Breaking the Urban Carbon Lock-In: The Effects of Heterogeneous Science and Technology Innovation Policies on Urban Carbon Unlocking Efficiency

Abstract

Digital technologies such as big data are reshaping resource allocation, raising interest in whether and how heterogeneous science and technology innovation (STI) policies can help unlock urban carbon lock-in. Using panel data for 286 prefecture-level cities in China from 2009 to 2023, this paper examines the relationship between heterogeneous STI policy intensity—classified as supply-side, demand-side, complementary-factor, and institutional-reform policies—and urban carbon unlocking efficiency. We develop a mechanism-based framework and empirically assess (i) the moderating roles of digital infrastructure, science and technology finance, and government green attention, and (ii) spatial spillover effects using spatial econometric models. The results show that all four policy types show a significant positive association with local carbon unlocking efficiency, with institutional-reform policies exhibiting the strongest association. When the four types are included jointly, only supply-side and demand-side policies retain statistically significant direct associations. Heterogeneity analyses indicate that demand-side, complementary-factor, and institutional-reform policies are more strongly associated with efficiency gains in low-pollution cities, whereas supply-side and demand-side policies have a stronger association in high energy-consuming cities. Mechanism analysis reveals that regional digital infrastructure exerts a selective moderating effect on the relationship between heterogeneous sci-tech innovation policies and urban carbon emission reduction efficiency. It positively reinforces the effectiveness of supply-side, demand-side, and institutional reform-oriented policies, while its interaction with complementary policies is statistically insignificant. Technology finance and government green policies function as a “resource catalyst” and an “institutional guarantee” respectively, significantly enhancing the correlation between heterogeneous sci-tech innovation policies and urban carbon emission reduction efficiency. Finally, carbon unlocking efficiency displays significant spatial dependence: the intensity of supply-side and institutional-reform policies is positively associated with carbon unlocking efficiency in neighboring cities, while complementary-factor policies exhibit a negative spatial association. Overall, the findings provide empirical evidence to inform the design and coordination of heterogeneous STI policy portfolios aimed at improving urban carbon unlocking efficiency.

1. Introduction

Unlocking urban carbon lock-in has become a critical global challenge in the transition toward low-carbon and high-quality development. Carbon emission rigidity, deeply embedded in long-term extensive growth trajectories, has increasingly constrained cities’ ability to achieve sustained decarbonization [1]. This challenge has attracted broad international recognition. From the United Nations Sustainable Development Goals (SDGs) to the Paris Agreement, breaking carbon lock-in and improving carbon unlocking efficiency are widely regarded as essential pathways for mitigating climate change and advancing sustainable development. Existing studies demonstrate that factors such as infrastructure upgrading, industrial structure adjustment, and the development of green finance play important roles in enhancing carbon unlocking efficiency [2,3,4,5]. Recent studies have further advanced this line of inquiry by explicitly examining policy-driven mechanisms of urban carbon lock-in and unlocking. For example, Chao and Qi (2024) employ a quasi-natural experiment based on China’s carbon emission trading policy and demonstrate that market-based environmental regulation can significantly alleviate urban carbon lock-in through channels such as green technological innovation and industrial structure upgrading [6]. This emerging evidence highlights the importance of institutional and policy instruments in shaping urban decarbonization pathways. Beyond carbon trading policies, recent studies have also documented that other policy instruments—such as green finance reforms, innovative city pilot programs, and big data oriented digital policies—can significantly enhance urban carbon unlocking efficiency through channels including industrial upgrading, energy intensity reduction, technological innovation, and public environmental awareness [7,8,9]. However, despite growing evidence on the role of specific environmental or market-based policies, existing studies have largely focused on single-policy instruments and paid limited attention to the systemic effects of heterogeneous science and technology innovation (STI) policy portfolios. Although these studies provide valuable causal evidence, they typically examine isolated policy interventions—such as carbon trading, green finance, or digital pilot policies—without considering the joint and potentially interacting effects of heterogeneous STI policy instruments operating simultaneously within urban innovation systems. In particular, how differentiated STI policy instruments, shaped by digitalization and evolving innovation ecosystems, jointly influence urban carbon unlocking efficiency remains insufficiently explored. As core policy tools guiding the allocation of innovation resources, heterogeneous STI policy instruments can better align differentiated innovation demands across sectors, accelerate the research, development, and commercialization of green technologies, and ultimately activate new momentum for urban carbon unlocking. This perspective is consistent with recent advances in green transition research that emphasize portfolio-based and multi-objective approaches, in which environmental constraints are treated as integral components of complex decision systems rather than as marginal considerations [10]. Extending such portfolio-based thinking from firm-level optimization to the domain of public STI policy design provides a useful conceptual lens for understanding urban carbon unlocking.

As the world’s second-largest economy, China bears substantial responsibility in global decarbonization efforts [11,12]. The country has articulated clear “dual carbon” goals—peaking carbon dioxide emissions by 2030 and achieving carbon neutrality by 2060—and has introduced a comprehensive set of STI policies tailored to urban low-carbon transformation [13,14]. China’s national strategies emphasize green development as a central pillar of modernization [7], with enhancing urban carbon unlocking efficiency representing a concrete operational pathway [15]. This context provides a valuable empirical setting to examine the relationship between heterogeneous STI policies and urban carbon unlocking, offering insights for other developing economies. Against this backdrop, this study departs from existing policy evaluations in three important ways. First, rather than examining the effects of a single environmental regulation, it adopts a heterogeneous and portfolio-based perspective to assess how different STI policy instruments jointly affect urban carbon unlocking efficiency. Second, by explicitly integrating digitalization into the STI policy framework, this study captures the evolving role of innovation systems in shaping low-carbon transitions. Third, it develops an efficiency-based measure of urban carbon unlocking that emphasizes the transformation capacity of cities, thereby complementing conventional indicators such as carbon intensity or green total factor productivity.

However, existing research exhibits notable gaps in examining the relationship between STI policies and urban carbon unlocking. On the one hand, the majority of studies focus primarily on the challenges faced by STI policies [16] or their macroeconomic effects [17], often providing aggregate assessments of their impacts on indicators such as green total factor productivity or carbon emissions [18,19]. Such analyzes tend to overlook the inherently heterogeneous nature of policy supply. In practice, STI policies encompass fundamentally different instruments—including supply-side policies (e.g., R&D subsidies and mission-oriented technological programs), demand-side policies (e.g., green public procurement and subsidies for low-carbon products), complementary measures (e.g., innovation platform construction and incentives for industry–university–research collaboration), and institutional reforms (e.g., strengthened intellectual property protection and the establishment of low-carbon standards). These policy types differ substantially in their underlying mechanisms, implementation contexts, and regulatory intensity [20], implying that their effects on urban carbon unlocking are unlikely to be uniform. This heterogeneity contrasts with the predominant focus of recent carbon unlocking studies on single policy experiments, such as innovative city construction [7], big data pilot zones [8], green finance reform initiatives [9], and voluntary carbon inclusion mechanisms [21], which, while informative, provide limited insights into coordinated policy design. On the other hand, existing studies largely confine their analysis to estimating the direct effects of STI policies on carbon unlocking outcomes [7], with limited attention paid to the underlying transmission mechanisms. In particular, insufficient consideration has been given to how enabling conditions in the digital economy—such as digital infrastructure and science and technology finance—may moderate policy effectiveness. Moreover, the potential spatial interactions of policy effects across cities have rarely been incorporated into empirical analyzes. Although a few recent studies have begun to document spatial spillover effects of specific policies on carbon unlocking efficiency [7,9], such analyses remain fragmented and policy-specific, leaving open the question of how heterogeneous STI policy portfolios generate spatial externalities across cities. This omission constrains both the theoretical understanding of policy functioning and the empirical basis for designing more targeted and effective STI policy mixes.

Building on this foundation, our study seeks to extend the current literature by examining these underexplored dimensions. China’s empirical context, marked by its explicit “dual carbon” goals [13,14] and active deployment of a diverse STI policy mix [7,15], offers a valuable setting for this investigation. Against this backdrop, this study employs panel data from 286 Chinese prefecture-level cities (2009–2023) to examine the association between the intensity of four heterogeneous STI policy types and urban carbon unlocking efficiency (UCE). Specifically, it investigates: (1) the baseline relationships; (2) the moderating roles of digital infrastructure, science and technology finance, and government green attention; and (3) the spatial spillover effects across cities using a spatial Durbin model.

Compared with the existing literature, this study makes three contributions. First, it provides disaggregated empirical evidence on the distinct relationships between different STI policy types and urban carbon unlocking efficiency, moving beyond a homogeneous policy perspective. Second, it develops and applies a dedicated measure of UCE, offering a more policy-sensitive indicator for evaluating low-carbon transition efficiency. Third, it extends the analysis to incorporate key moderators from the digital economy and spatial interactions, offering a more contextual and spatial understanding of the policy-efficiency relationship. These findings aim to inform the design of more targeted STI policy portfolios.

2. Theoretical Analysis and Research Hypotheses

2.1. Direct Effects of Heterogeneous STI Policy Intensity on Urban Carbon Unlocking Efficiency

Urban carbon lock-in denotes a multidimensional rigidity shaped by the interplay of technological path dependence, sunk capital accumulation, institutional inertia, and market failures. Breaking this lock-in necessitates policy instruments capable of intervening across multiple dimensions simultaneously, rather than relying on homogeneous policy approaches [22,23]. Science and technology innovation (STI) policies emerge as a critical means to enhance urban carbon unlocking efficiency, as they target the technological, market, systemic, and institutional roots of carbon lock-in. From a functional perspective, heterogeneous STI policies can be categorized into four types—supply-side, demand-side, complementary, and institutional reform policies—each differing fundamentally in intervention logic, application contexts, and intensity, and each operating through distinct causal pathways to address core mechanisms of carbon lock-in. Consequently, variations in the intensity of these policy categories are expected to exert differentiated effects on the efficiency and pathways of urban carbon unlocking. This classification aligns with the framework proposed by Borrás and Edquist (2013) [24], who systematically examine the typology and selection logic of innovation policy instruments and emphasize the need for policies to precisely target different stages of the innovation process, providing an important theoretical foundation for distinguishing heterogeneous STI policies. Moreover, empirical evidence from Chinese cities confirms that innovation-oriented urban development policies significantly boost carbon unlocking efficiency [7], further validating the practical relevance of STI policies as a core instrument for breaking carbon lock-in. Below is a detailed articulation of the causal mechanisms linking each policy type to urban carbon unlocking efficiency, alongside testable propositions.

Specifically, supply-side STI policies—including R&D subsidies, tax incentives, basic research programs, and talent development initiatives—focus on strengthening the upstream technological supply of low-carbon innovation by directly increasing input into low-carbon R&D and human capital accumulation. The intensity of such policies directly determines the depth and robustness of the technological foundation for carbon unlocking: higher policy intensity translates to greater resource injection into core low-carbon technologies (e.g., renewable energy, carbon capture), which in turn enhances the maturity and competitiveness of these technologies. Edquist (2001) argues that innovation policy constitutes a form of public action that shapes technological change and innovation activities [25], with supply-side policies playing a pivotal role in alleviating technological path dependence—a core mechanism of carbon lock-in. Specifically, by boosting the R&D capacity of enterprises and research institutions, supply-side policies help develop alternative low-carbon technologies that can compete with incumbent high-carbon technologies, thereby weakening the lock-in effect of path-dependent high-carbon technological systems. Through this mechanism of enhancing low-carbon technological capabilities, stronger supply-side STI policies help break the vicious cycle of “high-carbon technological path dependence–insufficient low-carbon technology supply,” as improved low-carbon technology supply reduces reliance on high-carbon paths and further drives technological upgrading, ultimately improving urban carbon unlocking efficiency [26].

Demand-side STI policies, such as green public procurement, consumption subsidies, demonstration projects, and carbon pricing mechanisms, aim to cultivate market demand for low-carbon technologies by creating predictable market signals and reducing the cost of low-carbon technology adoption. The intensity of these policies directly affects the scale and speed of technology diffusion and commercialization: higher policy intensity (e.g., higher consumption subsidies, stricter green procurement quotas, or more stringent carbon pricing) increases the market attractiveness of low-carbon technologies, prompting more enterprises and consumers to adopt them. Insufficient technology adoption is a key mechanism through which carbon lock-in persists, as limited market demand for low-carbon technologies inhibits their large-scale commercialization and reinforces dependence on mature high-carbon technologies. Edler (2010) systematically develops the theory of demand-based innovation policy (DBIP) [27], highlighting how demand-side instruments stimulate innovation and application by creating stable market opportunities and clear commercialization scenarios. By expanding market demand and providing real-world application contexts for low-carbon technologies, stronger demand-side STI policies drive economies of scale in low-carbon technology production, which reduces production costs and further expands market adoption. This demand-pull mechanism induces carbon unlocking from the demand side by breaking the lock-in of high-carbon technologies rooted in insufficient low-carbon adoption, thus enhancing carbon unlocking efficiency [28].

Complementary STI policies—including green finance support, technology trading platforms, cross-regional innovation collaboration, and supporting infrastructure—focus on building an effective innovation ecosystem by addressing systemic coordination failures that hinder low-carbon technology transformation. Their intensity determines the degree of coordination and synergy in low-carbon innovation activities: higher policy intensity strengthens the connectivity between innovation subjects (enterprises, universities, governments) and optimizes the allocation of scattered innovation resources. Mohnen and Röller (2005) demonstrate that policy complementarity is crucial for improving innovation performance, as coordinated policy mixes can significantly enhance overall effectiveness by eliminating systemic bottlenecks [29]. A core mechanism linking complementary policies to carbon unlocking is their ability to integrate fragmented innovation resources across regions and sectors, thereby alleviating coordination failures in the innovation process—such failures often lead to disjointed technological R&D, transformation, and diffusion, which reinforce carbon lock-in by preventing low-carbon technologies from reaching the market. By removing these bottlenecks in technology transformation and diffusion [30], complementary STI policies ensure that low-carbon technologies developed via supply-side policies and demanded via demand-side policies can smoothly flow through the innovation chain, thus bridging the gap between technology supply and market demand and promoting urban carbon unlocking efficiency.

Finally, institutional reform-oriented STI policies, such as strengthened intellectual property protection, reforms in government performance evaluation, improvements in carbon trading systems, and the establishment of low-carbon standards, focus on optimizing the institutional environment for innovation by dismantling institutional inertia—a deep-rooted mechanism of carbon lock-in. The intensity of these policies directly influences the long-term stability and sustainability of carbon unlocking: higher policy intensity promotes more profound institutional adjustments that align governance incentives and market rules with low-carbon development goals. At a deeper level, carbon lock-in is rooted in institutional inertia—such as GDP-oriented governance incentives and persistent dependence on high-carbon industries—which encourages local governments and enterprises to maintain high-carbon development models despite technological and market changes. This inertia can only be dismantled through targeted institutional change: for example, reforming government performance evaluation to include low-carbon indicators reduces the incentive for GDP-driven high-carbon investment, while improved carbon trading systems internalize the environmental costs of high-carbon activities. Integrating innovation system approaches into institutional design can effectively weaken path dependence by establishing long-term institutional guarantees for low-carbon innovation [31], thereby providing durable institutional support for carbon unlocking and enhancing urban carbon unlocking efficiency.

Based on the above theoretical analysis, this study proposes the following hypothesis:

Hypothesis 1 (H1). 

Greater intensity of heterogeneous STI policies is positively associated with urban carbon unlocking efficiency.

2.2. Mechanisms Through Which Heterogeneous STI Policy Intensity Affects Urban Carbon Unlocking Efficiency

The positive effects of heterogeneous STI policies on urban carbon unlocking efficiency do not operate in isolation; rather, they are significantly shaped by regional development conditions. In particular, regional digital infrastructure development, science and technology finance development, and government green attention strengthen the effectiveness of heterogeneous STI policies through three distinct channels: improving factor mobility, enhancing resource support, and reinforcing institutional enforcement. By improving the external environment in which policies operate, these conditions reduce implementation costs and amplify policy transmission efficiency, enabling heterogeneous STI policies to more precisely target the core mechanisms underlying urban carbon lock-in.

2.2.1. Regional Digital Infrastructure Development: An Amplifier of Policy Transmission Through Enhanced Factor Mobility

Regional digital infrastructure—represented by 5G networks, industrial internet platforms, and data centers—relaxes spatial and temporal constraints, reduces information asymmetries, and improves factor allocation efficiency [32,33]. Through these mechanisms, digital infrastructure significantly amplifies the positive effects of heterogeneous STI policies on urban carbon unlocking efficiency, with its moderating role extending across the entire policy transmission chain. According to Tobler (1970) first law of geography, spatial interaction weakens with distance [34]. Digital infrastructure mitigates this constraint through a “time–space compression” effect, thereby providing a technological foundation for scaling up policy effectiveness. For supply-side STI policies, digital infrastructure magnifies technological supply effects by accelerating knowledge spillovers [35]. As the core objective of supply-side policies is to stimulate low-carbon R&D, digital infrastructure facilitates the formation of virtual R&D communities, reducing the costs of cross-firm and cross-regional collaboration. As a result, the marginal returns to policy instruments such as R&D subsidies are substantially enhanced. Empirical evidence further suggests that digital technologies improve collaborative R&D efficiency and strengthen the contribution of low-carbon technological supply to carbon unlocking [2]. For demand-side STI policies, policy effectiveness critically depends on the breadth of market coverage. Digital infrastructure helps overcome the geographic limitations of traditional demand-side instruments [28], expanding policy reach and alleviating “local market lock-in.” By enlarging market access and accelerating information diffusion [36], digital infrastructure enhances the translation of demand-side policies into effective low-carbon technology adoption. With respect to complementary STI policies, digital infrastructure strengthens policy effectiveness through more precise resource matching [37]. Digital finance platforms improve the efficiency of green credit approval, while the digitalization of technology trading platforms facilitates cross-regional technology transactions [38]. Big data technologies further enhance coordination among policy instruments by enabling more accurate matching between financial resources and innovation needs [8], thereby alleviating problems of financial misallocation and transaction barriers that characterize traditional policy implementation. Finally, for institutional reform-oriented STI policies, digital infrastructure enhances the diffusion and enforcement of regulatory standards. Digital government platforms improve the cross-departmental and cross-regional dissemination of institutional arrangements such as carbon accounting standards and intellectual property protection rules [39], thereby strengthening the consistency and credibility of emission-reduction constraints. Taken together, by optimizing factor flows along the entire policy transmission chain, regional digital infrastructure functions as a key amplifier of the carbon unlocking effects of heterogeneous STI policies. Based on this reasoning, we propose the following hypothesis:

Hypothesis 2a (H2a). 

The level of regional digital infrastructure development is positively associated with the strength of the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency.

2.2.2. Science and Technology Finance Development: Strengthening Financial Support as a Resource Catalyst

The level of science and technology finance development—characterized by green credit, venture capital, and green bonds—provides essential financial support for the effective implementation of heterogeneous STI policies by lowering financing costs, diversifying innovation risks, and guiding capital toward low-carbon innovation activities. Through these mechanisms, science and technology finance significantly strengthens the positive effects of heterogeneous STI policies on urban carbon unlocking efficiency. For supply-side STI policies, R&D subsidies exhibit a “seed capital” nature. This role is consistent with Jaffe et al. (1995), who emphasize that innovation policies require financial instruments to amplify their intervention effects [40]. A well-developed science and technology finance system enhances the leverage effect of supply-side policies through mechanisms such as risk compensation and loan–investment linkage, enabling relatively limited public R&D subsidies to mobilize substantial private R&D investment [41]. This process improves the efficiency of technological transformation and amplifies the contribution of supply-side policies to urban carbon unlocking efficiency. For demand-side STI policies, science and technology finance activates market demand through consumer finance instruments. Financial incentives on the consumption side can substantially expand the market scale for low-carbon products and form synergies with demand-side STI policies [42], ensuring that policy-related financial inputs are more precisely directed toward carbon unlocking processes. With respect to complementary STI policies, the integration of diversified financial instruments enhances policy coordination through more effective resource matching. The science and technology finance system serves as a critical platform for integrating complementary policy tools [8], enabling synergy-enhancing effects through coordinated resource allocation. In particular, the deep integration of green finance instruments with science and technology finance forms a full-cycle financing chain—covering R&D financing, commercialization financing, and scale-up financing—which increases the implementation rate of low-carbon technologies. Moreover, green funds within the science and technology finance system participate in cross-regional innovation platforms through equity investment, mitigating potential “siphoning effects” whereby core cities attract excessive financial resources from neighboring areas. By promoting cross-regional capital allocation, science and technology finance helps strengthen the regional collaborative carbon unlocking effects of complementary STI policies [43]. Finally, science and technology finance reinforces the effectiveness of institutional reform oriented and demand-side STI policies by enhancing enforcement credibility and market activation. Financial innovation strengthens the implementation of institutional rules [44]. In particular, carbon derivatives such as carbon futures and options supported by science and technology finance improve the price discovery function of carbon markets, reduce excessive price volatility, and enhance the stability and binding force of institutional policies. Based on the above analysis, this study proposes the following hypothesis:

Hypothesis 2b (H2b). 

The level of science and technology finance development is positively associated with the strength of the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency.

2.2.3. Government Green Attention: Strengthening Policy Implementation Through Institutional Support

Government green attention—reflected in the weighting of green performance indicators, targeted supervision of green policy implementation, and the establishment of dedicated leadership mechanisms—enhances the effectiveness of heterogeneous STI policies by strengthening enforcement capacity, improving policy coordination, and reducing implementation costs [45,46]. For supply-side STI policies, higher levels of government green attention enhance policy effectiveness by increasing enforcement rigidity. Governments with strong green orientation are more likely to establish dedicated supervision mechanisms for low-carbon R&D subsidies, improving fund disbursement rates and enhancing the emission-reduction effectiveness of public subsidies [47]. Woolthuis et al. (2005) emphasize that policy design must address systemic failures in implementation, and green-attention-driven supervision mechanisms represent targeted solutions that improve the technological conversion efficiency of supply-side policies [48]. For demand-side STI policies, government green attention enhances effectiveness by improving market cultivation and implementation precision. The success of demand-side policies depends heavily on local governments’ promotional efforts and supporting infrastructure provision. Governments with high green attention tend to implement initiatives such as green consumption campaigns and dedicated funds for charging infrastructure, increasing public awareness of policy incentives and enabling dynamic policy adjustment mechanisms. These efforts help avoid “policy fatigue” and enhance the marginal effectiveness of demand-side policies [28], thereby amplifying their contribution to urban carbon unlocking efficiency. With respect to complementary STI policies, government green attention strengthens resource integration through enhanced coordination mechanisms. Complementary policies typically involve multiple departments, including finance, science and technology, and housing and urban development. Governments with strong green orientation are more likely to establish cross-departmental coordination bodies that integrate processes such as green credit approval, technology transaction registration, and infrastructure support into “one-stop” services. This integration shortens project implementation cycles and improves the synergistic effectiveness of complementary STI policies. For institutional reform oriented STI policies, strong government green attention facilitates the implementation of binding constraints, such as carbon reduction veto mechanisms. By strengthening institutional enforcement, government green attention helps dismantle institutional inertia associated with high-carbon development paths [49]. As a result, compliance rates in carbon trading systems increase, and enforcement efficiency in intellectual property protection related to low-carbon technologies improves, significantly enhancing the effectiveness of institutional policies. Overall, by combining rigid enforcement, coordinated governance, and adaptive policy adjustment, government green attention provides critical institutional support for amplifying the carbon unlocking effects of heterogeneous STI policies [50,51]. Accordingly, this study proposes the following hypothesis:

Hypothesis 2c (H2c). 

The level of government green attention is positively associated with the strength of the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency.

2.3. Spatial Effects of Heterogeneous STI Policy Intensity on Urban Carbon Unlocking Efficiency

Tobler (1970) first law of geography states that all things are related, but nearby things are more strongly related than distant ones, implying that spatial spillover effects decay with distance [34]. In the context of digital technologies weakening traditional spatial constraints, the cross-regional mobility of innovation factors such as knowledge and capital has increased substantially [52,53], providing favorable conditions for the spatial transmission of heterogeneous STI policies. By influencing factor mobility costs and alleviating information asymmetries, heterogeneous STI policies affect not only local urban carbon unlocking efficiency but also that of neighboring cities through channels such as factor flows and policy imitation. Importantly, these spatial spillover effects are heterogeneous across different types of STI policies. Supply-side STI policies, including R&D subsidies, basic research support, and talent development programs, exhibit positive spatial spillover effects. Regional policy imitation and coordination play a critical role in enhancing overall policy effectiveness [54]. Through transmission channels such as knowledge sharing, technological imitation, and collaborative innovation, the visible outcomes of supply-side policies—such as the strong linkage between R&D subsidies and patent output—are easily observed by neighboring cities, encouraging policy imitation and promoting regional carbon unlocking. In contrast, demand-side STI policies, such as green procurement, consumption subsidies, and demonstration projects, are constrained by local market lock-in and regional demand heterogeneity. The effectiveness of these policies depends heavily on local consumption structures and infrastructure conditions, making it difficult for their effects to transmit across regions and generate significant spatial spillovers. Complementary STI policies, including green finance support, technology trading platforms, and infrastructure provision, tend to exhibit negative spatial spillover effects. As highlighted by Oughton et al. (2002) in the “regional innovation paradox,” inherent imbalances exist between core and lagging regions in the allocation of innovation resources [55]. Complementary policies rely heavily on scarce factors such as capital and advanced services, which core cities are better positioned to attract due to agglomeration advantages. Consequently, these policies may intensify resource siphoning from neighboring cities, inhibiting their carbon unlocking processes. Finally, institutional reform-oriented STI policies generate positive spatial spillover effects due to the public-good nature of institutions and their high degree of compatibility and imitability. Effective institutional arrangements implemented in core cities often serve as benchmarks that diffuse to neighboring areas through regional cooperation and policy learning, reducing institutional barriers to cross-regional factor mobility. Standardized carbon accounting rules, intellectual property protection frameworks, and carbon trading systems can be directly adopted by neighboring cities, lowering institutional coordination costs. Moreover, institutional diffusion reflects systemic synergy [20], as integrated regional markets amplify the binding force of institutional constraints and enhance regional carbon unlocking efficiency. Based on the above analysis, this study proposes the following hypotheses:

Hypothesis 3a (H3a).  

Supply-side STI policy intensity is positively associated with urban carbon unlocking efficiency in neighboring cities.

Hypothesis 3b (H3b). 

Demand-side STI policy intensity is not significantly associated with urban carbon unlocking efficiency in neighboring cities.

Hypothesis 3c (H3c). 

Complementary STI policy intensity is negatively associated with urban carbon unlocking efficiency in neighboring cities.

Hypothesis 3d (H3d). 

Institutional reform oriented STI policy intensity is positively associated with urban carbon unlocking efficiency in neighboring cities.

The theoretical analysis framework of this study is illustrated in Figure 1.

Figure 1. Theoretical analysis diagram of this study.

3. Research Design

3.1. Empirical Model

To examine the effect of heterogeneous STI policy intensity on urban carbon unlocking efficiency, this study constructs a fixed-effects panel regression model based on prefecture-level city data, as specified below:

$$UC{E}_{it}={\alpha }_0+{\alpha }_{1}Policy\_S_{it}+\lambda contro{l}_{it}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}$$

(1)

$$UC{E}_{it}={\alpha }_0+{\alpha }_{1}Policy\_D_{it}+\lambda contro{l}_{it}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}$$

(2)

$$UC{E}_{it}={\alpha }_0+{\alpha }_{1}Policy\_C_{it}+\lambda contro{l}_{it}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}$$

(3)

$$UC{E}_{it}={\alpha }_0+{\alpha }_{1}Policy\_I_{it}+\lambda contro{l}_{it}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}$$

(4)

$UC{E}_{it}$ is the dependent variable, representing the carbon unlocking efficiency of city $i$ in year $t$. The core explanatory variables $Policy\_S_{it}$, $Policy\_D_{it}$, $Policy\_C_{it}$, and $Policy\_I_{it}$ denote the intensities of heterogeneous STI policies in city $i$ in year $t$, including supply-side policy intensity, demand-side policy intensity, complementary policy intensity, and institutional reform policy intensity, respectively. $contro{l}_{it}$ represents a set of control variables, including financial development ($Fd$), economic development ($Ed$), urbanization level ($Urban$), local government expenditure ($Ge$), per capita fixed-asset investment ($Ifa$), and informatization level ($Id$). ${\delta }_t$ and ${\gamma }_i$ denote year-fixed effects and city-fixed effects, respectively, and ${\epsilon }_{it}~N(0,{\delta }^2)$ is the stochastic error term. By simultaneously controlling for both city and year fixed effects, the coefficient ${\alpha }_1$ captures the marginal effect of heterogeneous STI policy intensity on urban carbon unlocking efficiency.

3.2. Variable Design

3.2.1. Dependent Variables

Urban carbon unlocking efficiency (UCE_it). With regard to urban carbon unlocking efficiency, this study refers to the research of Chen et al. (2024) [2] and adopts the slack-based measure (SBM) directional distance function with undesirable outputs for measurement. The calculation principle is as follows: each city is treated as a decision-making unit (DMU) for production. It is assumed that each city has N types of inputs for carbon unlocking efficiency, mainly including environmental regulation, technological funds, and technical personnel, which can generate M types of desirable outputs and 1 type of undesirable output. The desirable output is the city’s gross regional product (GRP), and the undesirable output is the annual CO₂ emissions. In year t, the environmental production technology of each city that includes both desirable and undesirable outputs is defined as: $P_t(X)=\left\{\left(x_t,y_t,b_t\right)\left|x_{\mathrm{t}}\right.\right.$ can produce $\left.\left(y_t,b_t\right)\right\}$. Referring to the research of Xia and Xu (2020) [37], assuming variable returns to scale (VRS), the non-radial and non-oriented SBM directional distance function for a city in year t, which incorporates both desirable and undesirable outputs, is expressed as:

$$D_V^t\left(x^t,y^t,b^t;g_x,g_y,g_b\right)=\underset{\beta ,s^{-},s^{+},s^b}{\min }{\displaystyle \frac{1-\left[\frac{1}{N}{\displaystyle \sum _{n=1}^N\frac{s_n^{-}}{g_n^x{x}_n^t}}\right]}{1+\left[\frac{1}{M+1}\left({\displaystyle \sum _{m=1}^M\frac{s_m^{+}}{y_m^t{g}_m^y}}+\frac{s^b}{g_b^t{b}^t}\right)\right]}}$$

(5)

$$\left\{\begin{array}{c}s.t.\begin{array}{c}{x}_n^t=X_n^t{z}_n^t+s_n^{-},n=1,2,\dots ,N\end{array}\\ y_m^t=Y_m^t{z}_m^t-s_m^{+},m=1,2,\dots ,M\\ b^t=B^t{z}^t+s^b\\ z^t\ge 0,{\displaystyle {\sum }_{k=1}^K{z}_{k=1}^K=1}\\ s_n^{-}\ge 0,s_m^{+}\ge ,s^b\ge 0\end{array}\right.$$

(6)

Among them, $D_V^t$ denotes the directional distance function under the assumption of variable returns to scale (VRS). $\left(x^t,y^t,b^t\right)$ represent the input vector, desirable output vector, and undesirable output vector of the city, respectively. $\left(g_x,g_y,g_b\right)$ is the directional vector. In this study, we set $g_x=x^t$, $g_y=y^t$, and $g_b=b^t$, meaning that input factors are reduced at their current scale, desirable outputs are expanded at their current scale, and undesirable outputs are reduced at their current scale. $\beta$ is the potential improvement margin of carbon unlocking efficiency and serves as the core optimization target of the objective function; $s^{-}$, $s^{+}$, and $s^b$ are the slack variables for excess inputs, insufficient desirable outputs, and excess undesirable outputs, respectively. These slack variables are only used to characterize the deviation degree between the decision-making unit (DMU) and the production frontier in the constraint conditions and do not directly participate in the optimization of the objective function; $z^t$ is the weight vector of DMUs, which is used to depict the relative importance of different cities in constructing the production frontier and satisfies the weight normalization constraint ${\displaystyle {\sum }_{k=1}^K{z}_{k=1}^K=1}$ (where $K$ is the number of cities in the sample).

To incorporate dynamic changes over time, and following the geometric mean approach, this study constructs the SBM–Malmquist–Luenberger (SBM-ML) index by comparing two adjacent periods, $t$ and $t+1$, as follows:

$$\begin{array}{ccc}{\left(SBM-ML\right)}_t^{t+1}& =& {\left[{\displaystyle \frac{D_V^t\left(x^{t+1},y^{t+1},b^{t+1}\right)}{D_V^t\left(x^t,y^t,b^t\right)}}\times {\displaystyle \frac{D_V^{t+1}\left(x^{t+1},y^{t+1},b^{t+1}\right)}{D_V^{t+1}\left(x^t,y^t,b^t\right)}}\right]}^{1/2}\hfill \\ & =& {\displaystyle \frac{D_V^{t+1}\left(x^{t+1},y^{t+1},b^{t+1}\right)}{D_V^t\left(x^t,y^t,b^t\right)}}\times {\left[{\displaystyle \frac{D_V^t\left(x^{t+1},y^{t+1},b^{t+1}\right)}{D_V^{t+1}\left(x^{t+1},y^{t+1},b^{t+1}\right)}}\times {\displaystyle \frac{D_V^t\left(x^t,y^t,b^t\right)}{D_V^{t+1}\left(x^t,y^t,b^t\right)}}\right]}^{1/2}\hfill \\ & =& E_{c_t^{t+1}}\times T_{c_t^{t+1}}\hfill \end{array}$$

(7)

Based on the above estimation framework, inputs include environmental regulation and technological inputs. Environmental regulation is proxied by a text-based measure derived from the frequency of environment-related keywords in prefecture-level government work reports. Technological inputs mainly consist of the number of technical personnel and expenditures on science and technology at the city level. In terms of outputs, both desirable and undesirable outputs are considered. Desirable output is measured by each city’s gross domestic product (GDP), while undesirable output is proxied by urban CO₂ emission intensity.

To precisely define the economic implications of “carbon-unlocking efficiency,” we have structured its measurement logic by clearly specifying its input and output dimensions. An input-output correspondence table (now presented as Table 1 in the revised manuscript) has been constructed to visually outline the core components and measurement logic of the UCE index.

Table 1. Input-Output Indicators for Measuring Carbon Unlocking Efficiency.

Dimension	Specific Indicator	Definition	Economic Significance
Input Variables	Environmental Regulation Input	Measured based on the word frequency of environmental regulation terms in prefectural-city government work reports.	Represents the “institutional input” that constrains and guides carbon unlocking, forming the institutional foundation for breaking carbon lock-in.
	Input of technical personnel	Proportion of R&D personnel in industrial enterprises above a designated size to the total urban employment.	Embodies the “human capital” supporting carbon unlocking, serving as the core element for low-carbon technology innovation and diffusion.
	technical investment	Proportion of internal R&D expenditure of industrial enterprises above a designated size to the city’s Gross Regional Product (GRP).	Represents the “capital investment” enabling carbon unlocking, directly impacting the processes of low-carbon technology R&D, piloting, and industrialization.
Desirable Output	expected output	Proportion of the city’s GRP to the provincial GRP.	Reflects the “quality of economic growth” during the carbon-unlocking process, highlighting the synergy between low-carbon transition and economic development.
	unintended output	Proportion of the city’s annual CO2 emissions to its GRP.	Reflects the “carbon emission reduction” achieved during the carbon-unlocking process, representing the attainment of the core decarbonization objective.

Input Dimension—The “Foundational Support System”: This dimension focuses on the key resources required to break carbon lock-in, providing the necessary institutional guarantees and technological foundations. Institutional input, via environmental regulation, establishes emission reduction constraints and policy direction to overcome the “institutional inertia of high-carbon pathways.” Technological input, through human and financial capital, addresses the bottleneck of “insufficient low-carbon technology supply.” Together, they form an indispensable resource base for carbon unlocking.

Output Dimension—The “Objective Achievement System”: This dimension assesses the attainment of core carbon-unlocking goals, adhering to a “win-win for economy and environment” logic for low-carbon transition. The desirable output (economic share) demonstrates that carbon unlocking does not come at the cost of economic growth but rather promotes high-quality development through innovation and optimized resource allocation. The undesirable output (carbon intensity) captures the core emission reduction target. Their synergistic optimization is a key feature distinguishing carbon unlocking from traditional “degrowth-based emission reduction.” Therefore, UCE is conceived as a comprehensive efficiency metric that evaluates how effectively a city translates its foundational resource commitments (inputs) into balanced economic-environmental outcomes (outputs), offering a more holistic and policy-relevant assessment than indicators focusing solely on outcomes or single ratios.

As the carbon unlocking efficiency index obtained from the SBM–ML approach reflects a growth rate—capturing year-on-year changes relative to the previous period—we follow Zhang and Wang (2020) to transform it into a level-based index [56]. Specifically, 2009 is set as the base year with a value of 1. The carbon unlocking efficiency for 2010 is calculated as the base-year value multiplied by the 2010 efficiency index. The value for 2011 is obtained by multiplying the 2010 level by the 2011 efficiency index, and this process is iteratively applied to subsequent years, thereby yielding the level of urban carbon unlocking efficiency for each city over time.

Figure 2 reports the spatial distribution of urban carbon unlocking efficiency (UCE) in 2010, 2014, 2019, and 2023. From a spatial perspective, UCE exhibits a pronounced east–west disparity, characterized by a clear “high-in-the-east, low-in-the-west” pattern. High-efficiency clusters (darker shades) are persistently concentrated in eastern coastal regions and core cities in central China, whereas most western regions remain in low-efficiency areas (lighter shades or missing values). This spatial configuration closely corresponds to regional differences in economic foundations and innovation resource endowments. The observed disparity likely reflects the uneven distribution of the foundational inputs defined in our framework: eastern regions typically benefit from stronger institutional governance, denser concentrations of technical talent, and more robust R&D funding, which collectively enhance their efficiency in transforming inputs into carbon-unlocking outcomes.

Figure 2. Spatial distribution of urban carbon unlocking efficiency in 2010, 2014, 2019, and 2023.

From a temporal perspective, the spatial coverage of high UCE values shows a continuous expansion over time. In 2010, high-efficiency areas were sporadically distributed in parts of eastern China. By 2014, these areas had expanded across most eastern cities and parts of central China, accompanied by an increased density of cities within the high-efficiency interval (6.55–10.52). In 2019, several cities in Northeast China entered the high-efficiency range, and high-value clusters in central China became more contiguous. By 2023, the agglomeration of high UCE values in eastern and central core cities became more pronounced, while some cities in central and western China also experienced efficiency improvements. Nevertheless, western regions as a whole remained at relatively low efficiency levels. Overall, during the period 2010–2023, urban carbon unlocking efficiency in China exhibited a trend of overall improvement, spatial diffusion, yet persistent regional disparities. The expansion of high-efficiency areas reflects the regional transmission of carbon unlocking capabilities, whereas the lagging performance of western regions highlights the constraining role of uneven regional development foundations in the carbon unlocking process.

3.2.2. Core Explanatory Variables

The core explanatory variables are the intensities of heterogeneous STI policies, denoted as, $Policy\_D$, $Policy\_C$ and $Policy\_I$. Given that innovation policies differ substantially in their functional orientation, existing studies have proposed several classification schemes. Early research distinguishes innovation policies into supply-side, demand-side, and environmental policies [57]. Subsequent studies classify innovation policies according to their mode of innovation, such as incremental, radical, autonomous, and adaptive policies [58]. Barrow (1996) further extends this framework by introducing complementary policies that emphasize industry–university collaboration and the mobilization of firms’ internal and external resources [59]. Building on these strands, Stcinmucller (2010) incorporates institutional reform oriented policies that provide stable institutional support and promote the commercialization of scientific and technological outcomes [60]. Following the classification framework proposed by Stcinmucller (2010) [60], this study categorizes STI policies into four types: supply-side policies, demand-side policies, complementary policies, and institutional reform policies. Specifically, supply-side STI policies aim to address underinvestment and bottlenecks in the innovation process by directly strengthening scientific research and technological development through instruments such as subsidies and dedicated funding programs. These policies primarily target constraints related to insufficient R&D investment and impeded technology diffusion. Demand-side STI policies emphasize demand-driven innovation by stimulating market demand for new technologies and providing stronger expected returns, thereby incentivizing firms to increase innovation efforts. Such policies focus on aligning technological innovation with market needs and accelerating technology adoption. Complementary STI policies seek to expand the availability of resources for innovation activities by enhancing coordination among financial, technological, and infrastructural elements, thereby reducing innovation bottlenecks and improving overall innovation efficiency. Finally, institutional reform-oriented STI policies focus on optimizing the institutional environment for innovation by clarifying the division of responsibilities in R&D activities, strengthening incentives for firms’ technological innovation, and promoting the construction of platforms that facilitate the transformation and commercialization of scientific and technological achievements.

To measure the strength of these categorized policies at the city level, this study employs the annual count of newly issued policy documents within each prefecture-level city as the core proxy variable. This measurement approach, while widely used in policy analysis literature [16] for its feasibility and comparability across regions and over time, is acknowledged as a proxy that primarily captures the breadth and frequency of governmental activity within a policy domain. This approach is predicated on the rationale that the frequency and density of policy promulgation serve as a visible and comparable indicator of the priority and administrative emphasis a local government places on fostering innovation within a specific domain. By applying the aforementioned four-type classification, we refine the simple aggregate count and construct separate intensity variables for each policy category (e.g., supply-side policy intensity). This allows the measurement to capture variation in governmental focus across different policy instruments. Consequently, these variables primarily operationalize the degree of governmental attention or priority allocated to different STI policy portfolios, reflecting the breadth of policy activity rather than assuming homogeneity in the fiscal scale or absolute impact of individual policies. This focus on policy attention aligns with the study’s primary aim of investigating whether and how different directions of policy emphasis (supply-push, demand-pull, etc.) affect carbon unlocking efficiency, rather than precisely quantifying the return on investment of specific budgetary outlays. This methodological choice is necessitated by the limited public availability of consistent, granular implementation data (e.g., detailed budgetary allocations) across all Chinese prefectural cities for the studied period, while aiming to ensure cross-sectional and longitudinal comparability for empirical analysis. Potential refinements using weighted intensity indicators based on implementation-level data are discussed as a valuable direction for future research in the Limitations section.

3.2.3. Control Variables

Urban carbon unlocking efficiency may be influenced not only by heterogeneous STI policy intensity but also by a range of city-level characteristics. Following Wu et al. (2023) [61], this study includes a set of control variables to account for potential confounding factors, namely financial development ($Fd$), economic development ($Ed$), urbanization level ($Urban$), local government expenditure ($Ge$), per capita fixed-asset investment ($Ifa$), and informatization level ($Id$).

Specifically, financial development ($Fd$) is measured by the ratio of year-end deposits and loans of financial institutions to GDP. Economic development ($Ed$) is proxied by GDP per capita (in 10,000 yuan). Urbanization level ($Urban$) is measured as the proportion of urban population to total population. Local government expenditure ($Ge$) is captured by the ratio of general public budget expenditure to GDP. Per capita fixed-asset investment ($Ifa$) is measured by the ratio of total fixed-asset investment to total population. Informatization level ($Id$) is proxied by the logarithm of the number of Internet users in each city. Detailed definitions of all variables are reported in Table 2.

Table 2. Variable Definitions.

	Variable Name	Symbol	Definition
Dependent variable	Urban carbon unlocking efficiency	$UCE$	Measured using the non-radial SBM directional distance function with undesirable outputs
Core explanatory variables	Supply-side STI policy intensity	$Policy\_S$	Number of policies related to supply-side STI policies at the city level
	Demand-side STI policy intensity	$Policy\_D$	Number of policies related to demand-side STI policies at the city level
	Complementary STI policy intensity	$Policy\_C$	Number of policies related to complementary STI policies at the city level
	Institutional reform oriented STI policy intensity	$Policy\_I$	Number of policies related to institutional reform oriented STI policies at the city level
Control variables	Financial development	$Fd$	Ratio of year-end deposits and loans of financial institutions to GDP
	Economic development	$Ed$	Logarithm of GDP per capita
	Urbanization level	$Urban$	Ratio of urban population to total population
	Local government expenditure	$Ge$	Ratio of general public budget expenditure to GDP
	Per capita fixed-asset investment	$Ifa$	Ratio of total fixed-asset investment to total population
	Informatization level	$Id$	Logarithm of the number of Internet users

3.3. Data Sources

The empirical analysis is based on an unbalanced panel dataset covering 286 prefecture-level cities in China over the period 2009–2023. Data for the dependent variable—urban carbon unlocking efficiency—and the control variables, including financial development, economic development, urbanization level, local government expenditure, per capita fixed-asset investment, and informatization level, are primarily obtained from the China City Statistical Yearbook. Data on the core explanatory variables—heterogeneous STI policy intensity—are manually collected from the official websites of prefecture-level municipal governments and their science and technology departments, as well as from policy databases such as Peking University’s Law Database. The classification of STI policies follows the framework proposed by Stcinmucller (2010) [60], which distinguishes supply-side, demand-side, complementary, and institutional reform oriented policies. For a small number of cities and years with missing observations, interpolation methods are applied to ensure data continuity. After data processing, the final sample consists of 4290 city–year observations. Descriptive statistics for all variables are reported in Table 3.

Table 3. Descriptive statistics.

	Variable	Obs	Mean	SD	Min	Max
Dependent variable	$UCE$	4290	1.2270	1.4310	0.1070	10.5200
Independent variable	$Policy\_S$	4290	1.6490	1.9710	0.0000	24.0000
	$Policy\_D$	4290	0.8710	1.3810	0.0000	23.0000
	$Policy\_C$	4290	1.1490	1.4420	0.0000	22.0000
	$Policy\_I$	4290	1.0580	1.4330	0.0000	16.0000
Control variables	$Fd$	4290	2.5600	1.2000	0.9900	6.9990
	$Ed$	4290	10.7200	0.6140	9.2320	12.0900
	$Urban$	4290	0.3950	0.2100	0.1070	0.9960
	$Ge$	4290	0.1990	0.0985	0.0743	0.6060
	$Ifa$	4290	0.2640	0.2560	0.0294	1.7590
	$Id$	4290	13.4400	1.0400	9.2100	17.7600

4. Empirical Analysis

4.1. Baseline Results

Table 4 reports the baseline regression results examining the effects of heterogeneous STI policy intensity on urban carbon unlocking efficiency. Columns (1)–(5) present the estimated effects of supply-side, demand-side, complementary, and institutional reform oriented STI policies, as well as their joint effects when all policy types are included simultaneously. In terms of model specification, the Hausman specification test yields statistically significant results, rejecting the assumption of no correlation between individual effects and explanatory variables. Therefore, the fixed effects model is preferred over the random effects model. The results in column (1) show that supply-side STI policy intensity ($Policy\_S$) has a significantly positive effect on urban carbon unlocking efficiency ($UCE$) at the 1% significance level, with an estimated coefficient of 0.0749. This implies that a one-unit increase in supply-side policy intensity is associated with an approximately 7.49% increase in urban carbon unlocking efficiency. Column (2) indicates that demand-side STI policy intensity ($Policy\_D$) also exerts a significantly positive impact on $UCE$ at the 1% level, with a coefficient of 0.0885, suggesting a stronger marginal effect relative to supply-side policies. Column (3) shows that complementary STI policy intensity ($Policy\_C$) has a positive effect on $UCE$ at the 5% significance level, with an estimated coefficient of 0.0609. Similarly, column (4) reveals that institutional reform oriented STI policy intensity ($Policy\_I$) is positively associated with $UCE$ at the 5% level, with a coefficient of 0.0929.

Table 4. Baseline regression results of heterogeneous STI policy intensity on urban carbon unlocking efficiency.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)	(5)
$Policy\_S$	0.0749 *** (4.405)				0.0467 ** (2.335)
$Policy\_D$		0.0885 *** (3.519)			0.0466 * (1.713)
$Policy\_C$			0.0609 ** (2.532)		0.0010 (0.040)
$Policy\_I$				0.0929 *** (3.418)	0.0315 (1.000)
$\_cons$	2.0664 (1.304)	2.0716 (1.287)	2.3856 (1.477)	2.3400 (1.458)	1.7888 (1.124)
Hausman test	41.150 *** (0.000)	40.290 *** (0.000)	45.330 *** (0.000)	39.080 *** (0.000)	54.930 *** (0.000)
Control Variables	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
Observations	4290	4290	4290	4290	4290
R2	0.4258	0.4257	0.4243	0.4254	0.4268

Note: Adjusted robust standard errors for city-level and year-level clustering are in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1; this note applies to all subsequent tables.

When all four types of STI policies are included simultaneously in column (5), only supply-side ($Policy\_S$) and demand-side ($Policy\_D$) policy intensities remain significantly positive, whereas the coefficients on complementary ($Policy\_C$) and institutional reform oriented ($Policy\_I$) policies become statistically insignificant. This pattern suggests that supply-side and demand-side policies are directly and positively associated with urban carbon unlocking efficiency, possibly through the mechanisms of strengthening low-carbon technology supply and stimulating market demand. Specifically, supply-side policies reduce R&D costs and accelerate technological breakthroughs, while demand-side policies expand market demand through instruments such as green procurement, consumption subsidies, and carbon pricing, thereby facilitating the diffusion and adoption of low-carbon technologies. In contrast, complementary and institutional reform oriented STI policies tend to operate through longer-term and indirect channels. Complementary policies primarily focus on building supporting service systems and innovation ecosystems, while institutional reform policies emphasize rule-setting and long-term incentive alignment. Their effects may require coordination between both supply and demand sides to materialize fully. As a result, when these policies are included alongside direct supply- and demand-side instruments, their immediate effects on urban carbon unlocking efficiency are less likely to be captured in the baseline specification.

From a deeper economic perspective, the stronger marginal effect of institutional reform policies in single-variable regression reflects the core role of institutional design in breaking carbon lock-in. China’s long-standing “GDP-first” performance appraisal and high-carbon industrial path dependence can hardly be resolved by short-term technological or market incentives alone. Institutional reform policies reshape the behavioral logic of local governments and enterprises through intellectual property protection, carbon trading rules, and green performance assessments, providing long-term institutional guarantees for carbon unlocking—this “bottom-line rule reconstruction” is more effective than temporary policy stimuli. The insignificance of complementary and institutional reform policies in joint regression also reveals the sequential nature of policy effects: complementary policies rely on the foundation of supply and demand policies, and institutional policies have a time lag in effect transmission, which is consistent with the realistic logic of China’s low-carbon transition.

From an international comparison perspective, the effect characteristics of China’s four types of policies share both commonalities and differences with developed countries. The commonality lies in the core guarantee role of institutional policies: the EU’s Carbon Border Adjustment Mechanism (CBAM) and the U.S. Inflation Reduction Act (IRA) both rely on institutional design to advance low-carbon transition, echoing the strong marginal effect of China’s institutional reform policies. The difference is reflected in the prominent role of supply-side policies: as a developing country, China still needs to strengthen independent innovation in low-carbon technologies, so supply-side policies are more necessary to break technological lock-in. In contrast, developed countries such as the EU and the U.S. focus more on demand-side regulation, which provides a reference for China to adjust its policy focus according to its development stage.

Overall, the baseline regression results provide strong empirical support for Hypothesis H1, confirming that greater intensity of heterogeneous STI policies—particularly supply-side and demand-side policies—is significantly positively associated with urban carbon unlocking efficiency.

4.2. Endogeneity Treatment

The estimated relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency may be subject to potential endogeneity concerns. On the one hand, unobservable factors—such as local governance capacity or latent innovation culture—may simultaneously affect both STI policy intensity and urban carbon unlocking efficiency, leading to omitted variable bias. Although this study controls for a rich set of macroeconomic and city-level characteristics, the possibility of residual omitted variables cannot be fully ruled out. On the other hand, reverse causality may also arise: cities with higher carbon unlocking efficiency may be more likely to attract or implement stronger heterogeneous STI policies. While the baseline regression results indicate that heterogeneous STI policy intensity significantly enhances urban carbon unlocking efficiency, improvements in carbon unlocking efficiency may in turn feed back into policy formulation and intensity. To address these potential endogeneity issues, this study adopts an instrumental variable (IV) approach, which is widely used in the literature to mitigate endogeneity concerns [62]. The specific estimator employed is the Two-Stage Least Squares (2SLS) method, which provides consistent estimates by isolating the exogenous variation in the policy variables through the selected instruments. The implementation of the 2SLS algorithm in our study proceeds in two distinct stages: First-Stage Regression: The endogenous explanatory variables (i.e., the intensities of the heterogeneous STI policies) are regressed on the selected instrumental variables—specifically, the one-period lagged policy intensities and the interaction term between regional university student enrollment and a time trend—alongside the full set of control variables. This ordinary least squares (OLS) regression yields the predicted values for the policy intensity variables. Second-Stage Regression: The original endogenous policy variables in the main equation are replaced by their predicted values from the first stage. This revised specification, which includes the first-stage fitted values and the control variables, is then estimated via OLS to obtain the consistent coefficients for the effect of policy intensity on urban carbon unlocking efficiency.

4.2.1. Lagged Instrumental Variable Strategy

Following existing studies, lagged values of the explanatory variables are employed as instruments under the panel data setting [63]. The underlying rationale is that heterogeneous STI policy intensity exhibits a certain degree of persistence over time, such that policy intensity in the previous period is highly correlated with current policy intensity, while being plausibly exogenous to contemporaneous shocks to urban carbon unlocking efficiency. Accordingly, the one-period lag of heterogeneous STI policy intensity is used as the instrumental variable. This strategy has been widely adopted in recent empirical studies [64]. The corresponding IV estimation results are reported in Table 5. Diagnostic tests indicate that the Kleibergen–Paap rk LM statistic and the Cragg–Donald Wald F statistic strongly reject the presence of under-identification and weak instruments across all model specifications, confirming the validity and relevance of the selected instruments. After instrumenting heterogeneous STI policy intensity with its lagged value, the estimated effects of supply-side STI policy intensity $Policy\_S$) and demand-side STI policy intensity ($Policy\_D$) on urban carbon unlocking efficiency ($UCE$) remain significantly positive. In contrast, the effects of complementary STI policy intensity ($Policy\_C$) and institutional reform oriented STI policy intensity ($Policy\_I$) remain statistically insignificant, consistent with the baseline regression results. Overall, these findings suggest that the main conclusions of this study are robust after accounting for potential endogeneity.

Table 5. Endogeneity treatment: lagged variables as instruments.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
$Policy\_S$	0.1253 *** (4.366)
$Policy\_D$		0.1482 *** (3.246)
$Policy\_C$			0.1578 *** (3.273)
$Policy\_I$				0.1416 *** (3.208)
$\_cons$	2.1906 (1.362)	2.1450 (1.321)	2.4295 (1.492)	2.6911 * (1.654)
Kleibergen Paap rk LM statistic	67.670 *** [0.000]	20.875 *** [0.000]	32.084 *** [0.000]	49.491 *** [0.000]
Cragg Donald Wald F statistic	2184.342 {16.38}	2427.823 {16.38}	1577.434 {16.38}	2813.794 {16.38}
Control Variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	4004	4004	4004	4004
R2	0.4535	0.4530	0.4507	0.4535

Notes: (1) * and *** denote statistical significance at the 10% and 1% levels, respectively. (2) Robust standard errors are reported in parentheses. (3) The under-identification test is based on the Kleibergen–Paap rk LM statistic; p-values for the Chi-square (2) test are reported in square brackets [ ]. (4) The weak-instrument test is conducted using the Cragg–Donald Wald F statistic; the Stock–Yogo critical value at the 10% significance level is reported in curly brackets { }. The same notation applies to the following tables.

4.2.2. Alternative Instrumental Variable Strategy Based on University Student Scale

After using lagged heterogeneous STI policy intensity as an instrumental variable, this study further adopts an alternative identification strategy by employing the scale of university students in the region where firms are located as an instrumental variable, following Xu et al. (2022) [19]. The rationale for this instrument is twofold. First, regions with a higher concentration of universities tend to be focal areas for government innovation resource allocation. Such regions are more likely to introduce supporting supply-side policies—such as R&D subsidies and talent incentive programs—and to attract firms through abundant human capital, thereby facilitating the formulation and implementation of demand-side STI policies. As a result, the scale of university students is strongly correlated with the intensity of heterogeneous STI policies, satisfying the relevance condition of a valid instrument. Second, the scale of university students is primarily determined by historical patterns of educational resource allocation and higher-education enrollment planning. These factors are plausibly exogenous to contemporaneous shocks affecting urban carbon unlocking efficiency and are unlikely to influence carbon unlocking outcomes through channels other than STI policy intensity. Therefore, this instrument satisfies the exclusion restriction. Accordingly, this study constructs the instrumental variable as the interaction between the number of enrolled university students in the city and a linear time trend. The corresponding estimation results are reported in Table 6. Diagnostic tests based on the Kleibergen–Paap rk LM statistic and the Cragg–Donald Wald F statistic indicate that none of the model specifications suffer from under-identification or weak instrument problems, confirming the validity of the chosen instrument. After instrumenting heterogeneous STI policy intensity with the regional university student scale, the estimated effects of heterogeneous STI policies on urban carbon unlocking efficiency remain consistent with the baseline regression results. This finding suggests that the main conclusions of this study are robust after further addressing potential endogeneity concerns. It is noteworthy that the estimated coefficients using this instrument are generally larger in magnitude than those obtained with the lagged policy instrument (see Table 5 and Table 6). This difference can be attributed to the distinct sources of exogenous variation each instrument captures. The lagged policy variable primarily isolates variation due to policy inertia, while the student-scale instrument captures variation linked to broader, long-term regional innovation capacities and human capital foundations, which may encompass a wider set of complementary factors that amplify the policy effect. Critically, both strategies yield coefficient estimates that are positive and statistically significant, reinforcing the core finding of a positive association.

Table 6. Endogeneity treatment: regional university student scale as an instrument.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
$Policy\_S$	0.5627 *** (3.965)
$Policy\_D$		0.8087 *** (4.469)
$Policy\_C$			0.7482 *** (3.699)
$Policy\_I$				0.7071 *** (4.217)
$\_cons$	−1.9938 (−1.151)	−3.2978 * (−1.775)	−1.7738 (−1.033)	−0.4417 (−0.283)
Kleibergen Paap rk LM statistic	23.321 *** [0.000]	41.284 *** [0.000]	21.151 *** [0.000]	30.584 *** [0.000]
Cragg Donald Wald F statistic	190.716 {16.38}	134.892 {16.38}	188.611 {16.38}	222.168 {16.38}
Control Variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	4290	4290	4290	4290
R2	0.3230	0.2743	0.3076	0.3361

Notes: *** p < 0.01, * p < 0.1.

4.3. Robustness Checks

To verify the robustness of the baseline regression results, this study conducts a series of robustness checks using five alternative approaches, including replacing the measurement of the dependent variable, adopting alternative regression specifications, excluding directly administered municipalities, controlling for additional policy shocks, and accounting for public health emergencies. This subsection reports the results from replacing the measurement of the dependent variable.

4.3.1. Alternative Measurement of Urban Carbon Unlocking Efficiency

In the baseline analysis, urban carbon unlocking efficiency is calculated using the SBM approach, while environmental regulation—one of the key input variables—is proxied by a text-based measure derived from the frequency of environment-related terms in government work reports. To further enhance measurement robustness and mitigate potential bias arising from text-based indicators, this study employs an alternative measure of environmental regulation based on data from the Environmental Penalty Research Database (EPRD), which records officially disclosed environmental penalties at the city level. Specifically, environmental penalty data from the EPRD are used to reconstruct the environmental regulation input, and the SBM–GML model is subsequently applied to re-estimate urban carbon unlocking efficiency. Based on this alternative measurement, the baseline regression model is re-estimated to examine the effects of heterogeneous STI policy intensity on urban carbon unlocking efficiency. The corresponding results are reported in Table 7. The results indicate that, after replacing the measurement of urban carbon unlocking efficiency, supply-side STI policy intensity ($Policy\_S$) and demand-side STI policy intensity ($Policy\_D$) continue to exhibit significantly positive effects on urban carbon unlocking efficiency. In contrast, the estimated effects of complementary STI policy intensity ($Policy\_C$) and institutional reform oriented STI policy intensity ($Policy\_I$) remain statistically insignificant. These findings are fully consistent with the baseline regression results, providing strong evidence for the robustness of the main conclusions.

Table 7. Robustness Checks: Alternative measurement of urban carbon unlocking efficiency.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
$Policy\_S$	0.0262 * (1.823)
$Policy\_D$		0.0674 *** (3.635)
$Policy\_C$			0.0369 * (1.854)
$Policy\_I$				0.0640 *** (2.789)
$\_cons$	−6.2530 *** (−4.262)	−6.5581 *** (−4.419)	−6.2512 *** (−4.217)	−6.3186 *** (−4.266)
Control Variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	4290	4290	4290	4290
R2	0.5686	0.5697	0.5687	0.5693

Notes: *** p < 0.01, * p < 0.1.

4.3.2. Alternative Regression Model

Given the distributional characteristics of urban carbon unlocking efficiency, the dependent variable may be subject to censoring at the lower bound, which could lead to biased estimates under linear regression. This is because UCE, measured as an efficiency score, is naturally bounded below (theoretically non-negative), and in practice, a non-trivial proportion of cities in our sample cluster near the lower end of the distribution—a pattern consistent with left-censored data. To address this concern, and following Long et al. (2024) [65], this study adopts an alternative estimation strategy by re-estimating the baseline specification using a Tobit model. The estimation results are reported in Table 8. Consistent with the baseline regression findings, supply-side STI policy intensity ($Policy\_S$) and demand-side STI policy intensity ($Policy\_D$) continue to exhibit significantly positive effects on urban carbon unlocking efficiency ($UCE$). In contrast, the estimated coefficients of complementary STI policy intensity ($Policy\_C$) and institutional reform oriented STI policy intensity ($Policy\_I$) remain statistically insignificant, which is also in line with the baseline results. The close agreement between the Tobit estimates and the baseline OLS results indicates that potential censoring in the efficiency measure does not materially alter the inferred relationships. Overall, the consistency between the Tobit estimates and the baseline regression results indicates that the main conclusions are not sensitive to the choice of estimation method, thereby providing further support for the robustness of the empirical findings.

Table 8. Robustness Checks: Alternative regression model.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
$Policy\_S$	0.0749 *** (4.256)
$Policy\_D$		0.0885 *** (4.132)
$Policy\_C$			0.0609 *** (2.615)
$Policy\_I$				0.0929 *** (3.908)
$\_cons$	1.4823 (0.993)	1.4346 (0.960)	1.7074 (1.143)	1.6930 (1.136)
Control Variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	4290	4290	4290	4290
Log	−6434.7106	−6435.2292	−6440.3325	−6436.1269

Note: *** p < 0.01.

4.3.3. Alternative Sample

In the baseline analysis, the sample includes all prefecture-level cities, including the four centrally administered municipalities—Beijing, Tianjin, Shanghai, and Chongqing. However, compared with other prefecture-level cities, these municipalities exhibit substantial advantages in terms of economic scale, population size, land area, and administrative authority [48]. Including them in the full sample may therefore introduce potential bias in estimating the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency. This exclusion follows explicit and reproducible criteria, specifically removing observations associated with the administrative division codes 110,000, 120,000, 310,000, and 500,000. This refinement ensures the sample focuses on comparable administrative units, enhancing the comparability and focus of the analysis. To address this concern, this study excludes the four centrally administered municipalities from the sample and re-estimates the baseline regression model. Consequently, the sample size is reduced from the baseline of 4290 observations (286 cities × 15 years) to 4230 observations (282 cities × 15 years). The corresponding results are reported in Table 9. The estimation results show that heterogeneous STI policy intensity continues to exert a significantly positive effect on urban carbon unlocking efficiency ($UCE$) after excluding Beijing, Tianjin, Shanghai, and Chongqing. Moreover, the estimated coefficients remain qualitatively consistent with those obtained from the baseline regressions. Overall, these findings suggest that the main conclusions are not driven by the inclusion of centrally administered municipalities, providing further evidence for the robustness of the empirical results.

Table 9. Robustness Checks: Alternative sample.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
$Policy\_S$	0.1034 *** (4.993)
$Policy\_D$		0.1782 *** (5.184)
$Policy\_C$			0.1104 *** (3.009)
$Policy\_I$				0.1557 *** (4.247)
$\_cons$	2.1185 (1.325)	2.0222 (1.251)	2.4607 (1.509)	2.4296 (1.497)
Control Variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	4230	4230	4230	4230
R2	0.4262	0.4275	0.4244	0.4265

Note: *** p < 0.01.

4.3.4. Excluding the Influence of Other Policy Interventions

Since 2008, China has successively launched pilot programs for innovation-oriented cities, and by 2022, a total of 103 cities (districts) had been approved as pilot areas [20]. These pilot cities typically receive preferential policy support and greater resource allocation, which may independently enhance urban carbon unlocking efficiency through pilot-related policy dividends rather than through general STI policies. As a result, failing to account for innovation-city pilot programs may confound the estimated effects of heterogeneous STI policy intensity. To isolate the effect of the baseline heterogeneous STI policies and mitigate this potential confounding factor, cities officially designated as National Innovative City Pilots were identified based on the official roster published by the Ministry of Science and Technology and subsequently excluded from this robustness check. To address this concern and isolate the effect of baseline STI policies, this study further controls for the potential influence of innovation-oriented city pilot programs. Specifically, cities designated as innovation-oriented pilots are excluded from the sample, and the baseline regression model is re-estimated. This exclusion results in a sample comprising only non-pilot cities for this specific test. The final sample size for this regression is 4230 observations, which is consistent with the sample after excluding municipalities, as the pilot city list and the four municipalities are mutually exclusive for this robustness check. The corresponding results are reported in Table 10. The estimation results indicate that, after controlling for innovation-oriented city pilot policies, the estimated effects of heterogeneous STI policy intensity on urban carbon unlocking efficiency remain consistent with the baseline regression results. This finding suggests that the main conclusions are not driven by the presence of innovation-oriented city pilot programs, thereby providing further evidence for the robustness of the empirical findings.

Table 10. Robustness Checks: Controlling for innovation-oriented city pilot programs.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
$Policy\_S$	0.0769 *** (4.480)
$Policy\_D$		0.0901 *** (3.531)
$Policy\_C$			0.0635 *** (2.609)
$Policy\_I$				0.0956 *** (3.490)
$\_cons$	2.4343 (1.535)	2.4344 (1.512)	2.7429 * (1.697)	2.7110 * (1.689)
Control variables	YES	YES	YES	YES
Innovation-oriented city pilots	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	4230	4230	4230	4230
R2	0.4266	0.4264	0.4250	0.4262

Notes: *** p < 0.01, * p < 0.1.

4.3.5. Excluding Years Affected by Major Exogenous Events

Studies examining the effects of innovation policies on urban carbon unlocking efficiency inevitably involve the period affected by the COVID-19 pandemic. However, pandemic-related shocks may introduce external disturbances to the estimated relationship between heterogeneous STI policies and urban carbon unlocking efficiency. On the one hand, containment measures implemented during the pandemic—such as production shutdowns, mobility restrictions, and traffic controls—may have temporarily reduced energy consumption and carbon emissions, leading to an artificial increase in measured carbon unlocking efficiency and potentially masking the true effects of STI policies. On the other hand, governments introduced temporary countercyclical measures during the pandemic, including short-term fiscal stimulus and green recovery policies, which may have interacted with existing STI policies and biased the evaluation of their effectiveness. Therefore, to assess the policy effects under typical socioeconomic conditions and to prevent these extraordinary temporary distortions from driving the results, the years 2020 and 2021 are excluded in this specific robustness test. To ensure the accuracy and reliability of the empirical results, this study excludes the pandemic years 2020 and 2021 from the sample and re-estimates the baseline regression model. This adjustment reduces the temporal coverage to 13 years (2009–2019 and 2022–2023). Based on the full set of 286 cities, the sample size is therefore reduced from 4290 observations to 3718 observations (286 cities × 13 years). The corresponding results are reported in Table 11. The estimation results show that, after excluding pandemic-affected years, the effects of heterogeneous STI policies—namely $Policy\_S$, $Policy\_D$, $Policy\_C$ and $Policy\_I$—on urban carbon unlocking efficiency ($UCE$) remain consistent with those obtained from the baseline analysis. Overall, these findings indicate that the main conclusions of this study are not driven by pandemic-related shocks, thereby further confirming the robustness of the empirical results.

Table 11. Robustness Checks: Excluding years affected by major exogenous events.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
$Policy\_S$	0.0826 *** (3.939)
$Policy\_D$		0.0842 *** (2.963)
$Policy\_C$			0.0743 ** (2.362)
$Policy\_I$				0.0952 *** (2.809)
$\_cons$	−0.2047 (−0.120)	−0.1140 (−0.066)	0.0376 (0.022)	0.1294 (0.075)
Control variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	3718	3718	3718	3718
R2	0.3862	0.3856	0.3845	0.3855

Notes: *** p < 0.01, ** p < 0.05.

4.4. Heterogeneity Analysis

Baseline regression results indicate that increasing the intensity of heterogeneous STI policies significantly enhances urban carbon unlocking efficiency. However, China exhibits substantial regional disparities in economic foundations, pollution intensity, and energy consumption patterns [65], which may lead to heterogeneous policy effects across cities. To comprehensively examine whether and how heterogeneous STI policy intensity affects urban carbon unlocking efficiency under different regional conditions, this study conducts heterogeneity analyses along two dimensions: urban pollution intensity and energy consumption intensity.

4.4.1. Heterogeneity by Urban Pollution Intensity

To investigate whether the effects of heterogeneous STI policy intensity vary with urban pollution conditions, this study measures urban pollution intensity as the ratio of city-level carbon dioxide emissions to gross regional product. Cities are then divided into high- and low-pollution groups based on the median value (P50) of pollution intensity. Using a group-wise regression approach, this study examines differences in the effects of heterogeneous STI policy intensity on urban carbon unlocking efficiency across pollution regimes. The corresponding results are reported in Table 12. Columns (1) and (2) show that supply-side STI policy intensity ($Policy\_S$) exerts a significantly positive effect on urban carbon unlocking efficiency ($UCE$) in both high- and low-pollution cities. A coefficient difference test further indicates that the effect of supply-side policies does not significantly differ across pollution levels. This finding suggests that technological innovation supported by supply-side policies constitutes a universal foundation for carbon unlocking, regardless of the initial pollution intensity of a city. In contrast, columns (3)–(8) reveal pronounced heterogeneity in the effects of demand-side ($Policy\_D$), complementary ($Policy\_C$), and institutional reform oriented ($Policy\_I$) STI policies. Based on coefficient estimates and corresponding statistical tests, these policy effects are significantly stronger in low-pollution cities than in high-pollution cities. This pattern can be explained by differences in industrial structure, market conditions, and institutional environments. From the perspective of supply-side policies, technological innovation plays a central role in carbon unlocking across all pollution regimes: in high-pollution cities, low-carbon technological breakthroughs are critical for mitigating emissions from heavy industrial sectors, while in low-pollution cities, technological upgrading primarily improves energy efficiency and production processes. By contrast, the regional heterogeneity of non-supply-side policies is more pronounced. Low-pollution cities tend to have lighter industrial structures, higher levels of marketization, and more mature innovation ecosystems, allowing demand-side policies (e.g., green procurement) to rapidly stimulate market demand, complementary policies (e.g., science–finance coordination) to efficiently allocate innovation resources, and institutional reform oriented policies (e.g., carbon trading rules) to be effectively enforced. These conditions are consistent with Nordhaus’s (2019) argument that emission reduction efficiency depends on policy–market coordination [66]. In contrast, high-pollution cities are often constrained by heavy industrial inertia, weak market demand for low-carbon products, and high institutional implementation costs, which limit the short-term effectiveness of demand-side, complementary, and institutional reform oriented STI policies. As a result, non-supply-side policies are less able to translate into immediate improvements in urban carbon unlocking efficiency in highly polluted regions.

Table 12. Heterogeneity Analysis: Urban pollution intensity.

Panel (A)
	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
	High-Pollution Cities	Low-Pollution Cities	High-Pollution Cities	Low-Pollution Cities
$Policy\_S$	0.0827 *** (2.723)	0.0697 *** (3.511)
$Policy\_D$			0.0307 (0.723)	0.0947 *** (3.083)
$\_cons$	4.4871 ** (2.254)	3.1745 (1.311)	4.3404 ** (2.173)	2.8783 (1.156)
Control variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	2145	2145	2145	2145
R2	0.3311	0.4776	0.3293	0.4781
Empirical p-value	0.442		0.000 ***
Panel (B)
	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(5)	(6)	(7)	(8)
	High-Pollution Cities	Low-Pollution Cities	High-Pollution Cities	Low-Pollution Cities
$Policy\_C$	0.0001 (0.002)	0.0693 ** (2.417)
$Policy\_I$			0.0102 (0.193)	0.1110 *** (3.481)
$\_cons$	4.2340 ** (2.135)	3.6257 (1.463)	4.2709 ** (2.138)	3.2596 (1.332)
Control variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	2145	2145	2145	2145
R2	0.3291	0.4765	0.3291	0.4783
Empirical p-value	0.000 ***		0.000 ***

Notes: *** p < 0.01, ** p < 0.05.

4.4.2. Heterogeneity by Energy Consumption Intensity

To examine whether the effect of heterogeneous STI policy intensity on urban carbon unlocking efficiency varies with cities’ energy consumption conditions, this study classifies cities into high energy-intensity and low energy-intensity groups based on the median (P50) of energy consumption intensity. A coefficient-difference approach is then employed to assess heterogeneity in the effects of heterogeneous STI policies across the two groups. The corresponding results are reported in Table 13. The results indicate that, based on empirical p-values, the effects of supply-side STI policy intensity ($Policy\_S$) and demand-side STI policy intensity ($Policy\_D$) on urban carbon unlocking efficiency ($UCE$) are significantly stronger in high energy-intensity cities. In contrast, the effects of complementary STI policy intensity ($Policy\_C$) and institutional reform oriented STI policy intensity ($Policy\_I$) are more pronounced in low energy-intensity cities. This heterogeneity can be explained by differences in industrial structure, technological constraints, and institutional environments. High energy-intensity cities are typically dominated by heavy and energy-intensive industries, where carbon reduction critically depends on technological breakthroughs that relax binding constraints on energy use and emissions. In this context, supply-side policies—such as R&D subsidies—accelerate the deployment of low-carbon technologies, while demand-side policies facilitate firm-level transformation by stimulating market demand. This mechanism is consistent with Acemoglu et al.’s (2012) theory of directed technical change, which emphasizes the role of targeted innovation policies in steering technological progress toward cleaner technologies [67]. The stronger mitigation effects observed in high energy-intensity cities thus reflect the higher marginal returns to technology-oriented policy interventions. By contrast, low energy-intensity cities tend to exhibit more advanced industrial structures and relatively lower dependence on energy-intensive production. In these cities, the focus of carbon unlocking shifts toward improving innovation ecosystems and strengthening institutional arrangements. Complementary policies enhance the efficiency of technology diffusion and resource coordination, while institutional reform oriented policies stabilize emission-reduction expectations and improve regulatory enforcement. This pattern aligns with the view that emission reduction effectiveness depends on policy–market coordination and institutional capacity.

Table 13. Heterogeneity Analysis: Energy consumption intensity.

Panel (A)
	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
	High Energy-Intensity Regions	Low Energy-Intensity Regions	High Energy-Intensity Regions	Low Energy-Intensity Regions
$Policy\_S$	0.0859 *** (4.312)	0.0830 ** (2.513)
$Policy\_D$			0.0716 *** (3.093)	0.2002 *** (3.373)
$\_cons$	7.7739 *** (3.834)	−0.2660 (−0.094)	7.8782 *** (3.855)	−1.6829 (−0.569)
Control variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	2115	2115	2115	2115
R2	0.3592	0.4760	0.3565	0.4790
Empirical p-value	0.000 ***		0.000 ***
Panel (B)
	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(5)	(6)	(7)	(8)
	High Energy-Intensity Regions	Low Energy-Intensity Regions	High Energy-Intensity Regions	Low Energy-Intensity Regions
$Policy\_C$	0.0420 ** (1.969)	0.1359 ** (2.310)
$Policy\_I$			0.0729 *** (2.896)	0.1872 *** (3.149)
$\_cons$	8.1100 *** (3.958)	−0.8392 (−0.283)	8.1787 *** (3.997)	−0.9036 (−0.311)
Control variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	2115	2115	2115	2115
R2	0.3542	0.4765	0.3557	0.4785
Empirical p-value	0.000 ***		0.000 ***

Notes: *** p < 0.01, ** p < 0.05.

5. Mechanism Analysis of the Effects of Heterogeneous STI Policy Intensity on Urban Carbon Unlocking Efficiency

Based on the preceding theoretical analysis, regional digital infrastructure development, the level of science and technology finance, and government green attention are expected to shape the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency. Following Edwards and Lambert (2007) [68], a moderated regression model is constructed as follows:

$$\begin{array}{cc}UC{E}_{it}& ={\alpha }_0+{\alpha }_{1}Digita{l}_{it}+{\alpha }_{2}Policy\_S_{it}/Policy\_D_{it}/Policy\_C_{it}/Policy\_I_{it}\hfill \\ & +{\alpha }_{3}Digita{l}_{it}\times Policy\_S_{it}/Policy\_D_{it}/Policy\_C_{it}/Policy\_I_{it}\hfill \\ & +\lambda contro{l}_{it}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}\hfill \end{array}$$

(8)

$$\begin{array}{cc}UC{E}_{it}& ={\alpha }_0+{\alpha }_{1}Financ{e}_{it}+{\alpha }_{2}Policy\_S_{it}/Policy\_D_{it}/Policy\_C_{it}/Policy\_I_{it}\hfill \\ & +{\alpha }_{3}Financ{e}_{it}\times Policy\_S_{it}/Policy\_D_{it}/Policy\_C_{it}/Policy\_I_{it}\hfill \\ & +\lambda contro{l}_{it}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}\hfill \end{array}$$

(9)

$$\begin{array}{cc}UC{E}_{it}& ={\alpha }_0+{\alpha }_{1}Gree{n}_{it}+{\alpha }_{2}Policy\_S_{it}/Policy\_D_{it}/Policy\_C_{it}/Policy\_I_{it}\hfill \\ & +{\alpha }_{3}Gree{n}_{it}\times Policy\_S_{it}/Policy\_D_{it}/Policy\_C_{it}/Policy\_I_{it}\hfill \\ & +\lambda contro{l}_{it}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}\hfill \end{array}$$

(10)

where $UC{E}_{it}$ denotes the dependent variable, measuring the urban carbon unlocking efficiency of city $i$ in year $t$. $Policy\_S_{it}$, $Policy\_D_{it}$, $Policy\_C_{it}$, and $Policy\_I_{it}$ are the core explanatory variables, capturing the intensity of heterogeneous STI policies in city $i$ at time $t$, including supply-side policies, demand-side policies, complementary-factor policies, and institutional reform policies, respectively. $Digita{l}_{it}$, $Finac{e}_{it}$, and $Gee{n}_{it}$ are moderating variables, representing the level of regional digital infrastructure development, the development of science and technology finance, and government green attention. $contro{l}_{it}$ denotes a vector of control variables consistent with the baseline regression. Year fixed effects and city fixed effects are denoted by ${\delta }_t$ and ${\gamma }_i$, respectively. The error term ${\epsilon }_{it}$ is assumed to follow a normal distribution, ${\epsilon }_{it}~N(0,{\delta }^2)$.

Regarding the measurement of regional digital infrastructure development ($Digita{l}_{it}$), this study follows the approach of Chao et al. (2021) [69]. Specifically, government work reports of prefecture-level cities from 2009 to 2023 are first collected, and keywords related to new digital infrastructure are identified. Python (Version 3.12.12) is then employed to conduct text segmentation and frequency analysis of the reports, calculating both the total number of words and the frequency of digital-infrastructure-related terms. The ratio of digital infrastructure keywords to total words is used as a baseline indicator. Considering the strong correlation between digital infrastructure development and employment in information-related sectors, the baseline indicator is further weighted by the share of information industry employment in the total population of each city. The interaction of these two components is taken as a comprehensive measure of regional digital infrastructure development. The level of science and technology finance development ($Finac{e}_{it}$) is measured by the logarithm of the number of newly registered technology-oriented enterprises at the prefecture-level city. Government green attention ($Gee{n}_{it}$) is proxied using text analysis of government work reports. Specifically, keywords related to green development are identified, and Python is used to compute their frequencies. The ratio of green-related keywords to total word count in each report is employed to capture the intensity of government green attention.

5.1. Mechanism Test: Regional Digital Infrastructure Development

Table 14 reports the mechanism test results concerning the role of regional digital infrastructure development in moderating the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency. The results indicate that the interaction between regional digital infrastructure development ($Digital$) and supply-side STI policy intensity ($Policy\_S$) is positive and statistically significant at the 1% level. This finding suggests that a higher level of digital infrastructure development strengthens the positive effect of supply-side policies on urban carbon unlocking efficiency. Similarly, the interaction term between digital infrastructure development and demand-side STI policy intensity ($Digital\times Policy\_S$) is significantly positive at the 10% level, indicating that well-developed digital infrastructure also amplifies the effectiveness of demand-side policies in promoting carbon unlocking efficiency. In contrast, the interaction between digital infrastructure development and complementary-factor STI policy intensity ($Digital\times Policy\_C$) is statistically insignificant, implying that digital infrastructure does not significantly condition the effect of complementary-factor policies on urban carbon unlocking efficiency. The core reasons are as follows: Complementary-factor STI policies focus on building physical innovation collaboration carriers and coordinating interests, relying on the adaptability of local industrial bases; while digital infrastructure concentrates on universal information circulation and factor optimization, making it difficult to break through the non-digital barriers in policy implementation. Additionally, some policies are merely formalistic, which further undermines the practical effect of digital empowerment. Finally, the interaction term between digital infrastructure development and institutional-reform STI policy intensity ($Digital\times Policy\_I$) is positive and significant at the 5% level, suggesting that digital infrastructure development is significantly positively associated with the effectiveness of institutional reform policies in improving urban carbon unlocking efficiency. Taken together, these results indicate that regional digital infrastructure development plays a significant moderating role in the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency. The above results indicate that regional digital infrastructure exhibits heterogeneous moderating effects on the relationship between heterogeneous STI policies and urban carbon unlocking efficiency. It exerts a positive reinforcing impact on supply-side, demand-side, and institutional-reform STI policies, while its interaction effect with complementary-factor STI policies is statistically insignificant. Overall, these findings suggest that the moderating role of digital infrastructure is selective across different policy types. This provides empirical support for Hypothesis H2a, while acknowledging that the interaction terms demonstrate a contextual strengthening of the association rather than directly testing specific transmission mechanisms such as factor mobility.

Table 14. Mechanism Analysis: The moderating role of regional digital infrastructure development.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
$Digital$	−1.3573 *** (−3.698)	−1.1459 *** (−3.518)	−1.0803 *** (−2.983)	−1.2411 *** (−3.685)
$Policy\_S$	0.0323 (1.006)
$Digital\times Policy\_S$	0.3071 *** (2.643)
$Policy\_D$		0.0992 (1.554)
$Digital\times Policy\_D$		0.3421 * (1.673)
$Policy\_C$			0.0686 (1.120)
$Digital\times Policy\_C$			0.1923 (1.182)
$Policy\_I$				0.0882 (1.568)
$Digital\times Policy\_I$				0.3375 ** (2.041)
$\_cons$	2.1626 (1.362)	2.1835 (1.347)	2.7494 * (1.689)	2.4612 (1.517)
Control variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	4209	4209	4209	4209
R2	0.4335	0.4348	0.4311	0.4341

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1.

5.2. Mechanism Test: Regional Science and Technology Finance Development

Table 15 presents the mechanism test results regarding the moderating role of regional science and technology finance development in the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency. The results show that the interaction term ($Finance\times Policy\_S$) between regional science and technology finance development ($Finace$) and supply-side STI policy intensity ($Policy\_S$) is positive and statistically significant at the 1% level. This finding indicates that a higher level of science and technology finance development strengthens the positive effect of supply-side policies on urban carbon unlocking efficiency by enhancing financial support for low-carbon technological innovation. Similarly, the interaction between science and technology finance development and demand-side STI policy intensity ($Finance\times Policy\_D$) is significantly positive at the 1% level, suggesting that well-developed science and technology finance systems amplify the effectiveness of demand-side policies in stimulating market demand for low-carbon technologies and accelerating their diffusion. Moreover, the interaction term between science and technology finance development and complementary-factor STI policy intensity ($Finance\times Policy\_C$) is also positive and statistically significant at the 1% level. This result implies that science and technology finance development is significantly positively associated with the coordination effect of complementary-factor policies by facilitating resource matching and reducing financing constraints along the innovation chain. Finally, the interaction between science and technology finance development and institutional-reform STI policy intensity ($Finance\times Policy\_I$) is positive and significant at the 1% level, indicating that the development of science and technology finance significantly strengthens the observed positive association across different policy types, suggesting its role as a key contextual enhancer. The results provide strong empirical support for Hypothesis H2b. It is important to note that while these results are consistent with a theoretical role of science and technology finance in mitigating financial constraints, the interaction analysis primarily establishes the significance of this contextual factor rather than directly testing the underlying financial channel.

Table 15. Mechanism Analysis: The moderating role of regional science and technology finance development.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
$Finace$	−0.0383 (−0.953)	−0.0567 (−1.394)	−0.1115 ** (−2.152)	−0.1118 ** (−2.491)
$Policy\_S$	−0.0533 * (−1.747)
$Finance\times Policy\_S$	0.0518 *** (5.348)
$Policy\_D$		−0.0713 (−1.429)
$Finance\times Policy\_D$		0.0929 *** (5.559)
$Policy\_C$			−0.1180 ** (−2.465)
$Finance\times Policy\_C$			0.0964 *** (5.227)
$Policy\_I$				−0.0692 (−1.568)
$Finance\times Policy\_I$				0.1011 *** (6.249)
$\_cons$	0.2551 (0.165)	−0.3914 (−0.251)	0.0762 (0.049)	−0.3918 (−0.253)
Control variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	4226	4226	4226	4226
R2	0.4339	0.4372	0.4347	0.4384

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1.

5.3. Mechanism Test: Government Green Attention

Table 16 reports the mechanism test results concerning the moderating role of government green attention in the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency. The empirical results indicate that the interaction ($Green\times Policy\_S$) between government green attention ($Green$) and supply-side STI policy intensity ($Policy\_S$) is positive and statistically significant at the 1% level. This finding suggests that a higher level of government green attention strengthens the effectiveness of supply-side policies in promoting urban carbon unlocking efficiency by enhancing policy enforcement and prioritizing low-carbon technological innovation. Similarly, the interaction term between government green attention and demand-side STI policy intensity ($Green\times Policy\_D$) is significantly positive at the 1% level. This result implies that strong governmental emphasis on green development amplifies the impact of demand-side policies by improving policy credibility, increasing public awareness, and facilitating the market diffusion of low-carbon products. Moreover, the interaction between government green attention and complementary-factor STI policy intensity ($Green\times Policy\_C$) is also positive and statistically significant at the 1% level. This finding indicates that heightened government green attention is significantly positively associated with the coordination effect of complementary policies by reducing administrative fragmentation and improving cross-departmental policy alignment, thereby strengthening their contribution to carbon unlocking efficiency. Finally, the interaction term between government green attention and institutional-reform STI policy intensity ($Green\times Policy\_I$) is positive and significant at the 1% level. Overall, these results demonstrate that government green attention plays a significant positive moderating role, indicating that the positive association between STI policies and carbon unlocking efficiency is more pronounced in contexts where the government prioritizes green development. These findings provide robust empirical support for Hypothesis H2c. The analysis shows that government green attention is a significant contextual factor that strengthens the observed association, aligning with theoretical expectations regarding policy enforcement and credibility.

Table 16. Mechanism Analysis: The moderating role of government green attention.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)	(3)	(4)
$Green$	−53.9221 *** (−10.071)	−54.0788 *** (−10.862)	−54.9788 *** (−10.629)	−55.4769 *** (−10.794)
$Policy\_S$	0.0307 * (1.737)
$Green\times Policy\_S$	3.1918 ** (2.554)
$Policy\_D$		0.0276 (1.543)
$Green\times Policy\_D$		6.3757 *** (3.656)
$Policy\_C$			0.0035 (0.168)
$Green\times Policy\_C$			5.2339 *** (3.160)
$Policy\_I$				0.0195 (0.739)
$Green\times Policy\_I$				6.1095 *** (3.495)
$\_cons$	1.4106 (0.903)	1.3992 (0.886)	1.7194 (1.085)	1.7495 (1.109)
Control variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	4290	4290	4290	4290
R2	0.4478	0.4490	0.4471	0.4486

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1.

6. Empirical Analysis of the Spatial Effects of Heterogeneous STI Policy Intensity on Urban Carbon Unlocking Efficiency

The preceding analyses indicate that heterogeneous STI policy intensity significantly promotes urban carbon unlocking efficiency. However, given the strong interregional linkages across Chinese cities, urban carbon unlocking efficiency may exhibit spatial dependence. Meanwhile, increases in heterogeneous STI policy intensity in one city may influence not only local carbon unlocking outcomes but also those of neighboring cities through factor mobility, policy diffusion, and interregional interactions. Therefore, building on the baseline and mechanism analyses, this section incorporates spatial interaction effects into the analytical framework to examine whether heterogeneous STI policy intensity generates spatial spillover effects on urban carbon unlocking efficiency. By explicitly modeling spatial dependence, this analysis aims to provide a more comprehensive assessment of the overall policy impact and to further verify the role of heterogeneous STI policies in shaping regional carbon unlocking dynamics.

6.1. Specification of the Spatial Econometric Model

From the perspective of spatial econometric modeling, commonly used specifications include the spatial autoregressive model (SAR), the spatial error model (SEM), and the spatial Durbin model (SDM). LeSage and Pace (2009) argue that, compared with SAR and SEM, the SDM offers greater flexibility by simultaneously capturing spatial dependence in both the dependent variable and explanatory variables [70], thereby effectively mitigating omitted-variable bias arising from unobserved spatial spillovers. Following this rationale, and drawing on the empirical strategy of Jia et al. (2021), this study employs the spatial Durbin model to investigate the spatial effects of heterogeneous STI policy intensity on urban carbon unlocking efficiency [71]. In the empirical analysis, Wald tests are further conducted to verify the appropriateness of the SDM specification. Accordingly, the spatial Durbin model is specified as follows:

$$\begin{array}{c}UC{E}_{it}=\theta \times W_{ij}\times UC{E}_{jt}+\beta \times Policy\_S_{it}+\sigma \times W_{ij}\times Policy\_S_{ij}\\ +{\displaystyle \sum _k{\gamma }_k}contro{l}_{it}+{\displaystyle \sum _k{\nu }_k}{W}_{ij}contro{l}_{ij}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}\end{array}$$

(11)

$$\begin{array}{c}UC{E}_{it}=\theta \times W_{ij}\times UC{E}_{jt}+\beta \times Policy\_D_{it}+\sigma \times W_{ij}\times Policy\_D_{ij}\\ +{\displaystyle \sum _k{\gamma }_k}contro{l}_{it}+{\displaystyle \sum _k{\nu }_k}{W}_{ij}contro{l}_{ij}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}\end{array}$$

(12)

$$\begin{array}{c}UC{E}_{it}=\theta \times W_{ij}\times UC{E}_{jt}+\beta \times Policy\_C_{it}+\sigma \times W_{ij}\times Policy\_C_{ij}\\ +{\displaystyle \sum _k{\gamma }_k}contro{l}_{it}+{\displaystyle \sum _k{\nu }_k}{W}_{ij}contro{l}_{ij}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}\end{array}$$

(13)

$$\begin{array}{c}UC{E}_{it}=\theta \times W_{ij}\times UC{E}_{jt}+\beta \times Policy\_I_{it}+\sigma \times W_{ij}\times Policy\_I_{ij}\\ +{\displaystyle \sum _k{\gamma }_k}contro{l}_{it}+{\displaystyle \sum _k{\nu }_k}{W}_{ij}contro{l}_{ij}+{\delta }_t+{\gamma }_i+{\epsilon }_{it}\end{array}$$

(14)

where $UC{E}_{it}$ denotes the dependent variable, representing the urban carbon unlocking efficiency of city $i$ in year $t$. $Policy\_S_{it}$, $Policy\_D_{it}$, $Policy\_C_{it}$, and $Policy\_I_{it}$ are the core explanatory variables, capturing the intensity of heterogeneous STI policies in city $i$ in year $t$, including supply-side, demand-side, complementary-factor, and institutional-reform policy intensity, respectively. $contro{l}_{it}$ represents a set of control variables, including financial development ($Fd$), economic development ($Ed$), urbanization level ($Urban$), local government expenditure ($Ge$), per capita fixed-asset investment ($Ifa$), and informatization level ($Id$). $W$ denotes the spatial weight matrix. Accordingly, $W\times Policy\_S$, $W\times Policy\_D$, $W\times Policy\_C$, and $W\times Policy\_I$ represent the spatially lagged terms of heterogeneous STI policy intensity, reflecting the spillover effects of supply-side, demand-side, complementary-factor, and institutional-reform policies across neighboring cities. Similarly, $W\times UCE$ represents the spatial lag of urban carbon unlocking efficiency, capturing spatial dependence in the dependent variable. Following Xu and Zhang (2017) [72], this study incorporates economic proximity into the construction of the spatial weight matrix and adopts an economic-distance-based spatial weight matrix. Specifically, economic distance is embedded into the geographical framework to better capture spatial interactions driven by economic similarity rather than purely physical distance.

$$W_{ij}^e=\left\{\begin{array}{c}{W}_{ij}^d\ast diag({\overline{Y}}_1/\overline{Y},{\overline{Y}}_2/\overline{Y},\cdots {\overline{Y}}_n/\overline{Y}),i\ne j\\ 0,\hspace{1em}i=j\end{array}\right.$$

(15)

where $W_{ij}^d$ denotes the inverse of the straight-line Euclidean distance between city $i$ and city $j$. ${\overline{Y}}_i=1/(t_1-t_0+1){\displaystyle {\sum }_{t_0}^{t_1}{Y}_{it}}$ represents the average GDP of the city $i$ over the observation period; $\overline{Y}=1/n(t_1-t_0+1){\displaystyle {\sum }_{i=1}^n{\displaystyle {\sum }_{t_0}^{t_1}{Y}_{it}}}$ denotes the average GDP of all cities during the observation period. The geographical data used to construct the spatial weight matrix are obtained from the National Geomatics Center of China.

6.2. Spatial Autocorrelation Test

To further verify whether urban carbon unlocking efficiency exhibits spatial dependence, a spatial autocorrelation test is conducted. Table 17 reports the Moran’s I statistics for urban carbon unlocking efficiency. The results show that, under the economic–geographical distance-based spatial weight matrix, urban carbon unlocking efficiency does not exhibit significant spatial autocorrelation in 2010 and 2011. However, from 2012 to 2023, the Moran’s I statistics are significantly positive, indicating the presence of positive spatial dependence in urban carbon unlocking efficiency across Chinese cities. Overall, the Moran’s I statistics for the period 2010–2023 are predominantly positive and statistically significant, although their magnitudes fluctuate over time. This pattern suggests that urban carbon unlocking efficiency is not randomly distributed across space, but instead displays evolving spatial dependence and clustering characteristics. The observed spatial agglomeration implies that neighboring cities tend to exhibit similar levels of carbon unlocking efficiency, highlighting the necessity of explicitly accounting for spatial interactions when assessing the effects of heterogeneous STI policies.

Table 17. Moran’s I Statistics of Urban Carbon Unlocking Efficiency.

	Moran’s I	Z-Value	p-Value
2010	0.000	0.582	0.280
2011	−0.002	0.303	0.381
2012	0.005 *	1.341	0.090
2013	0.009 **	1.990	0.023
2014	0.006 *	1.505	0.060
2015	0.014 ***	2.715	0.003
2016	0.012 ***	2.525	0.006
2017	0.007 **	1.675	0.047
2018	0.011 ***	2.316	0.010
2019	0.017 ***	3.208	0.001
2020	0.045 ***	7.590	0.000
2021	0.039 ***	6.609	0.000
2022	0.027 ***	4.797	0.000
2023	0.042 ***	6.975	0.000

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1.

6.3. Spatial Spillover Effects of Heterogeneous STI Policy Intensity on Urban Carbon Unlocking Efficiency

Based on the spatial analysis of urban carbon unlocking efficiency, the results indicate that urban carbon unlocking efficiency exhibits significant spatial effects. Accordingly, this section further employs the spatial Durbin model (SDM) to empirically examine the spatial spillover effects of heterogeneous STI policy intensity on urban carbon unlocking efficiency. Following Jia et al. (2021) [71], Wald tests are conducted to assess the appropriateness of the SDM specification. The results reported in Table 18 show that both the Wald SAR and Wald SEM test statistics are statistically significant, indicating that the SDM cannot be simplified into either the spatial autoregressive model (SAR) or the spatial error model (SEM) (To verify the robustness of the findings regarding spatial spillover effects, this study constructs an innovation-geographic distance nested matrix as an alternative specification. The resulting estimates are consistent with the baseline model utilizing the economic-geographic distance weight matrix.). This confirms the suitability of adopting the SDM for the spatial analysis. In addition, the goodness-of-fit statistics, including $Sigma2$, $Log$ and $R^2$, suggest that the estimated models exhibit satisfactory explanatory power and reliability, providing mutual support for the robustness of the empirical results. Moreover, the spatial autoregressive coefficient ($\rho$) is statistically significant at conventional confidence levels (above the 10% significance level), indicating the presence of significant spatial spillover effects in the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency. Specifically, the estimated coefficients of supply-side STI policy intensity ($Policy\_S$), demand-side STI policy intensity (Policy_D), complementary-factor STI policy intensity (Policy_C), and institutional-reform STI policy intensity (Policy_I) are all positive and statistically significant. These results imply that, after explicitly accounting for spatial interactions, heterogeneous STI policy intensity continues to exert a positive effect on urban carbon unlocking efficiency. Overall, the findings confirm that heterogeneous STI policies is not only significantly positively associated with local carbon unlocking performance but also generate positive spatial spillover effects across neighboring cities.

Table 18. Estimation results for the spatial spillover effects of heterogeneous STI policy intensity.

	Dependent Variable: Urban Carbon Unlocking Efficiency ($\mathit{U}\mathit{C}\mathit{E}$)
	(1)	(2)
$Policy\_S$	0.0700 *** (4.000)
$W\times Policy\_S$	0.4277 *** (4.468)
$Policy\_D$		0.0833 *** (3.853)
$W\times Policy\_D$		0.1247 (1.056)
$Policy\_C$			0.0695 *** (2.948)
$W\times Policy\_C$			−0.2231 ** (−2.092)
$Policy\_I$				0.0724 *** (2.963)
$W\times Policy\_I$				1.5486 *** (4.660)
$\rho$	0.1754 * (1.656)	0.3000 *** (3.154)	0.3000 *** (3.150)	0.2261 * (1.906)
$Sigma 2$	1.1544 *** (46.401)	1.1662 *** (46.146)	1.1671 *** (46.150)	1.1577 *** (46.304)
Wald SAR (p)	41.09 *** [0.000]	22.73 *** [0.001]	41.09 *** [0.000]	22.73 *** [0.001]
Wald SEM (p)	47.85 *** [0.000]	26.34 *** [0.000]	47.85 *** [0.000]	26.34 *** [0.000]
Control variables	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	4290	4290	4290	4290
$R^2$	0.0058	0.0001	0.0030	0.0000

Notes: The findings are robust to the use of this alternative matrix and that detailed results are available upon request. *** p < 0.01, ** p < 0.05, * p < 0.1.

The above results indicate that, after accounting for spatial dependence, heterogeneous STI policy intensity continues to show a positive association with urban carbon unlocking efficiency. However, the baseline spatial regression does not explicitly distinguish between the relationship of heterogeneous STI policies within local cities and their associations with outcomes in neighboring cities. Therefore, following LeSage and Pace (2009) [70], this study further decomposes the spatial Durbin model (SDM) to obtain the direct, indirect, and total effects. The estimation results are reported in Table 19. The results in column (1) show that the indirect effect of supply-side STI policy intensity ($Policy\_S$) is significantly positive. This finding suggests that, while supply-side policies enhance carbon unlocking efficiency in the local city, they also generate positive spatial spillover effects that significantly improve carbon unlocking efficiency in neighboring cities. Column (2) reports that the indirect effect of demand-side STI policy intensity ($Policy\_D$) is not statistically significant, indicating a lack of empirical evidence supporting the existence of significant spatial spillover effects associated with demand-side policies. Column (3) shows that the indirect effect of complementary-factor STI policy intensity ($Policy\_C$) is significantly negative. This result implies that increases in complementary-factor policy intensity may be significantly negatively associated with improvements in carbon unlocking efficiency in surrounding cities, possibly due to resource reallocation or competitive effects. Finally, the results in column (4) indicate that the indirect effect of institutional-reform STI policy intensity ($Policy\_I$) is significantly positive. This finding suggests that institutional reform policies are not only positively associated with local carbon unlocking efficiency but also correlated with improvements in neighboring cities through positive spatial spillover effects.

Table 19. Direct, indirect, and total effects of heterogeneous STI policy intensity on urban carbon unlocking efficiency.

	(1)	(2)	(3)
	Direct Effects	Indirect Effects	Total Effects
$Policy\_S$	0.0713 *** (3.975)	0.5319 *** (4.743)	0.6032 *** (5.357)
$Policy\_D$	0.0845 *** (3.807)	0.2129 (1.250)	0.2975 * (1.718)
$Policy\_C$	0.0698 *** (2.884)	−0.2946 ** (−1.977)	−0.2248 (−1.496)
$Policy\_I$	0.0764 *** (3.062)	2.0367 *** (4.209)	2.1131 *** (4.393)

Notes: (1) *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. (2) Z-values are reported in parentheses.

7. Conclusions and Discussion

As digital technologies such as big data increasingly reshape resource allocation processes, understanding the relationship between heterogeneous science and technology innovation (STI) policy intensity and urban carbon unlocking efficiency has become a critical issue for achieving low-carbon and high-quality urban development. Against this backdrop, this paper investigates this relationship using panel data for 286 prefecture-level cities in China over the period 2009–2023. The analysis proceeds in several steps. First, the theoretical foundations and transmission mechanisms are articulated. Second, the relationship is empirically examined, with heterogeneity analyses conducted along pollution intensity and energy consumption intensity dimensions. Third, the moderating roles of regional digital infrastructure development, science and technology finance development, and government green attention are explored. Finally, spatial dependence is explicitly incorporated to assess spatial spillover effects.

The main findings can be summarized as follows. First, baseline regression results show that all four types of heterogeneous STI policy intensity—supply-side, demand-side, complementary-factor, and institutional-reform policies—are significantly positively associated with urban carbon unlocking efficiency, with institutional-reform policies exhibiting the strongest association. When all four policy types are included simultaneously, only supply-side and demand-side policies retain statistically significant direct associations. A series of endogeneity treatments and robustness checks confirm the stability of these core findings. Second, heterogeneity analyses reveal notable regional differences. From the perspective of pollution intensity, the association of supply-side policies does not differ significantly between high- and low-pollution regions, whereas demand-side, complementary-factor, and institutional-reform policies show a stronger association with efficiency gains in low-pollution regions. From the perspective of energy consumption intensity, supply-side and demand-side policies have a stronger association in high-energy-consumption regions, while complementary-factor and institutional-reform policies are more prominently linked to efficiency in low-energy-consumption regions. Third, mechanism analysis shows that regional digital infrastructure exerts heterogeneous moderating effects on the relationship between heterogeneous sci-tech innovation policies and urban carbon emission reduction efficiency. It has a positive reinforcing effect on supply-side, demand-side, and institutional reform-oriented sci-tech innovation policies, while the interaction effect with complementary sci-tech innovation policies is not statistically significant. These findings reveal that the moderating role of digital infrastructure exhibits selective differences across different policy types. Additionally, the development of technology finance acts as a “resource catalyst,” and government green policies play the role of “institutional guarantee,” both significantly enhancing the correlation between the intensity of heterogeneous sci-tech policies and urban carbon emission reduction efficiency. Fourth, spatial analyses show that urban carbon unlocking efficiency exhibits significant spatial autocorrelation. Results from the spatial Durbin model indicate that all four types of STI policies are significantly positively associated with local carbon unlocking efficiency. Further decomposition of spatial effects reveals that supply-side and institutional-reform policies show positive spatial associations, demand-side policies do not exhibit significant spatial associations, and complementary-factor policies show negative spatial associations.

The main findings can be summarized as follows. First, baseline regression results show that all four types of heterogeneous STI policy intensity—supply-side, demand-side, complementary-factor, and institutional-reform policies—are significantly positively associated with urban carbon unlocking efficiency, with institutional-reform policies exhibiting the strongest association. When all four policy types are included simultaneously, only supply-side and demand-side policies retain statistically significant direct associations. A series of endogeneity treatments and robustness checks confirm the stability of these core findings. Second, heterogeneity analyses reveal notable regional differences. From the perspective of pollution intensity, the association of supply-side policies does not differ significantly between high- and low-pollution regions, whereas demand-side, complementary-factor, and institutional-reform policies show a stronger association with efficiency gains in low-pollution regions. From the perspective of energy consumption intensity, supply-side and demand-side policies have a stronger association in high-energy-consumption regions, while complementary-factor and institutional-reform policies are more prominently linked to efficiency in low-energy-consumption regions. Third, mechanism analyses indicate that the levels of regional digital infrastructure development, science and technology finance development, and government green attention are all significantly and positively associated with the strength of the relationship between heterogeneous STI policy intensity and urban carbon unlocking efficiency. This suggests these contextual factors strengthen the observed positive association, though their specific intermediary roles are not directly tested here. Fourth, spatial analyses show that urban carbon unlocking efficiency exhibits significant spatial autocorrelation. Results from the spatial Durbin model indicate that all four types of STI policies are significantly positively associated with local carbon unlocking efficiency. Further decomposition of spatial effects reveals distinct patterns: supply-side and institutional-reform policies show positive spatial associations, complementary-factor policies show a negative spatial association, and demand-side policies exhibit no statistically significant spatial association.

The findings provide strong empirical support for the role of heterogeneous STI policy intensity and yield several important policy implications. First, given the differentiated associations, policymakers should construct a coordinated policy mix characterized by “supply–demand leadership and complementary–institutional synergy.” On the one hand, supply-side and demand-side policies should be strengthened as direct drivers. On the supply side, increasing R&D subsidies, improving innovation infrastructure, and cultivating high-skilled technical talent can reduce the costs of green technology development and accelerate low-carbon technology commercialization [73]. On the demand side, expanding green public procurement, improving carbon pricing mechanisms, and enhancing low-carbon consumption subsidies can stimulate market demand and better align technological supply with carbon unlocking needs [74,75]. On the other hand, the lagged and interactive nature of complementary-factor and institutional-reform policies should be addressed by building cross-departmental innovation service platforms and establishing long-term evaluation mechanisms. Strengthening coordination between these policies and supply- and demand-side instruments can activate their indirect enabling effects. In addition, dynamic policy monitoring systems should be established to regularly assess effects and synergies.

Second, greater emphasis should be placed on reinforcing the enabling roles of digital infrastructure, science and technology finance, and government green attention. Accelerating the deployment of new digital infrastructures, promoting data openness and sharing, developing a multi-tiered green finance support system, and incorporating carbon unlocking efficiency into government performance evaluation can all help amplify the effectiveness of innovation policies. We encourage banks to provide dedicated loans for low-carbon technology R&D, establish government-guided green innovation funds, and pilot innovative instruments such as carbon-backed financing and carbon options.

Third, differentiated policy strategies should be implemented based on regional differences. For high-pollution regions, supply-side policies should continue to focus on breaking technological bottlenecks in industrial emission reduction. For low-pollution regions, greater emphasis should be placed on demand-side, complementary-factor, and institutional-reform policies. High-energy-consumption regions should prioritize coordinated supply-side and demand-side interventions, while low-energy-consumption regions should focus on optimizing complementary-factor and institutional-reform policies. Finally, interregional policy coordination mechanisms should be established to facilitate knowledge sharing and cross-regional learning.

8. Limitations and Future Research

This study acknowledges certain limitations that point to valuable directions for future research.

First, regarding the measurement of core variables, it is important to reiterate that the core explanatory variable, policy intensity, is operationalized as the count of policy documents within distinct categories. This approach fundamentally captures the degree of governmental attention or priority allocated to different STI policy portfolios, rather than the absolute financial weight or ex-post impact of individual policies. Consequently, the primary aim and interpretation of our findings center on revealing the differential directional associations of various types of policy emphasis, not on providing a precise, quantified assessment of the effect size per unit of fiscal expenditure. While the use of policy counts, refined by categorical disaggregation, provides a feasible and comparable proxy for such governmental prioritization across a large sample, it does not directly capture the granular intensity or implementation efficacy of individual policies. Variations in fiscal commitment, enforcement rigor, and practical scope among policies bearing the same formal title are not fully accounted for. Future research could significantly enhance the precision of policy measurement by integrating diverse data sources, such as detailed fiscal expenditure records, specific subsidy values, procurement volumes under demand-side policies, or regulatory compliance rates, to construct composite, weighted indices.

Second, the theoretical interpretations offered for the observed spatial spillover patterns—particularly the negative association linked to complementary-factor policies—remain propositions that require more direct empirical validation. Our analysis identifies significant spatial associations but does not test the specific channels (e.g., cross-city factor flows, knowledge diffusion, or competitive dynamics) that may underlie them. Future studies could significantly advance this line of inquiry by incorporating more granular data on inter-city resource mobility, firm linkages, or policy diffusion networks to empirically examine the mechanisms driving these spatial correlations.

Finally, while the study employs panel and spatial econometric models alongside multiple robustness checks to strengthen causal inference, the potential for endogeneity due to omitted variables or reverse causality cannot be entirely dismissed. Future work could benefit from more stringent quasi-experimental designs or the utilization of natural policy experiments to further isolate the causal effects of heterogeneous STI policies on urban carbon unlocking efficiency.