovation, it does not distinguish the heterogeneous impacts of different types of data elements. Second, the analysis focuses on whether data are open, without further exploring the role of data openness quality. Future research could extend this work by incorporating cross-country samples, evaluating the classification of data elements, and examining the mechanisms through which data quality influences low-carbon innovation.